Вы находитесь на странице: 1из 137

Lightweight Robotic Excavation

Krzysztof Skonieczny
April 17, 2013

School of Computer Science


Carnegie Mellon University
Pittsburgh, PA 15213
Thesis Committee:
David Wettergreen, Co-Chair
William (Red) L. Whittaker, Co-Chair
Dimitrios Apostolopoulos
Karl Iagnemma, MIT
Submitted in partial fulfillment of the requirements
for the degree of Doctor of Philosophy.

c 2013 Krzysztof Skonieczny


Copyright

Abstract
Planetary excavators face unique and extreme engineering constraints relative to
terrestrial counterparts. In space missions mass is always at a premium because it
is the main driver behind launch costs. Lightweight operation, due to low mass and
reduced gravity, hinders excavation and mobility by reducing the forces a robot can
effect on its environment.
This thesis shows that there is a quantifiable, non-dimensional threshold that
distinguishes the regimes of lightweight and heavy excavation. This threshold is
crossed at lower weights for continuous excavators (bucket-wheels, bucket chains,
etc.) than discrete excavators (loaders, scrapers, etc.). The lightweight threshold
relates payload ratio (weight of regolith payload collected to empty robot weight),
excavation resistance (force imparted on an excavator by cutting and collecting soil),
and excavation thrust (force supplied by an excavator that is available for cutting
soil).
Experiments and simulation herein show that payload ratio governs productivity
of lightweight excavators. Reducing weight (due to low mass, reduced gravity, or
both) decreases an excavators thrust to resistance ratio, especially in cohesive soils.
There is a predictable regime in the operating space where this ratio is low enough
that it limits an excavators payload ratio and, ultimately, productivity. Discrete
excavators cross into this regime more readily than continuous excavators, because
soil accumulation on their blades increases their excavation resistance.
This research introduces novel experimentation that for the first time subjects
excavators to gravity offload (a cable pulls up on the robot with 5/6 its weight, to
simulate lunar gravity) while they dig. A 300 kg excavator offloaded to 1/6 g successfully collects 0.5 kg/s using a bucket-wheel, with no discernable effect on mobility. For a discrete excavator of the same weight, production rapidly declines as rising
excavation resistance stalls the robot; in total the discrete bucket collects less than
20 kg of regolith. These experiments demonstrate that discrete excavation crosses
the lightweight threshold under conditions where continuous excavation does not.
They also suggest caution in interpreting low gravity performance predictions based
solely on testing in Earth gravity.
This work develops a novel robotic bucket-wheel excavator. It features unique
direct transfer from a bucket-wheel to a high payload ratio dump bed, as well as a
high traction and high speed mobility system. Past lightweight excavator prototypes
were too slow or carried too little regolith payload. Some used bucket-wheels or
bucket-ladders to dig continuously, but transported regolith using exposed chains or
conveyors that would not withstand harsh lunar conditions.
Future research on lightweight excavation would benefit from testing in reduced
gravity flights. These provide the most representative test environment short of actually operating on a planetary surface, as excavator and regolith are both subject
to reduced gravity. Another important direction for future study is deep excavation
in the presence of submerged rocks, which pose challenges for lighweight continuous and discrete excavators alike. Experiments to confirm the generality of results

herein are recommended, including studying the scaling of excavation resistance in


cohesive soils, and comparing a broad variety of discrete and continuous excavator
tools.

iv

Contents
1

Introduction
1.1 Motivation for planetary excavation . . . . . . . . . . . . . . . . . . . . . . .
1.2 The Load-Haul-Dump cycle at the core of excavation tasks . . . . . . . . . . .
1.3 Lightweight is low mass in low gravity . . . . . . . . . . . . . . . . . . . . . .
1.4 Definitions of important terms and concepts . . . . . . . . . . . . . . . . . . .
1.4.1 Continuous vs. discrete excavators . . . . . . . . . . . . . . . . . . . .
1.4.2 Payload ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4.3 Excavation thrust . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4.4 Excavation resistance . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.5 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.6 The problem of distinguishing productive lightweight excavator configurations
1.7 Thesis Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.8 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.

1
1
2
4
6
6
9
9
10
10
11
11
11

.
.
.
.
.
.
.
.
.
.
.
.
.
.

13
13
14
15
16
16
17
17
18
19
19
21
23
23
24

Hauling and Payload Ratio


3.1 Task-level site work modeling . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.1 Traction modeling (wheels) . . . . . . . . . . . . . . . . . . . . . . . .
3.1.2 Excavation models . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

27
27
28
30

Background and Related Work


2.1 Fundamental mechanics of excavation . . . . . . . . . . . . . . . . .
2.1.1 Gravity and cohesion forces included in all excavation models
2.1.2 Adhesion and inertial forces can usually be neglected . . . . .
2.1.3 Surcharge forces arise due to soil accumulation . . . . . . . .
2.1.4 Discrete Element Models for excavation . . . . . . . . . . . .
2.2 Experimentation for Lunar and Planetary Excavation . . . . . . . . .
2.2.1 The large impact of soil accumulation on discrete excavation .
2.2.2 Soil properties and gravity are important conditions to control
2.3 Applicability of excavation resistance models to planetary excavation
2.4 Lunar excavation trade studies . . . . . . . . . . . . . . . . . . . . .
2.5 Lightweight excavator prototypes . . . . . . . . . . . . . . . . . . . .
2.6 Soil loosening methods and mechanisms . . . . . . . . . . . . . . . .
2.7 Autonomous Earthmoving and Tele-Operation . . . . . . . . . . . . .
2.8 Conclusions Based on Related Work . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

3.2

3.3
3.4
3.5
4

3.1.3 Operations modeling . . . . . . . . . . . . . . .


3.1.4 Power modeling . . . . . . . . . . . . . . . . .
3.1.5 Parametric sensitivity analysis . . . . . . . . . .
Experiments with a small robotic excavator . . . . . . .
3.2.1 Experimental setup . . . . . . . . . . . . . . . .
3.2.2 Predicted sensitivity of experimental parameters
3.2.3 Experimental results . . . . . . . . . . . . . . .
Comparison of simulated and experimental results . . . .
Hauling dominates task productivity . . . . . . . . . . .
Conclusions from sensitivy experiments and simulations

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

33
34
35
36
39
41
43
45
48
49

Thrust and resistance in lightweight excavation


4.1 Relationship of mass and scale . . . . . . . . . . . . . . . . . . . . .
4.2 Light weight reduces excavation thrust coefficient . . . . . . . . . . .
4.3 Predicted effects of light weight on excavation resistance coefficient .
4.3.1 Effects of light weight operation on surcharge . . . . . . . . .
4.4 Excavation scaling experiments . . . . . . . . . . . . . . . . . . . . .
4.4.1 Experimental setup . . . . . . . . . . . . . . . . . . . . . . .
4.4.2 Preliminary investigation of soil preparation . . . . . . . . . .
4.4.3 Soil preparation and force measurement . . . . . . . . . . . .
4.4.4 Experimental results . . . . . . . . . . . . . . . . . . . . . .
4.5 Conclusions regarding thrust and resistance for lightweight resistance

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

51
51
53
55
59
61
62
63
64
67
71

The lightweight threshold


5.1 A non-dimensional Lightweight number . . . .
5.1.1 L for continuous and discrete excavation
5.2 Gravity offloaded excavation experiments . . . .
5.2.1 Experimental setup . . . . . . . . . . . .
5.2.2 Predicted lightweight numbers . . . . . .
5.2.3 Experimental results . . . . . . . . . . .
5.3 Conclusions regarding the lightweight threshold .

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

75
75
77
79
80
82
84
86

.
.
.
.

89
89
92
96
96

.
.
.
.
.

99
99
103
103
104
104

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

Lightweight excavator development


6.1 Excavation tooling configuration . . . . . . . . . . . . . . . . . .
6.1.1 Testing transverse bucket-wheels . . . . . . . . . . . . . .
6.2 Excavator mobility system . . . . . . . . . . . . . . . . . . . . .
6.3 Conclusions regarding lightweight robotic excavator development
Conclusions and future work
7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2 Contributions . . . . . . . . . . . . . . . . . . . . . . . . .
7.2.1 Bringing planetary excavation missions forward . . .
7.2.2 Establishing resources and direction for future work
7.3 Future work . . . . . . . . . . . . . . . . . . . . . . . . . .
vi

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.

Bibliography

109

A Extensions
119
A.1 Regolith shaping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
A.2 Non-tractive excavation thrust . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
B Soil flow imaging and grouser spacing

123

vii

viii

1 Introduction
1.1

Motivation for planetary excavation

Excavation of regolith enables in situ resource utilization (ISRU) on the Moon and Mars. ISRU
reduces the cost of exploration by producing consumables (including oxygen, water, and fuel)
from native regolith and building earthwork infrastructure (such as trenches as berms). NASA
highlights five motivations for excavation:
(1) excavation for oxygen production, (2) excavation and material handling for landing pad and berm fabrication, (3) excavation for habitat protection (e.g., radiation
and micrometeoroids), (4) excavation for mission element emplacement (e.g., nuclear reactor burial), and (5) excavation for science (e.g., trenching for stratigraphy
evaluation). The most important task identified to date is regolith excavation and
transport for oxygen production [58].
China intends to build a lunar base for taikonauts. Russia and Japan plan to extablish robotic
lunar outposts. Early ISRU missions will be fully robotic, demonstrating ISRU and excavation technology while carrying out scientific inquiry. Regolith excavation and processing will
continue to be performed robotically as the technology matures, even when supporting human
exploration.
Excavation can expose buried ice by removing overburden. Figure 1.1 shows multiple regions
at the Lunar South Pole that may harbor buried deposits of water ice. Studies suggest ice could
1

Figure 1.1: The Lunar south pole has areas cold enough to sustain water ice (shown red
through blue) even in accessible areas well outside of permanently shadowed craters (outlined in
white) [23]. Excavating down to these resources can uncover them for direct scientific measurement, characterization, or mining.
be found in accessible areas (outside crater rims) at depths of only tens of centimeters [23, 59].
Excavating down to these deposits can uncover them for direct scientific measurement, characterization, or mining. Characterizing and mapping these ice resources is another important goal
for ISRU [58], and Astrobotic Technology Inc. and Shackleton Energy Company intend to mine
these resources. Moon Express aims to mine platinum on the lunar surface.
The tasks requiring excavation on the Moon and Mars thus span mining, earthworking for
infrastructure, as well as direct scientific inquiry. Figure 1.2 and Figure 1.3 show visualizations
of some of these various excavation tasks.

1.2

The Load-Haul-Dump cycle at the core of excavation tasks

Regolith requires varying degrees of processing depending on the application. Three classes of
task are distinguished here based on this degree of processing.
2

Figure 1.2: Conceptual discrete excavation robot (with a front-loading bucket) building outpost
infrastructure on the Moon

Figure 1.3: Conceptual continuous excavation robot (with a bucket-wheel) digging a trench on
the Moon, while collecting regolith

1. Displacement. For some tasks the only requirement is moving the regolith out of the
way. These tasks (or sub-tasks) include removing overburden from buried ice, trenching
to expose stratigraphy, and digging holes to emplace mission elements.
2. Shaping. For another class of tasks - which includes building berms, covering habitats,
burying emplaced assets - the regolith is the desired material. Once moved into place, additional processing consists merely of physically shaping, molding, and perhaps compacting
or sintering the regolith.
3. Refinement. For some tasks, the desired resource makes up only a fraction of the regolith, and processing is required to refine and extract the resource. This includes oxygen
extraction as well as mining for ice or platinum.
Moving regolith from one place to another, in a load-haul-dump cycle, is central to the first
two of these task classes. The degree to which load-haul-dump is also central to the third depends
on where processing occurs. Options include a central processing plant designed to accept raw
regolith, a central plant that accepts beneficiated regolith (i.e. pre-processed to increase the
concentration of the desired resource), or a plant that runs entirely on the excavator. Onboard
processing introduces significant mass and extreme thermal requirements; oxygen extraction, for
example, requires heating regolith to between 900 C and 1600 C [58]. It is assumed that, for
these reasons, onboard processing will not be incorporated into excavators themselves during
prototypical excavation missions. Load-haul-dump is thus a paradigm that encompasses the key
aspects of all relevant regolith excavation tasks.

1.3

Lightweight is low mass in low gravity

Light weight can be attributed to low robot mass, reduced gravity, or both. In any space mission,
mass is always at a premium because it is the main driver behind launch costs. Small excavators
that can achieve mission goals are preferable to larger ones. Low mass machines operating in
4

Figure 1.4: Contraints of planetary excavation impose unique engineering challenges

reduced gravity (1/6 of Earth gravity on the Moon, 1/3 on Mars or Mercury) have limited weight
available to produce traction or plunge tools into regolith. Traction and plunge force are limited
to a fraction of robot weight. Figure 1.4 expresses how low mass and reduced gravity leads to low
traction and plunge force. Engineering challenges associated with lightweight excavation necessitate a rethink of excavation configurations, possibly beyond the dozers, loaders, and excavators
typical in terrestrial applications [12].
Excavation missions started small. For example, the Surveyor, Viking, and Phoenix landers
gathered samples of a few cubic centimeters at depths of a few centimeters with scoops mounted
to relatively heavy landers (see Figure 1.5). Next missions will likely escalate to excavating cubic
meters worth of regolith at depths of 10s of cms using lightweight mobile robots (e.g. digging
down to expose and collect water-ice in polar regions of the Moon). Finally, excavation will scale
up to production machines for ISRU. Having a configuration that scales well with increasing size
and mass allows subsequent missions to re-use existing technology, learn from past difficulties,
and reduce risk. This principle is exemplified in the similarities between Mars Sojourner, the
subsequent MERs, and now the Mars Science Laboratory (MSL), as seen in Fig. 1.6. Robotic
5

Figure 1.5: Lunar and Martian landers to date have gathered only small samples with scoops
mounted to relatively heavy landers. Surveyor (top left) on the Moon with scoop extending right
on scissor arm; Viking (top center) model with excavation boom deployed; Phoenix (top right)
artists concept; Trenches dug by each (bottom row), respectively [NASA].
excavators with high productivity across a range of light weights are essential.

1.4

Definitions of important terms and concepts

Before addressing the problem posed by lightweight robotic excavation, a few additional concepts and terms are defined.

1.4.1 Continuous vs. discrete excavators


Excavators can be classified as continuous or discrete, describing interactions with the soil while
taking multiple cuts.
Continuous excavators stay continually in contact with the soil as they take multiple cuts.
This necessitates having multiple cutting surfaces; by the time each surface or bucket has ac6

Figure 1.6: Configurations that scale well from small initial missions reduce risks in subsequent
missions. Sojourner (center), MER (left), and MSL (right) share a common suspension configuration for this reason [NASA JPL].
cumulated an appreciable amount of soil it clears the ground and the next, soil-free, bucket has
already started cutting. Continuous excavators include bucket-wheels, bucket-chains, and elevating scrapers.
Discrete excavators are those that must break contact with the soil before starting a new cut;
between cuts, the excavator may need to dump its load or clear the cutting surface, for example.
These excavators fill one large bucket with a single cut; the cutting edge has an ever-growing
accumulation of soil as the bucket is filled. Discrete excavators include front-end loaders, dozers,
mining shovels, and open bowl scrapers.
Figure 1.7 shows examples of both continuous (left box) and discrete (right box) mobile
excavators. The taxonomy lines in the figure also note another way of subdividing the excavators,
namely into trenchers, scrapers, and front-end loaders/pushers.
In this work, a discrete excavators blade or bucket (filled directly by the act of cutting)
is assumed to be the excavators only vessel for collecting and transporting regolith. Discrete
excavators that transfer load to a secondary collection bin (i.e. a dump-bed) are considered
separately in Appendix A.
7

8
Figure 1.7: Taxonomy of mobile excavators, with continuous excavators shown in the blue box (left), and discrete in the yellow
(right). The upper and lower rows show parallels between terrestrial and planetary machines, respectively.

1.4.2 Payload ratio


Payload ratio is the ratio of weight of regolith payload collected to empty robot weight; it is a
measure of pound-for-pound regolith moving capacity that turns out to govern the productivity
of lightweight robotic excavators. Terrestrial loaders and scrapers attain payload ratios as high as
80% to 100% [35]. Space systems are subject to additional constraints that make it challenging
to attain payload values that high; a payload ratio of 50% is considered relatively high in this
context. In this thesis, the non-dimensional quantity payload ratio is denoted P .

1.4.3 Excavation thrust


Excavation thrust is the force supplied by an excavator that is available for cutting soil. In
this work, excavation thrust is assumed to be provided by traction, as excavator configurations
typically considered for space applications cut by driving forward. Alternate modes of providing
excavation thrust, such as resisting articulation forces using a static base or using excavator
weight directly to cut vertically down, are considered separately as extensions to this work, in
Appendix A.
A vehicles drawbar pull is the net traction available for doing work, and is dependent upon
slip (or travel reduction, a caveat explained in Chapter 4). Drawbar pull at 20% slip is a good
measure of tractive performance, as pull begins to plateau around 20% slip for many wheels (or
tracks) while negative effects such as sinkage increase [68]. A non-dimensional quantity, P20 /W
(Drawbar pull at 20% slip, normalized by weight), has been used as a benchmark metric for lunar
wheel performance from the times of Apollo [24] to today [70, 80].
In this thesis, excavation thrust refers to drawbar pull at 20% slip, because of the assumption
of tractive thrust; it is denoted P20 . The non-dimensional ratio of excavation thrust to weight
is defined as the excavation thrust coefficient, and is denoted T. Under the tractive thrust
assumption, T = P20 /W .
9

1.4.4 Excavation resistance


Excavation resistance is the force imparted on an excavator by cutting and collecting soil. Only
the forces during the cut (once the excavation blade is already in the ground) are treated explicitly.
Penetration forces are neglected but, as Blouin assumes these forces are of the same nature as
the cutting forces [11], they can be treated as analagous to cutting forces and subsumed by them.
Crucially, resistance introduced by soil accumulation at a buckets cutting edge (which increases
the force required to move additional soil into the bucket) is accounted for as part of excavation
resistance.
Excavation resistance force is denoted Fex in this work. The non-dimensional ratio of excavation resistance to empty excavator weight is defined as the excavation resistance coefficient,
and is denoted F = Fex /Wrobot .

1.5

Scope

This thesis considers excavation tasks that involve load-haul-dump, with some additional processing such as shaping, compaction, and beneficiation treated as extensions to this central task.
Mining robots that fully process resources onboard are outside the scope of this work.
This work deals primarily with excavators that produce thrust for cutting by developing traction, with other sources of excavation thrust treated as extensions in Appendix A.
Excavation in gravity between 1 and 1/6 that of Earth is considered, to cover a range that
includes Earth, Mars, Mercury, and the Moon. Digging on asteroids is outside the scope of
this work. The range of robot mass considered in this work spans from 30 kg to approximately
300 kg. Larger, more massive, machines are unlikely to satisfy mass budgets of near-future
excavation missions. As machines get smaller than 30 kg or so, baseline components that do not
scale well (like computing and communications) take up an ever larger proportion of the mass,
leaving little room for productive excavation tooling. The scope of this work covers a range of
10

weights that is relevant across multiple space mission scenarios.

1.6

The problem of distinguishing productive lightweight excavator configurations

Excavation using lightweight machines is problematic because light weight puts severe limits on
forces an excavator can effect for traction and plunging tools into soil. Excavators for building
infrastructure and mining resources on the Moon and Mars will necessarily be lightweight, because they will be low mass machines (in space missions mass is always at a premium) operating
in reduced gravity.
No prior methodology exists for developing or even evaluating robotic configurations that are
lightweight and yet still productive.

1.7

Thesis Statement

This thesis substantiates that continuous excavators maintain high productivity at light weights,
where productivity for discrete excavators declines. All excavators have a lightweight threshold
in the operating space, below which their productivity is limited. This threshold is crossed at
lower weights for continuous excavators than discrete excavators. The lightweight threshold is
described by a non-dimensional quantity that relates payload ratio, excavation resistance, and
excavation thrust.

1.8

Overview

The remainder of this document is organized as follows:


Chapter 2 presents related work in lightweight excavation. This includes excavator configuration trade studies and a variety of prototypes. Proposed methods for reducing excavation
11

resistance, including percussion and raking, are discussed. The utility and limitations of analytical models and experimental techniques commonly used for lunar excavation research are
explored.
Chapter 3 shows that the haul stage of load-haul-dump cycles governs excavator productivity,
based on experiments and simulations. The resulting importance of payload ratio is discussed, as
are underlying assumptions regarding excavation thrust and resistance that underpin these results.
Excavation thrust, excavation resistance, and the relation of these terms, are the subject of
Chapter 4. This chapter also discusses how excavation resistance varies during excavation depending on excavator configuration (i.e. continuous vs. discrete excavation). Excavation resistance due to soil accumulation is explored. Analytical models are used to extend results to low
gravity environments, and these results are compared to the limited low gravity experimental
data available.
Chapter 5 develops the lightweight threshold, combining the concepts from the previous
chapters. Experimental results of excavation operations below and above this threshold are presented. The limitations of performing lightweight excavation experiments on Earth are discussed.
Chapter 6 presents practical considerations for implementing continuous excavation in space.
This is done in the context of the development of a prototype for a novel lightweight bucket-wheel
excavator robot.
Chapter 7 summarizes the major conclusions and contributions of this thesis, and proposes
relevant future work.

12

2 Background and Related Work


Literature related to lightweight robotic excavation research includes the principles of excavation
and attempts to model its mechanics, experimentation in analogue lunar and planetary conditions,
trade studies and prototypes of lunar excavator configurations, soil loosening methods, and automation of earthmoving and mining equipment.

2.1

Fundamental mechanics of excavation

The mechanics of excavation are based on the principles of passive earth pressure, adapted from
the design of retaining walls, as shown in Figure 2.1. Reece presents the following as the fundamental equation of earthmoving mechanics [32]:

PEx = N gd2 + Nc cd + Nq qd + Na Ca d

(2.1)

where PEx is excavation resistance force per unit width, and the four terms of the summation
represent (in order) forces due to frictional shearing (i.e. gravity), cohesion, surcharge, and soiltool adhesion. Inertial forces are explicitly ignored, as low cutting speed is assumed. The Ni are
non-dimensional coefficients pertaining to each of the four sources of force, respectively. Gravitational acceleration is denoted g, is soil density, d is cut depth, c is cohesion, q is surcharge
pressure, and Ca is soil-tool adhesion. The equation is for cutting with a flat plate. As this is
a two-dimensional formulation, a first order estimate of excavation resistance force for a cut of
13

Figure 2.1: Mechanics of excavation based on passive earth pressure [32]


finite width can be made by multipyling by said width, w:

FEx = wPEx

(2.2)

A wide variety of models have been investigated for their potential applicability to planetary
excavation [26, 39, 73, 74]. However, at their root, they are all just variations of Reeces fundamental equation (with the possible exception of Luth & Wismer). Models vary in which force
terms they do and dont include. Several models omit tool-soil adhesion and/or surcharge forces.
Some include inertial forces, which Reece explicitly omitted. Table 2.1 lists the array of models
and shows which force terms they include. Additionally, the models vary in their definitions of
the Ni coefficients.

2.1.1 Gravity and cohesion forces included in all excavation models


Excavation shears soil, and a soils shear strength is governed by its internal friction angle and
cohesion. These shear strength contributions are modelled for excavation resistance by gravity
and cohesion terms, respectively. All the models listed in Table 2.1 include at least some form
14

Model
Reece
Osman
Gill
Luth & Wismer
Godwin
Balovnev2
McKyes / Swick
Qinsen
Willman
Zeng

Gravity
X
X
X
X
X
X
X
X
X
X

Cohesion
X
X
X
1
X
X
X
X
X
X

Surcharge
X
X

X
X
X
X3

Adhesion
X
X

X
X
X

Inertia

X
1

Table 2.1: Models vary in which force terms they include, but gravity and cohesion are always
considered. 1 In Luth & Wismer, cohesion and inertia terms are multiplied by gravity terms,
rather than added to them. 2 Balovnev includes additional terms to account for sidewalls and a
blunt cutting edge. 3 Qinsen models a curved bulldozer blade, and explicitly models surcharge
due to soil accumulation. 4 Zeng treats acceleration directly, rather than inertia.

of these two terms, implying that their contribution to total excavation resistance is of primary
importance. In fact, Wilkinson and DeGennaro show that, for the McKyes / Swick model (which
includes all five typical terms), the gravity term (referred to as the depth term in their paper)
and/or cohesion are the dominant contributions to total excavation resistance force over a very
broad range of operating conditions [73].

2.1.2 Adhesion and inertial forces can usually be neglected


Compared to gravity and cohesion, adhesion and inertial forces tend to have minimal contribution
to excavation resistance force. Hettiaratchi and Reece note that the Na coefficient (for adhesion)
is small compared to the other Ni and that soil-tool adhesion is almost always smaller than
cohesion; they neglect inertial forces outright, arguing that cutting speeds are typically low [32].
Table 2.1 shows that adhesion and inertial terms are the two most often omitted from excavation
resistance force models.
15

2.1.3 Surcharge forces arise due to soil accumulation


The surcharge term can be used to account for soil that accumulates at the front edge of the bucket
during cutting. The surcharge increases as cutting proceeds. To account for this, Shmulevich [61]
models surcharge as:
q gx

(2.3)

where x is cut advance distance (and is soil density). Kobayashi [42], making different as
sumptions about the shape of the accumulating pile, proposes: q g dx where d is cut depth.
In both cases, surcharge increases with cut advance distance, linearly in the former and as the
square root in the latter. In both cases, the surcharge force is assumed to be only due to the
additional weight causing increased frictional shearing.
Qinsen [57], when modeling excavation with a bulldozer blade, accounts for soil accumulation directly. They model forces at steady state once soil accumulation has reached maximum
extent. The model considers not only the weight and frictional shearing of the cut soil, but also
its cohesion.
As discussed in Section 1.4.1, soil accumulation is particularly notable for discrete excavators such as front-end loaders and bulldozers. Section 2.2.1 discusses experimental results that
demonstrate how excavation resistance increases during discrete excavation, and relates these
results to the models discussed above.

2.1.4 Discrete Element Models for excavation


Discrete Element Modeling (DEM) provides greater promise for high fidelity modeling of excavation. This approach explicitly models interactions between particles, and produces resultant
flow fields and stresses for these soil particles. DEM could therefore model how soil flows and
accumulates in a bucket. The goal of current research in DEM [9, 69] is to produce excavation
flow fields as well as calibrated resultant forces. Experiments providing quantified visualizations
16

of excavation will drive development and validation of DEM. Bui et al [14] have performed soil
footing failure experiments in reduced gravity to provide data for tuning their DEM model for
excavation.
This modern approach is still being developed, and is not ready to incorporate into system
development optimizations, let alone online prediction and control. DEM development is a field
of research in its own right, and is outside the scope of this work.

2.2

Experimentation for Lunar and Planetary Excavation

Classical excavation experiments pull blades and buckets through soil bins, measuring how excavation resistance and other variables are affected by changes in excavation parameters; a recent
example is work at NASA Glenn Research Center [2]. Controlled soil bin experiments have also
been conducted with bucket-wheels [36].

2.2.1 The large impact of soil accumulation on discrete excavation


Agui shows that horizontal excavation resistance rises approximately linearly with cut distance,
as soil accumulates in a bucket [2]. These results agree with the general modeling assumptions
of Shmulevich presented in Section 2.1.3. Agui also showed though, that the shape and location
of a pile accumulating in a bucket is non trivially dependent upon time as well as cut depth, cut
angle, and possibly other parameters. Modeling soil accumulation in a bucket by a continuously
changing surcharge distribution is therefore difficult, and ideally would depend on knowing how
the soil flows as it enters the bucket.
A bulldozer blade also exhibits significant increase in horizontal force as surcharge increases
with cut distance, as demonstrated by King [39]. Comparing a variety of excavation models to
their data, they conclude that Qinsens model provides the best fit. This is not entirely unexpected, as Qinsens model was developed specifically for bulldozing (though one of the other
17

Figure 2.2: Gravity offload: a cable pulls up on an excavator with 5/6 its weight to simulate lunar
gravity

models compared against was specific to bulldozing as well).

2.2.2 Soil properties and gravity are important conditions to control


Controlled planetary excavation experiments make use of simulants that mimic the geotechnical
properties of Lunar or Martian soils. GRC-1 and GRC-3 are lunar simulants with properties
relevant for excavation [56]. JSC-1 is another lunar simulant often used for excavation experiments [18, 74]. JSC-1 has a particle size distribution that is similar enough to lunar regolith to
duplicate its compaction and relative density [83].
Simulating low gravity conditions is another important consideration for lightweight excavation experiments. Boles [13] showed that excavation resistance in 1/6 of Earth gravity (experienced during reduced gravity flights) could be anywhere between 1/6 and 1 of the resistance
experienced in full Earth gravity. Sample data shows excavation forces in 1/6 g that average 1/3
of the resistance in full Earth gravity.
Another way to simulate low gravity conditions (at least for the excavator if not the soil) is to
use a gravity offload mechanism. No excavator testing with gravity offload has been reported in
the literature to date.
18

2.3

Applicability of excavation resistance models to planetary


excavation

A common result from literature that attempts to compare excavation forces predicted by various
models (e.g. [39, 73, 74]) is that the models yield disparate predictions. This makes it inprudent
to rely on any one model for estimating excavation forces. As Section 2.1 showed, however, the
models share common fundamentals that are instructive when investigating planetary excavation.
Any estimate of excavation resistance must take into account soil weight (and thus friction) and
cohesion. Surcharge is also very important, particularly for discrete excavation; the weight, and
perhaps cohesion, of the accumulating soil comprise this surcharge.
Muff [53] reports that the Luth & Wismer model was tested against Martian telemetry from
the Viking sampling digs (a claim seemingly based on personal correspondance with those who
performed the analysis), giving this model flight heritage in a sense. The lack of published
quantitative comparisons, however, compels caution in interpreting this claim.

2.4

Lunar excavation trade studies

Trade studies have examined the applicability of various excavation robot options, specifically
for lunar outpost site work. However, these studies assume several metric tons are available for
excavation equipment; this is unlikely to be the case in the short or even medium term. The trade
studies also restrict themselves to predefined configuration options, potentially missing novel
designs that could fare better than those considered.
Boles et al. [12] compares the probable required launch mass of several construction machine suites. The study concludes that typical terrestrial excavation machines would not be as
effective as tripod cranes, sweeper leveler/excavators, and other innovative vehicles. Abu El
Samids work [1] continues along the lines of Boles, concentrating on tradeoffs between autonomous and tele-operated operation and between single vehicle and team configurations. A
19

Configuration options
Boom cranes, track
dozers, haulers, drills,
clamshell
diggers,
sweeper
excavator/levelers
Same as Boles (controlled manually), or
teams of autonomous
bulldozers,
bucket
loaders, or bucket
wheels
Multipurpose excavator, auger, bucket
ladder, bucket wheel,
dragline,
overshot
loader,
pneumatic
vacuum, scraper

Metrics
Launch mass

Launch mass

Productivity, reliability, dust generation,


power
efficiency,
maintainability

Selected option
All-purpose
super
cranes with drilling,
excavating,
leveling,
and hauling
capabilities
Team of autonomous
bulldozers

Ref.
[12]

Multipurpose excavator

[51]

[1]

Table 2.2: Trade studies examining options for large-scale lunar excavation (using several metric
tons of equipment) arrive at different conclusions, demonstrating the weakness of approaching
such a complex problem with a predefined set of solutions to choose from.

team of autonomous bulldozers is recommended for the task of berm building. Mueller and
Kings study [51] scores excavator designs on a number of quantitative and qualitative metrics
and decides a multi-purpose machine with bulldozing blade and excavator arm is most appropriate for lunar site work. The results of these trade studies are summarized in Table 2.2.
The aforementioned trade studies restrict themselves to predefined configuration options and
compare their relative merit for lunar operations; in that sense, they espouse a top-down approach
to configuration analysis. Each of the studies arrives at different conclusions regarding robot
designs. The varying results highlight effects of differing assumptions, models, and metrics
when approaching such a complex problem with a predefined set of solutions to choose from.
Assumptions of high mass machinery, as well as wide variability of the results, limit the
relevance of past trade studies to the development of lightweight robotic excavators. Metrics for
comparison of configurations in these studies are useful to consider, but the top-down approach
20

of studying a predefined set of solutions is not as useful.

2.5

Lightweight excavator prototypes

In recent years, several robot prototypes have been developed specifically for lunar excavation
and ISRU. There are tested, however, in full Earth gravity, so principles of lightweight excavation
are obscured. The taxonomy of mobile excavators introduced in Section 1.4.1 can be applied to
these robots as well, as seen in Figure 1.7. The figure shows samples of each of the following: a
bucket-wheel excavator, a bucket-ladder scraper, an open bowl scraper, as well as a loader and a
dozer.
Bucket-wheel excavators produce low resistance forces suitable for lightweight operation [36].
A past lunar bucket-wheel excavator prototype [52] has been configured like a trencher (see Figure 1.7). However, the small scale intended for the lightweight excavator made material handling
and tranfer prohibitively challenging [37]. A novel lightweight bucket-wheel excavator, with a
simplified material transfer approach, has been developed as part of this work and will be discussed in detail in Chapter 6. A Bucket-Drum Excavator, which is an adaptation of a bucket
wheel [17], has a novel regolith collection system with cutting buckets mounted directly around
the outside of the collection drum. Regolith Advanced Surface Systems Operations Robot (RASSOR) has counter-rotating front and rear bucket drums, making it possible to balance horizontal
excavation forces [50]. Figure 2.3 shows a Bucket Drum Excavator as well as RASSOR.
Due to past difficulties encountered transferring regolith from bucket-wheel to collection
bin, bucket-ladders have gained favor [37]. Bucket-ladders use chains to move buckets along
shapeable paths, easing transfer to a collection bin. Winners of the NASA Regolith Excavation
Challenge and subsequent Lunabotics mining competitions (competitions where lightweight excavators must collect as much regolith simulant as possible in 15 to 30 minutes) have all employed bucket-ladder trenchers driven by exposed chains or flexible conveyors. However, exposed chains and conveyors fare poorly in harsh lunar regolith and vacuum, making them inap21

Figure 2.3: Adaptations of bucket-wheel excavation: Bucket Drum Excavator (left) and RASSOR (right)

Figure 2.4: Juno rover with a small load-haul-dump scoop that achieves only low payload ratio.
propriate for operation on the Moon.
Cratos [16] is an open bowl scraper with a central bucket between its tracks, as seen in
Figure 1.7. It can carry a payload ratio of approximately 30% (in Earth gravity). Although
terrestrial scrapers buckets extend laterally beyond the outside of the wheel track, the central
bucket mounting is a key feature that leads to Cratos being classified as a scraper here. Juno
rovers [67] can be equipped with front-end load-haul-dump scoops, though these scoops can
carry only a small fraction of the rovers mass in regolith (see Figure 2.4).
Other lunar and planetary excavator prototypes include NASAs Chariot with LANCE bulldozer blade and Centaur II with front-loader bucket. These machines are very high mass (on the
order of tonnes) and low payload ratio, making their relevance to lightweight excavation mis22

sions limited. Robots that excavate by filling up with regolith as they burrow into the ground
have also been proposed [44].

2.6

Soil loosening methods and mechanisms

Lunar regolith is very strong below the top few centimeters from the surface [30]. The presence
of ice only makes this dense mass harder and more cohesive [27]. This has led researchers to
develop several methods to loosen regolith either prior to or during excavation.
Sture et al [40, 66] as well as Zacny et al [18] have shown that percussive/vibratory actuation
of diggins implements reduces excavation resistance forces. Specifically, percussion reduces
the shear strength of dry soil by removing the effects of soil dilatancy from the internal friction
angle along the shear failure boundary layer [28]. To date, the advantages of percussion have
been studied for bulldozers, small narrow scoops, and helical augers.
Gertsch et al. [29] have studied the applicability of cutterhead wheels and rippers for loosening frozen and compacted regolith in preparation for excavation. Iai showed that adding ripping
reduces total excavation energy (ripping + excavation) in soils with high density and low gravel
content [34]. An important contribution of Iais work is raising awareness of the often overlooked
contribution of gravel and rock content to excavation forces.
Bernold has suggesting using small explosive charges to loosen compacted regolith [8]. The
fact that these explosives are a consumable that cannot be manufactured in situ, though, limits
their applicability to ISRU missions.

2.7

Autonomous Earthmoving and Tele-Operation

The automation and tele-operation of earthmoving machines is a research field in its own right.
Singh [62] lists a taxonomy of the fields inter-related aspects: sensing, kinematic and dynamic
modeling, soil-tool interaction modeling [45], tool trajectory planning and control, and tele23

operation.
Dunbabin [22] investigates operating large-scale excavation machines in extra-terrestrial environments, and discusses operating modes ranging from manual, through various levels of abstracted tele-operation (remote, fly-by-wire, and copilot), to autonomous. Autonomous dig and
dump cycles are demonstrated (on Earth), with the goal of shifting as much control as possible
to the robotic excavator to avoid tele-operation challenges such as dealing with latency.
A theoretical lower bound on the round trip time of communications between the Earth and
the Moon, based on the speed of light and lunar perigee, is approximately 2.5 s. Even this amount
of latency makes direct remote control a psychologically tiring task for any expert operator,
which can greatly hamper the productivity of even the most capable machines [60].
The Lunokhod rovers were commanded directly via remote tele-operation from Earth. Despite the taxing effects of latency, the rovers regularly drove at speeds of 1 km/hr [38]. Of course,
the remote operators did not deal with any excavation tasks as the Lunokhods were not equipped
for them.
This work investigates aspects that arise when tele-operating bucket-wheel excavators. One
of the guiding principles is that continuous excavator configurations should lead to simpler control than discrete wide bucket excavators. A generalized investigation of autonomy for earthmoving equipment, beyond reviewing the literature, is outside the scope of this work.

2.8

Conclusions Based on Related Work

Review of literature related to lightweight robotic excavation leads to the following conclusions:
There is no consensus on appropriate excavation force modeling for lunar excavation. However, it is instructive to rise above the fray of contrasting models and focus on their commonly
shared features. Any estimate of excavation resistance must take into account soil weight (and
thus friction) and cohesion. Surcharge is also very important, particularly for discrete excavation; the weight, and perhaps cohesion, of the accumulating soil comprise this surcharge. These
24

common features provide a theoretical framework for broadly predicting dependence on key
variables such as soil density and cohesion as well as gravity and cut depth.
Excavation resistance varies significantly during a cut as soil accumulates in the bucket, and
classical models can only approximate this effect. They fail to capture excavation soil flows.
Modern Discrete Element Modeling (DEM) shows promise in modeling excavation soil flows.
Past experiments have studied the effects of many excavation parameters, and have shown
that bucket-wheel excavators produce low resistance forces suitable for lightweight operation.
Only preliminary efforts have been made to study excavation forces in reduced gravity. Experiments with excavator prototypes simulating low gravity constitute a novel contribution to the
field of study.
The wide variability in configurations resulting from lunar excavation trade studies and prototype developments highlight the lack of consensus on appropriate configurations for lightweight
excavators. An anecdotal consensus is the fact that bucket-ladder trenchers have won the Regolith
Excavation Challenge and Lunabotics mining competitions each of the 4 times such competitions
were held [49].

25

26

3 Hauling and Payload Ratio


The load-haul-dump cycle is central to lightweight robotic excavation tasks, as described in Section 1.2. This chapter will show that, for a nominally capable excavator, hauling productivity
dominates overall task performance. Payload ratio directly influences hauling productivity, making it an important design parameter.
Section 2.4 showed how excavator configuration trade studies utilizing a top-down approach
(i.e. comparing a predefined set of solutions) have produced widely varying results, limiting their
usefulness.
This work explores configurations for lightweight robotic excavators from the bottom up,
starting with system parameters that figure into analytical models of excavating and driving,
synthesizing them for analysis of task-level performance metrics. This approach distinguishes
design parameters (such as driving speed, payload ratio, or number of wheels) of appropriate
excavator configurations instead of picking between configurations themselves.

3.1

Task-level site work modeling

Regolith-moving machines are commonly characterized for elemental actions like digging or
driving [73], but it is also important to measure comprehensive performance combining digging
and driving. A task model is developed here for excavation tasks that includes digging, transporting, dumping, and shuttling for recharge (See Fig. 3.1).
The REMOTE (Regolith Excavation, MObility & Tooling Environment) task simulator [65]
27

Figure 3.1: Comprehensive task modeling for lunar site work that combines elemental actions of
digging and driving
, computes metrics including task completion time, production ratio (weight of regolith moved
per hour, normalized by robot weight), and production efficiency (weight of regolith moved per
unit of energy spent, normalized by robot weight), based on parameters describing the task, the
robotic system, and the environment. The novelty of comprehensive task simulation, combined
with sensitivity analysis, is that it identifies system parameters that are important for overall task
success. This determines what matters most for system design and tradeoffs.
Traction and excavation forces are modeled to determine admissible bucket geometries, and
transport and recharge times are estimated based on driving speed and power draw. Excavation
is assumed to occur on approximately flat ground (i.e. not digging on a large uphill or downhill
slope).

3.1.1 Traction modeling (wheels)


The underlying traction model is that of Bekker [7] and Wong [76], based on their empirical and
theoretical work. Net traction, also known as drawbar pull (DP), is obtained by calculating wheel
resistance and thrust.
28

Wheel resistance is assumed due to soil compaction. Gravitational resistance is ignored


because of the assumption of excavating on relatively flat ground. Bulldozing resistance is also
ignored; wheel bulldozing can be avoided with careful grouser design, as shown in Appendix B.
Following Bekker, compaction resistance of a single wheel, Ri , is estimated as:

Ri = b



kc
+ k
b

zin+1
n+1

Soil pressure-sinkage parameter values are based on estimates made for lunar regolith [30]:
kc = 1.4 kN/mn+1 , k = 820 kN/mn+2 , and n = 1. Wheel width is denoted b, and sinkage,
zi , is estimated as:


3Ni

zi =
b(3 n)(kc /b + k ) 2r

2/(2n+1)

where Ni is the normal load on a given wheel, and r is wheel radius. Slip sinkage is ignored,
for the sake of simplicity. New work in terramechanics [20] is developing modeling techniques
for slip sinkage which could be incorporated into future modeling work.
Wheel thrust, Hi , is estimated based on equations (and assumptions) presented by Bekker
and Wong:
Z0
Hi = rb (c + ((kc /b + k )(r(cos cos 0 ))n ) tan )
0

(1 exp(r/K[0 (1 j)(sin 0 sin )])) cos d


where c and are soil cohesion and internal friction angle, respectively, K is a shear deformation constant, j is wheel slip, and 0 = cos1 (1 z/r) is the angle from vertical to where
the wheel rim contacts level terrain, as shown in Fig. 3.2.
Within REMOTE, vehicle load is assumed to be evenly distributed between all wheels, so
drawbar pull is calculated as:
29

Figure 3.2: Wheel geometry terms. Wheel width, b, is into the page.

DP = Nw (Hi Ri )
where Nw is the number of wheels.

3.1.2 Excavation models


There is no consensus excavation resistance force model for lunar excavation, as discussed in
Chapter 2. REMOTE offers a choice of two underlying excavation models, Balovnev and LuthWismer. Balovnevs [4] is a 3-D bucket model developed from theory. It is of the fundamental
form proposed by Reece (discussed in Section 2.1). Luth-Wismer [46, 75] was developed empirically from separate experiments in cohesive clay and cohesionless sand. The Luth-Wismer
model represents an excavating bucket by a single plate, and may have been tested under Martian
conditions during the Viking missions [53]. The same parameters govern productivity, independent of the choice of model, as will be shown in Section 3.1.5.
Horizontal excavation resistances modeled by Luth and Wismer for (cohesionless) sand and
(cohesive) clay are:

FH,sand = gwl1.5

1.73

d
l sin

0.77
30

"

#
 1.1
v2
d
1.05
+ 1.26 + 3.91
w
gl

Figure 3.3: Excavation geometry terms. Bucket/plate width, w, is into the page.

FH,clay

1.21
d
d
= gwl
l sin
"
#
!
1.21  0.121
 0.78
11.5c
2v
v2
d

+ 0.065 + 0.64
0.055
gd
3w
w
gl
1.5 1.15

Bucket width is denoted w, cut depth is d, is the angle of the buckets cutting face (relative
to horizontal), and l is the length of cutting face interacting with the soil, as seen in Fig. 3.3. In
REMOTE, l is defined by d/ sin to avoid overconstrained geometry. The buckets horizontal
cut velocity is denoted v, and gravitational acceleration is g. Soil density is denoted , and c is
soil cohesion. Luth-Wismer does not explicitly include soil friction angle or external (soil-tool)
friction.

Balovnevs model includes typical force terms due to weight/friction, cohesion, and surcharge. It also includes additional terms: external friction contributes resistance on the bucket
sidewalls, and cutting edge thickness is also taken into account. The horizontal component of
excavation resistance is given by:
31

FH



1 sin
dg
+ c cot + gq + B (d l sin ) g
= wd(1 + cot tan )A1
2
1 + sin



eb g
1 sin
+ web (1 + tan cot )A2
+ c cot + gq + d g
2
1 + sin



dg
1 sin
+ 2sdA3
+ c cot + gq + B (d ls sin ) g
2
1 + sin



1 sin
dg
+ c cot + gq + B (d ls sin ) g
+ 4 tan A4 ls d
2
1 + sin


Common parameters are denoted the same as on page 31. The soil internal friction angle is
denoted , and is the external (soil-tool) friction angle. Surcharge is denoted q. Bucket side
thickness is s, side length is ls , eb is blunt edge thickness, and b is blunt edge angle. Ai are
non-dimensional coefficients specific to the model [4], and B is a boolean flag indicating if the
bucket is fully buried below soil level.
Within REMOTE, excavation is assumed to occur over a short distance, so that cut depth and
cutting face length do not change substantially. By this same assumption, traction parameters
that might in reality vary with time, such as slip, also remain constant for the duration of an
excavation cut. For longer cuts, one could account for soil accumulation by making surcharge
and/or cut depth depenedent on horizontal cut progress (and thus time).
Excavation with a forward-facing bucket is assumed, meaning the excavator can generate and
sustain an excavation force no greater than its net traction, or drawbar pull. Excavation at this
stall condition is subject to:

FH = DP
This equation is solved, by defining all parameters but one (for example, bucket width), to
find an admissable bucket geometry. The bucket is assumed to be of equilateral triangular prism
shape, as seen in Fig. 3.4. Combining this assumption with a bucket filling efficiency, b , gives
32

Figure 3.4: Bucket geometry with equilateral triangular prism shape


the volume of soil that can be excavated in a single cut:

1
V = b wl2 sin(/3)
2
To account for excavators that have secondary collection/dump beds, an overall payload capacity can be defined. In that case, several cuts may be required to reach capacity, and REMOTE
accounts for the time required for all of these cuts as well as the time for transfers from primary
bucket to collection bed.

3.1.3 Operations modeling


Traction and excavation modeling describe the dig portion of a task, but as Fig. 3.1 shows,
a general task also includes transporting and dumping regolith. To account for these aspects
of tasks, REMOTE includes operational parameters such as average distance between dig and
dump, driving speed, area and depth of the desired excavation, and operational efficiency (percentage of time spent actually performing work, as opposed to waiting for commands or performing computations).
The number of robots performing a task, and the mass of each robot, are further system
design parameters.
33

3.1.4 Power modeling


Energy is expended by both driving and excavating. There is also baseline power that is always
being dissipated in communication, computation, and other avionics tasks, even when not performing physical work. Over the class of small vehicles studied (100 kg to 300 kg), this baseline
power is assumed to be the same for each vehicle. Only steady state power is considered during
each phase of a task.
Power expended during driving is modeled by:

Pdrive = KP d mgvd

Where m is vehicle mass, g is gravitational acceleration, vd is driving velocity, and KP d is a


driving power coefficient. The KP d coefficient captures and sums several sources of power dissipation. Power required to overcome wheel rolling resistance can be estimated as a percentage of
vehicle weight [55]. Internal machine losses (in bearings, for example) are also proportional to
weight (acting as a radial load). Even undulations in the terrain can be captured by multiplying
weight by the sine of a representative terrain angle. KP d can thus be used to account for rolling
resistance, internal losses, and terrain losses.
Excavation power draw is modeled by:

Pexcav = KP ex FH vex

Where FH is excavation resistance force, vex is excavation velocity, and KP ex is an excavation


power coefficient that is nominally 1. Driving power is also expended (with vd = vex ) during
excavation.
Dumping power is ignored, as dumping comprises a very small portion of the overall task.
Batteries are assumed to be the primary power source for excavation robots. Each vehicle
is assumed to have a constant fraction of mass budget for batteries, meaning larger vehicles are
34

able to store more energy than smaller ones. A battery charging time is included in the model.
This charge time does not include the time required to shuttle to and from the power plant, which
is accounted for separately in the same way that shuttling to and from a digging site is.
Batteries can potentially be charged during operation by additional power sources such as
onboard solar panels. Such an additional power source is modeled as a negative power draw, and
denoted within REMOTE as trickle power.

3.1.5 Parametric sensitivity analysis


As the preceding sections show, modeling excavation tasks involves a large number of parameters
(over 25). A particularly instructive application of REMOTE is in performing sensitivity analyses
that compare the relative impact of variations in these parameters on output metrics. Here it is
not so much the values themselves of the calculated metrics that are paramount, but rather how
sensitive these calculations are to changes in system, concept of operations, and environmental
parameters.
Parameters for sensitivity analysis include system parameters (such as individual robot mass,
payload ratio, wheel radius, etc.) and concept of operations parameters (operational efficiency,
distance to recharge station, etc.) that could be variables in system/mission design. Sensitivity
analysis also includes regolith parameters (bulk density, cohesion, etc.) whose values are estimated within bounds. Each parameter is varied individually from its expected baseline value to
maximum and minimum values in turn. The resulting values of the metrics are calculated at each
variation. Although some parameters are not fully independent in reality, isolating each parameters individual contribution to productivity in this way is still a very useful guide for focusing
attention within such a broad design and operational space.
Sensitivity of production ratio to relevant parameters for an example berm building task is
presented in Figure 3.5 and Figure 3.6. The task involves shallow digging, to a total depth of
20 cm, over a large area (50 m diameter circle). Excavated material is moved to an arc along the
35

circle and dumped in a berm. Average distance between dig and dump is 25 m.
Figure 3.5 shows REMOTE sensitivity analysis results for the berm building task, with LuthWismer as the underlying excavation resistance model. Production ratio (mass of regolith moved
per hour, normalized by rover mass) is shown on the x-axis, while parameters that can affect it
are shown on the y-axis. Changing driving speed, from its baseline value of 20 cm/s to 50 cm/s,
for example, is predicted to increase production ratio from just over 2 to a little under 4. Driving
speed, payload ratio, and operational efficiency are predicted to have the strongest effects on
productivity. Other parameters, such as number of wheels and battery characteristics, have little
effect.
Figure 3.6 shows results for the same sensitivity analysis, but with the Balovnev excavation
model. Results broadly agree between the two models. Driving speed, payload ratio, and operational efficiency govern productivity. The next three most important parameters in the Balovnev
analysis are external friction angle, cohesion, and robot mass. Luth-Wismer also predicst cohesion and robot mass as the next two most important parameters (Luth-Wismer does not include
external friction angle).
These results demonstrate that task-level sensitivity analyses are not particularly dependent
on the choice of underlying excavation model. In both versions of the analysis, productivity
is governed by payload ratio, driving speed, and operational efficiency. These three parameters
figure prominently in the hauling part of excavation tasks, as will be discussed in Section 3.4. Cohesion and robot mass are also important parameters, and have been discussed in prior work [63].
In upcoming sections, additional sensitivity analysis is performed on parameters relevant to
a small robotic excavator, Lysander, and simulated results are compared to experimental data.

3.2

Experiments with a small robotic excavator

To develop effective lightweight robotic excavators, it is important to identify which design parameters have a significant effect on productivity. As described in the previous section, REMOTE
36

37
Figure 3.5: Sensitivity analysis using Luth-Wismer excavation model shows productivity governed by driving speed, payload ratio,
and operational efficiency

38
Figure 3.6: Sensitivity analysis using Balovnev excavation model shows productivity governed by driving speed, payload ratio, and
operational efficiency

Figure 3.7: Lysander is a robotic platform for sitework experimentation - shown here carrying
excavated lunar regolith simulant
simulations and analyses show that payload ratio (ratio of regolith payload mass to robot mass)
and driving speed govern the productivity of small robotic excavators; operational efficiency
also significantly affects productivity. The analysis also shows that other parameters, including
number of wheels, have little effect on productivity.
A prototype excavator, Lysander, enables experimental validation of sensitivity analyses as
well as of the simulator more broadly. Lysander is a low center-of-gravity scraper, and is shown
in Fig. 3.7 transporting excavated lunar regolith simulant.

3.2.1 Experimental setup


Load-haul-dump experiments measure productivity of the lightweight robotic excavator, Lysander,
in controlled conditions. A sandbox was set up with an excavation area, a dump area, and obstacles, as seen in Fig. 3.8. The entire experimental area was on flat ground, with a board in the
dump area to keep dumped soil separate for measurement. The setup represents a general loadhaul-dump task, and the layout provided an efficient way to incorporate all major elements of the
task, including driving approx. 6 m (roundtrip), and making turns to avoid obstacles and align
39

Figure 3.8: Experimental setup for a comprehensive excavation task including digging, dumping,
and shuttling between the two
for dig and dump. The soil used in the experiments is a mixture of general-purpose play sand and
a uniformly fine silica sand. This soil is not a lunar simulant, but its granular nature allows it to
be modeled similarly to regolith. Furthermore, the soil has an internal friction angle between 39
and 42 degrees, and cohesion up to approximately 3 kPa; the values of these strength parameters
lie within the ranges measured for lunar regolith [30]. Internal friction angle and cohesion are
measured using direct shear tests (ASTM D3080). Results of these tests are shown in Fig. 3.9.
The experimental setup fixes some of the parameters studied in the REMOTE simulations.
Some physical robot parameters, such as wheel radius and mass, are fixed. Battery and recharge
parameters are omitted because tethered power enables rapid repetition of experiments. Strength
parameters, i.e. internal friction angle and cohesion, are known within confidence bounds (as
described above). Soil strength parameters are kept within tight bounds with consistent soil
preparation. Between each test run, the soil conditions were reset using a technique developed
40

Shear Stress (kPa)

150

100

50

50

100

150

Normal Stress (kPa)

Figure 3.9: Direct shear test results for soil used in Lysander experiments: Internal friction angle
of 39 to 42 degrees, cohesion of 0 to 3 kPa
at NASA Glenn Research Center. First, the soil is fully loosened by plunging a shovel approximately 30 cm deep and then levering the shovel to fluff the soil to the surface; this is repeated
every 15-20 cm in overlapping rows. Next, the soil is leveled with a sand rake (first with tines,
then the flat back edge). The soil is then compacted by dropping a 10 kg tamper from a height
of approximately 15 cm; each spot of soil is tamped 3 times. Finally, the soil is lightly leveled
again for a smooth flat finish.

3.2.2 Predicted sensitivity of experimental parameters


Parameters that either varied during experiments, or were known only within bounds, are listed
on the y axis in Fig. 3.10. Aside from parameters already discussed (payload ratio, driving speed,
operational efficiency, number of wheels, and soil strength parameters: cohesion and friction
angle), cutting speed, slip, and shear deformation (K) could also potentially vary. While digging,
41

25 %

Payload ratio

50 %

18 cm/s

Driving speed
Operational efficiency

28 cm/s
65 %

80 %

3 kPa

Soil cohesion

10 cm/s

Cutting speed

30 cm/s

Shear deformation

1 cm

2.5 cm

Slip

60 %

90 %

39 deg

Soil friction angle

42 deg
4

Number of wheels
0

10

15

6
27
Production ratio (hr1)

Figure 3.10: Predicted sensitivity of Lysanders productivity to candidate experimental variables [65]. Payload ratio, driving speed, and operational efficiency govern productivity. Parameters that cannot be varied within experiments, such as fixed wheel radius, etc., are excluded.

soil accumulation increased excavation resistance. This higher resistance slowed progress (i.e.
cutting speed) due to increased slip. The ranges of values in cutting speed and slip account for
this variation. Shear deformation is a soil-specific parameter, which could not be measured for
these experiments. A range of possible values from 1 cm to 2.5 cm is considered, based on values
presented in literature for similar soils [33, 76].
Load-haul-dump task productivity is also dependent on driving distance between dig and
dump. This distance was kept the same for all experiments, at 3 m. This is at the low end
of distances that would be required for any long-term task (e.g. mining, trenching). Longer
distances would be expected to further increase the importance of payload ratio and driving
speed, as an even higher percentage of the task would be devoted to driving and transporting
regolith, as opposed to digging or dumping.
42

Figure 3.10 shows REMOTE sensitivity analysis results for parameters relevent to the experimentation campaign using Lysander. Production ratio (mass of regolith moved per hour,
normalized by rover mass) is shown on the x-axis, while parameters that can affect it are shown
on the y-axis. Changing Lysanders payload ratio in simulation, from its baseline value of 25% to
50%, for example, is predicted to increase production ratio from 15 to 27. As in earlier REMOTE
simulations for general lightweight excavators, payload ratio and driving speed are predicted to
have a strong effect on productivity while other parameters, such as number of wheels, are not.

3.2.3 Experimental results


The experimental campaign tested sensitivity of two high sensitivity parameters (payload ratio
and driving speed) and one low sensitivity parameter (number of wheels - on Lysander, the
two middle wheels can be removed with relative ease). Payload ratio was modified by simply
changing the amount of payload carried by the robot; this was implemented in practice by taking
either 1 or 2 cuts of soil to collect a payload ratio of 25% or 50%, respectively. Before taking a
second cut of soil, the soil from the first cut was shifted out of the way to the back of Lysanders
large bucket by tilting the bucket back. The large surface of the bucket and relatively shallow cut
angle, , kept the collected soil from sliding back to the front of the bucket during the second
cut. Clearing the cutting edge of the bucket in this fashion ensured similar excavation during
both cuts.
As operational efficiency was also predicted to be a relatively high sensitivity parameter, it
was monitored during each test. By maintaing operational efficiency within a range of 65%
and 80%, the expected effects of its variability were kept smaller than the expected effects of
varying payload ratio and driving speed, as Fig. 3.10 shows. Production ratio was measured
as the output for each test. Experimental test sets at each parameter setting were performed in
triplicate. Photos from a sample experiment are shown in Fig. 3.11, and results from all the tests
are summarized in Table 3.1.
43

Figure 3.11: Lysander at start location (top) and excavation area (bottom) during an experiment

44

Table 3.1: Experimental data from 4 sets of tests (each set in triplicate)
Test set

Baseline

Low speed

High payload
ratio

4 wheels

Speed
(cm/s)

Payload
ratio

No. of wheels

Operational
efficiency

Production
ratio (hr1 )

29

25%

73%

17.9

28

23%

66%

14.3

30

19%

72%

12.4

18

22%

75%

8.7

18

25%

77%

10.3

18

16%

80%

11.0

29

51%

76%

28.7

28

51%

72%

29.3

28

51%

72%

26.6

26

26%

77%

14.0

24

27%

79%

15.2

37

25%

74%

13.1

High payload ratio and low speed result in statistically significant differences in production
ratio relative to the baseline test set. Applying t-tests to the test sets, high payload ratio and
low speed results in p-values of 0.002 and 0.049, respectively. Both these parameters thus affect
productivity with 95% confidence of statistical significance (meaning there is less than 5% probability of the observed difference in productivity arising by chance, as opposed to there being a
real difference). Comparison of the 4 wheel tests with the baseline (6 wheels), on the other hand,
results in a p-value of 0.679, meaning no statistically significant difference in productivity was
observed.

3.3

Comparison of simulated and experimental results

Figure 3.12 (top) shows experimental results graphically, with error bars at each setting indicating
the standard error. The top bar extends from the mean production ratio value measured during
baseline tests (which had payload ratio at 25%), on the left, to the mean production ratio value
measured during tests with 50% payload ratio, on the right. The error bar around the right edge
45

represents the error in the tests with 50% payload ratio, while the error bar around the left edge
represents the error in the baseline tests. Similarly, the next bar extends from the mean production
ratio value measured during baseline tests (which had a driving speed of 28 cm/s), this time on
the right, to the mean production ratio value measured during tests with 18 cm/s driving speed,
on the left. The error bar around the baseline edge is the same as the bar above, because there
is only one set of baseline tests; these tests act as the baseline for each parameter change. For
the final parameter variation, the plot shows that not only is the mean production ratio achieved
with 4 wheels within the error for production with 6 wheels, but also the mean production ratio
achieved with 6 wheels is within the error for production with 4 wheels. This provides a visual
representation of the statistical results described in the previous section. For changes in payload
ratio and driving speed the error bars do not overlap, highlighting a stastistically significant
difference between these tests and the baseline. For changes in the number of wheels, the error
bars overlap fully and no statistically significant difference is observed.
The bottom of Fig. 3.12 shows simulated results for the same conditions as those tested
experimentally. This plot is a subsampling of Fig. 3.10, showing only payload ratio, driving
speed, and number of wheels.
The sensitivity to payload ratio, driving speed, and number of wheels observed experimentally aligns consistently with the simulated results. For each of the 4 test cases, the simulated
production ratio is within the error of the corresponding experimental case.
As described in previous sections, some modeling simplifications were introduced that do
not correspond exactly with all the details of the excavation tasks. Specifically, slip sinkage
is ignored, as is the time dependency of slip, cut depth (d) and cutting face length (l) during
digging. This makes it impossible to model the observed phenomenon of soil accumulation
(which increases effective d and l), and the subsequent increase in slip (which causes increased
sinkage and thus rolling resistance), using the current implementation of REMOTE.
The good correspondence between simulated and experimental results suggests that this mod46

25 %

Payload ratio

50 %

18 cm/s

Driving speed

28 cm/s
4

Number of wheels
0

10

6
15

25 %

Payload ratio
18 cm/s

Driving speed

27
1
Production ratio (hr )

50 %
28 cm/s

4 6

Number of wheels
0

10

15

27
Production ratio (hr1)

Figure 3.12: Top: Measured production ratio sensitivity to experimental variables. Bottom:
Production ratio sensitivity to experimental variables predicted by simulation. Experiments and
model show good correspondence. In both cases, payload ratio and driving speed have a significant effect on productivity, while number of wheels does not [65].

47

eling simplification did not diminish REMOTEs ability to predict excavator productivity for the
load-haul-dump task described in this work. Figure 3.10 shows that the effects of varying slip
are negligible for this specific task.
The effects of cut depth and slip on prodcutivity are somewhat more significant for trench
excavation, as simulations in prior work show [63]. This suggests that neglecting soil accumulation and slip sinkage may not be appropriate for all tasks, particularly ones involving deep
digging. Before applying REMOTEs modeling framework specifically to a deep excavation
task, additional experiments and possibly additional modeling are recommended.
A significant source of variability in the experimental results was the inability, in practice,
to keep operational efficiency precisely constant. A human tele-operator cannot replicate performance exactly between test runs. Errors were low enough to make clear observations, as
described above. Tighter error bounds, though, could increase the potential statistical power of
experiments. If such tighter bounds were to be required for future investigations, better control
over operational efficiency could be achieved with increased autonomy for the tests.

3.4

Hauling dominates task productivity

Experiments and simulations all show that payload ratio, driving speed, and operational efficiency govern the production ratio for lightweight robotic excavators. These parameters direcly
comprise the terms for hauling productivity:
Production ratio (hauling only) = op P v/d
where op is operational efficiency, P is payload ratio, v is driving speed, and d is hauling
distance. Note that hauling distance is not included as a parameter in above sensitivity analyses,
as it is dictated directly by a given task.
The prominence of the three hauling productivity parameters shows that hauling is the important part of load-haul-dump excavation tasks, in terms of productivity. In fact, hauling takes
up the vast majority of time in any of the tasks investigated, which is why improvements in this
48

area translate into the largest overall gains. An important underlying assumption leading to this
result is that the excavators in question can achieve a nominal level of excavation capability.
If an excavator gets stuck while digging, or must regularly correct and adjust its cuts, the time
spent digging would grow relative to time spent hauling. REMOTE does not model such potential problems, and they were not encountered in the experiments performed. Ensuring that basic
excavation capability is maintained is the subject of the next chapter.

3.5

Conclusions from sensitivy experiments and simulations

Payload ratio, driving speed, and operational efficiency govern productivity of excavation tasks
by small robots. This has been shown using simulation and sensitivity analysis. Experiments using Lysander validate these sensitivity analysis results quantitatively, by varying driving speed,
payload ratio, and number of wheels and measuring output productivity. The task had a relatively short travel distance (3 m) between dig and dump. Longer distances would be expected
to only further increase the importance of payload ratio and driving speed, as an even higher
percentage of the task would be devoted to hauling regolith, as opposed to digging or dumping.
The overarching importance of payload ratio and driving speed is also consistent with simulated
berm building tasks.
The prominence of the three hauling productivity parameters shows that hauling is the important part of load-haul-dump excavation tasks, in terms of productivity. An important underlying
assumption in this result is that the excavators in question can achieve a basic level of excavation
capability. If an excavator gets stuck while digging, or must regularly correct and adjust its cuts,
the time spent digging would grow relative to time spent hauling. Ensuring that basic excavation
capability is maintained is the subject of the next chapter.
Experiments lend credence to simulated results and highlight the reality of some potentially
counter-intuitive findings. It may not be surprising that payload ratio and driving speed should
govern productivity, as making each load-haul-dump cycle faster and carrying more load each
49

time directly speed up the task. The relatively negligible effect of changing the number of wheels
is not necessarily obvious though; one might have thought that added traction could significantly
speed up the digging process and thus overall task productivity. Both experiment and simulation
suggest that optimizing a traction system is not the best place to expend efforts that should rather
be used maximizing speed and payload ratio.
These results suggest that high payload ratio and high driving speed be included as key features of any future lightweight robotic excavators.
REMOTE endevours to strike an appropriate balance between complexity and fidelity for
task-level modeling. Some aspects of excavation tasks are ignored, including wheel slip sinkage
and soil accumulation at the bucket cutting edge. The good correspondence between simulated
and experimental results suggests that these modeling simplifications are acceptable for for high
level analysis of load-haul-dump tasks.

50

4 Thrust and resistance in lightweight


excavation
The previous chapter assumed a nominal level of excavation performance; namely, that excavators could dig without stalling or impeding mobility. For this assumption to hold, excavation
resistance must be kept below excavation thrust. This chapter will show that resistance exceeding
excavation thrust can quickly lead to high slip and sinkage.
This chapter investigates the effects of light weight (both in terms of lower mass and reduced gravity) on excavation thrust and resistance. It demonstrates how soil cohesion has the
detrimental effect of increasing excavation resistance coefficient.

4.1

Relationship of mass and scale

Excavator mass scales approximately as the cube of linear dimension. This is because mass is
related to a machines three-dimensional volume, while length, width, and height are each singledimensional. Fitting data for many terrestrial excavators of many scales [35] to a power law of
the form:
m = aw
one expects that 3. The fit yields exponents of 2.92 and 2.98 for the sample of terrestrial
loaders and scrapers, respectively (with R2 = 0.99 in both cases). In the formulation, m is
excavator mass, w is bucket width (treated as a characteristic length), and a are the power law
51

Figure 4.1: Excavator mass scales with the cube of bucket width, shown by the fact that the ratio
of these two quantities equals approximately 1.

exponent and coefficient, respectively. This approaches the idealized = 3 very closely. This
relationship is also illustrated by directly comparing the ratio of excavator mass to the cube of
bucket width in Figure 4.1. Each of the ratio values is normalized by the median value for easier
comparison. All values are within 15% of the median, with no obvious bias.
In this thesis, any changes in excavator mass are assumed to have corresponding changes in
scale according to the cubic relationship described here. This assumption does not hold for all
possible comparisons, particularly between terrestrial and planetary. Mass optimization may lead
to a planetary excavator with the same scale as a terrestrial one but with lower mass. However, the
more pertinent comparison in this work is instead between two or more planetary excavators. In
this case, all excavators being compared are already mass optimized, so the scaling relationship
is assumed to hold between them.
52

4.2

Light weight reduces excavation thrust coefficient

A vehicles drawbar pull is the net traction available for doing work, and is dependent upon slip
(or travel reduction). Slip is defined as:
Slip(%) = 100% 1

v
r

where v is the vehicles forward velocity, is the wheel angular speed, and r is the effective
wheel radius. In granular soils, effective radius is an equivalent radius where shear occurs between moving soil and static soil [21], which may not be well defined a priori. Compliant wheels
also introduce the possibility of time-varying wheel radius, further confounding measurement of
r and thus slip. For these reasons, an approximately equivalent parameter - travel reduction - is
used in the context of experimental wheel results, instead of slip. Travel reduction is defined as:
Travel reduction(%) = 100% 1

v
v0

where v0 is a baseline vehicle speed on flat ground, with no drawbar load applied. In any
comparisons of travel reduction, the same wheel angular speed is applied throughout.
Drawbar pull at 20% slip (or 20% travel reduction) is a good measure of tractive performance,
as pull begins to plateau around 20% slip for many wheels (or tracks) while negative effects
such as sinkage increase [68]. A non-dimensional quantity, P20 /W (Drawbar pull at 20% slip,
normalized by vehicle weight), has been used as a benchmark metric for lunar wheel performance
from the times of Apollo [24] to today [70, 80]. In this thesis, excavation thrust coefficient is
denoted T; under an assumption of tractive thrust, T = P20 /W .
The ratio P20 /W for any given wheel is approximately constant with changing load (i.e. W
changing but scale and gravity constant), up to a critical loading [24]. It is assumed that excavator
wheels are designed to ensure an excavator with full regolith payload loads the wheels less than
this critical point, so that T remains constant throughout a load-haul-dump excavation task.
Figure 4.2 shows the relationship between normalized drawbar pull (DP/W) and travel reduction for a set of compliant lunar-relevant tires at two different wheel loadings (980 N and 1790 N
per wheel). The value at which the curves cross 20% travel reduction, i.e. the excavation thrust
53

Figure 4.2: It is safest to keep wheel slip at or below 20%. Spring tires, for example, exhibit
very gradually rising travel reduction with increasing load until crossing above 20%, where travel
reduction jumps suddenly [NASA GRC].
coefficient T, is approximately the same in both cases (within 10% of one another). These tires
exhibit strongly nonlinear performance. Travel reduction increases very gradually with increasing load until it crosses above 20%, at which point it rapidly increases to unsafe conditions of
approximately 80% travel reduction. Other wheels experience similar non-linearities at approximately 20% to 30% slip [24]. To prevent unsafe slip and sinkage, drawbar pull should not exceed
T. The tires themselves are shown in Figure 4.3.
Reductions in T are predicted for both reduced gravity as well as reduced scale (and thus,
correspondingly, reduced mass). Figure 4.4 shows such predictions for a rigid wheel. Drawbar
pull vs. slip is calculated according to Bekkers equations, using the same procedure and assumptions described in Section 3.1.1. The baseline wheel (plotted in black) has a radius of 30 cm and
width of 15 cm, and is assumed to be for a 200 kg rover; the baseline condition is Earth gravity.
The same wheel and rover are used in the reduced gravity case (blue). The reduced mass case
(red) assumes a 33 kg rover, with wheel radius and width scaled to 17 cm and 8 cm, respectively,
according to the relationship presented in Section 4.1. In all cases, the internal friction angle is
54

Figure 4.3: Compliant, lunar-relevant spring tires on the Scarab robot [NASA GRC]
45 degrees. Results are plotted for cohesionless soil as well as soil with c = 3.8 kPa (at the high
end of estimates for lunar regolith).
Figure 4.4 shows that reduced gravity and reduced mass & scale are both predicted to decrease T (though the effect of gravity on T is somewhat mitigated in highly cohesive soil). In
other words, mobility performance decreases by more than just the decrease in weight when
either gravity or mass is reduced.
Kobayashis experiments with a rigid wheel in reduced-gravity flights also demonstrate that
lower gravity diminishes relative mobility performance [43]. Higher slip was observed in lower
gravity for otherwise identical test conditions.

4.3

Predicted effects of light weight on excavation resistance


coefficient

An excavators requirements for drawbar pull are driven by the excavation resistance it experiences while cutting soil. Chapter 2 showed that excavation resistance can be modeled in the
form:
55

100

90

90

80

80

70

70

60

60

slip (%)

slip (%)

100

50

50

40

40

30

30

20

20

10

10

0.2

0.4

0.6

0.8

DP/W

0.2

0.4

0.6

0.8

DP/W

Figure 4.4: Reducing gravity or wheel scale decreases the excavation thrust coefficient, T. Predicted T in cohesionless soil (left) and soil with c = 3.8 kPa (right), for a baseline rigid wheel
(black), the same wheel in 1/6 gravity (blue), and a 1/6 mass scaled wheel (red). The circles on
the slip-DP/W curves indicate T (on the DP/W axis) for each case.
Fex = N gwd2 + Nc cwd
where gravitational acceleration is denoted g, is soil density, d is cut depth, w is cut width,
and c is cohesion. The Ni are non-dimensional coefficients pertaining to each of the sources of
force. Soil-tool adhesion and inertial forces are ignored due to their insignificant contributions.
Surcharge is treated separately in Section 4.3.1.
Of interest is how Fex and F (excavation resistance coefficient, Fex /W ) vary in light weight
operation. If we first consider cohesionless soil (c = 0), the simple excavation resistance model
above predicts constant excavation resistance coefficient. F simplifies to:

N gwd2

F =
mg
N wd2
=
m
In this case F is no longer dependent on g, so it remains unchanged by a reduction in gravity.
Section 4.1 showed that mass varies as the cube of scale for excavators; reducing mass to a
fraction of the original (X) is equivalent to shortening linear dimensions to the cube root of that

fraction ( 3 X). Mass reduction thus also results in an unchanged F for cohesionless soil:
56

N gwd2
F =
mg
N wd2
=
m

N ( 3 Xw0 )( 3 Xd0 )2
=
Xm0

3
N ( X)3 w0 (d0 )2
=
Xm0
N w0 (d0 )2
=
m0
= F0

where the subscript 0 denotes original (pre-reduction) parameter values.


On the other hand, in a purely cohesive soil (N = 0) F is sensitive to changes in weight. In
this case Fex is independent of g:
Fex = Nc cwd
In reduced gravity, Fex remains unchanged while weight reduces proportionally. In other
words, a reduction in gravity to 1/6 results in a six-fold increase in F in purely cohesive soil.
Reducing mass has a slightly more complex effect on F due to the relationship with length
scaling:

Nc cwd
F =
mg

Nc c( 3 Xw0 )( 3 Xd0 )
=
Xm0 g
Nc cw0 d0
=
3
Xm0 g
F0
=
3
X
57

Figure 4.5: Reducing gravity to 1/6 or mass to 1/6 does not decrease excavation force to 1/6 if
the soil is cohesive. Effects of reduced weight on Fex and F depend on the proportion of original
excavation force from cohesion.
Soils with both frictional and cohesive properties exhibit a mix of the weight effects described
above, depending on the proportion of Fex due to cohesion. Figure 4.5 shows the effects of
reducing gravity to 1/6 or mass to 1/6, with dependence on cohesion. Effects on Fex and F
are both plotted, on the left and right y-axis, respectively, as each of these related metrics is
convenient to discuss in certain contexts. What is apparent from the figure is that reducing
weight (either by reducing gravity, mass, or both) increases F in any soil with cohesion.
Another important observation is that reducing an excavators scale magnifies the effects of
cohesion. Because the gravity term in Fex includes a product of 3 lengths (wd2 ) while the
cohesion term just 2 (wd), reductions in scale lower the gravity contribution more than they
lower the cohesion contribution. A related analysis with similar conclusions was put forward
by Zacny et al [81].
This analysis also provides a potential explanation for results observed by Boles et al [13].
When they compared excavation resistance forces measured in Earth gravity to resistance forces
measured during 1/6 gravity parabolic flights (for otherwise identical experiments), they were
58

puzzled that excavation forces reduced to an average of 1/3 the original force, rather than 1/6.
Figure 4.5 suggests that this simply implies that cohesion contributed approximately 20% of the
resistance force in the Earth gravity experiments. As the soil used in these tests was JSC-1 lunar
simulant, with a documented cohesion as high as 14.4 kPa [41], this is not a surprising result.

4.3.1 Effects of light weight operation on surcharge

Agui shows that longitudinal excavation resistance rises approximately linearly with cut distance,
as soil accumulates in a discrete bucket [2]. This agrees with the general modeling assumptions
of Shmulevich [61], as discussed in Sections 2.1.3 and 2.2.1. Experiments conducted as part of
this research also show approximately linearly increase in F with payload accumulation, as will
be shown in Section 4.4.4.
It is pertinent here to consider how the slope of this linear increase in F with increasing
payload is affected by reductions in gravity or mass. This slope is denoted dF /dP . Shmulevich
and Kobayashi both assume surcharge is caused only by the weight of the accumulating soil (see
Section 2.1.3). Plugging Shmulevichs surcharge model (Equation 2.3) into Reeces fundamental
earthmoving equation (2.1 and 2.2) results in a surcharge contribution of the form:

Fex,q = N gwdx

(4.1)

where x is the cut distance and N is a dimensionless coefficient. Payload also increases
with cut distance according to the density and volume of cut soil (gwdx), leading to a dF /dP
independent of g or length scaling:
59

dF /dt
dF
=
dP
dP /dt
Wrobot 1 dFex /dt
=
Wrobot 1 gwdx
dFex,q /dt
=
gwdx
N gwdx
=
gwdx
= N

However, this conclusion hinges on the form of Fex,q proposed by Shmulevich, which assumes that surcharge force depends only on the weight of accumulated soil, and not its cohesion.
Meanwhile, there are in fact reasons to believe cohesion plays a part in surcharge force.
Qinsen [57] demonstrates a precedent for considering cohesive surcharge. They model excavation for a bulldozer blade, accounting for soil accumulation directly. The model considers
not only the weight of accumulating surcharge, but also its cohesion. The models other aspects
relating it specifically to a curved bulldozer blade make it more complex than is expedient for
direct analysis here.
Surcharge models such as Shmulevichs and Kobayashis assume all surcharge force is due
to pressure applied (by accumulating soil) from above original ground level. Soil shearing below
ground level, accounted for by the gravity (frictional) and cohesive shear, is assumed to be unaffected by soil accumulation above. In reality, though, the two are not independent. Figure 4.6,
with photos by Shmulevich [61], shows how the shear plane failure angle (which can be measured from the photos) changes as soil accumulates. Early in the cut (displacement 5 mm, left),
the failure angle is just under 30 degrees; later (displacement 70 mm, center) it is approximately
25 degrees; finally (displacement 140 mm, right) it is less than 20 degrees. As cut depth is kept
constant, this shallower angle results in a final shear plane over 50% larger than at the start of the
cut. As shearing along the failure plane is both frictional and cohesive, it stands to reason that
60

Figure 4.6: Soil accumulation not only adds pressure from the weight of the soil, but also increases the size of the shear failure plane below the surface, as measured in a photo by Shmulevich [61]

Reduced gravity
Reduced mass / scale

Cohesionless (c = 0)
Constant
Constant

Cohesive (c 6= 0)
Large increase
Small increase

Table 4.1: Predicted effects of light weight on both F and dF /dP , depending on soil properties
cohesive forces thus arise at least indirectly with increasing surcharge.
Analyses analagous to those conducted in the previous section for cohesive soils suggest that
cohesive contributions to surcharge would lead to the slope dF /dP increasing with decreasing
gravity and/or mass. Testing this cohesive surcharge hypothesis experimentally is suggested as
future work.

4.4

Excavation scaling experiments

Table 4.1 summarizes the predicted effects of light weight on both F and dF /dP , based on
analyses presented above. The top-right quadrant describes conditions during experiments by
Boles et al [13]. Although they were not explicitly testing the hypothesis presented here, their
results qualitatively agree. Experiments in conditions shown in the bottom-left quadrant are the
subject of this section. Experiments covering the remaining two quadrants are recommended as
future work.
61

Figure 4.7: Bucket-wheel dimensions: wheel diameter D (A), blade width w (B), and blade
tangent extension e (C)

4.4.1 Experimental setup


Experiments with bucket-wheels (BWs) and flat-plates (FPs) of various scales were conducted
in cohesionless soil, GRC-1 [56], studying the effects of scaling on continuous and discrete excavation, respectively. The bucket-wheel experiments were conducted jointly with Diaz Lankenau [19].
To study the effects of length scaling (and thus the related mass scaling), 4 sizes each of
bucket-wheels and flat-plates were tested. All aspects of the test implements with a length dimension were scaled proportionally. For BWs this includes: wheel diameter D (A), blade width
w (B), blade tangent extension e (C), cut depth, and advance speed. Angular speed, number of
blades, and blade thickness were kept constant. A, B, and C are shown in Figure 4.7. For FP
tests, scaled parameters include: plate width, plate length l, and advance speed (blade thickness
again kept constant). Dimensions for bucket-wheels and flat-plates are summarized in Table 4.2.
Note that production rate, which can be approximated by wdv (where is soil density, w
is cut width, d is cut depth, and v is advance speed), is consistent between bucket-wheels and
flat-plates of the same scale. In fact, the scaling ensures that production ratio (production rate
normalized by a representative excavator mass) is constant between all tests.
62

Scale
1.0
1.3
1.7
2.3

v
0.07 cm/s
0.10 cm/s
0.13 cm/s
0.17 cm/s

BW D
27.6 cm
36.8 cm
48 cm
63 cm

BW w
5.7 cm
7.4 cm
9.8 cm
13 cm

BW d
BW e
2.0 cm 4.6 cm
2.7 cm 6.2 cm
3.5 cm 8.2 cm
4.7 cm 10.9 cm

FP w
11.4 cm
14.8 cm
19.6 cm
26 cm

FP d
FP l
1.0 cm 5.8 cm
1.3 cm 7.5 cm
1.7 cm 9.8 cm
2.3 cm 13.2 cm

Wrobot
54 N
124 N
283 N
645 N

Table 4.2: Scaled dimensions of bucket-wheel (BW) and flat-plate (FP) test implements

Figure 4.8: The four scales of bucket-wheel tested


All bucket-wheels share the same dodecagon-shaped hub designed to hold one L-shaped
blade extending out on each of its 12 sides, as shown in Figure 4.8. Each blade is cut and bent
from 1.6mm (1/16 in.) thick aluminum. Flat plates are cut from the same thickness aluminum,
and are bent to a cutting angle of 10 degrees below horizontal.
The scaling of advance speed and bucket-wheel diameter, with angular rate kept constant,
results in a tangential bucket-wheel cutting speed 5 times the advance speed during each test.
Cutting speeds were low enough that no dynamic effects need be considered. These values never
exceeded 1 cm/s (measured as the sum of advance speed plus tangential speed).

4.4.2 Preliminary investigation of soil preparation


Soil density is an important parameter of lunar regolith [15]. It also influences tool-soil interaction and must be controlled for repeatable experimental results [3]. There is direct interdependence of density with state of compaction and stress history [31], both of which can be modified
63

during soil preparation. Resistive forces encountered during penetration and cutting of soil are
of similar nature [82]. It is common to use a penetrating device to estimate cutting resistance in
the field; soil parameters can be determined from such experiments if tools are selected appropriately [3].
Penetration tests were done on GRC-1 prepared to different densities to determine the necessary level of compaction. GRC-1 that is too loose may have unpredictable behavior due to the
random localization of large voids in the soil. The test consisted of slowly increasing the vertical
load on a 3.8 cm diameter cylinder penetrating the soil surface until a depth of 7.5 cm is reached.
Throughout the movement of the cylinder load vs. sinkage data was collected and analyzed.
GRC-1 is expected to have a soil constant n of approximately 1.2 [56], which means that penetration pressure should increase almost linearly with depth. Linear behavior was observed during
experimentation when compaction was done by ten tamps or more with a 20 x 20 cm 6.7 kg steel
plate.
Not only are the pressure-sinkage results for the compacted (with 10 tamps) soil more linear,
they are also more consistent. Linear regression on each test with ten tamps gives an R2 value
of at least 0.987. The average soil stiffness constant is 2778 kPa/m with a standard deviation of
157 kPa/m; for loose soil those same values are 1625 kPa/m and 418 kPa/m respectively. For
these slopes an n soil constant of 1 was used. Plots for five penetration tests each in loose soil
(Figure 4.9) and compacted soil (Figure 4.10) are shown. A vertical offset between experiments
is due to preload sinkage and is not the main concern in testing. Obtaining similar slopes between
experimental runs was the main objective.

4.4.3 Soil preparation and force measurement


All tests were done on GRC-1 lunar regolith simulant contained in a 115 x 68 x 27 cm soil bin.
A consistent soil preparation procedure was followed before each test. First, soil was loosened using a gardening spade. The loose soil surface was made flat by dragging a straight-edged
64

Figure 4.9: Penetration curves for loosened soil showing wide variability. Each line is one
experimental run.

Figure 4.10: Penetration curves for compacted soil showing consistent linear response. Each line
is one experimental run.

65

tool across it. To check if the soil surface was horizontal a digital level was placed length-wise
on it. To be accepted as flat horizontal soil the measured angle must be less than 0.5 and the level
flush with the soil surface all through its 61 cm length. The soil was modified as necessary until
it lay in a satisfactory way, by leveling with the straight-edged tool. The entire surface of the soil
was compacted by tamping with a 20 x 20 6.7 kg steel plate with a handle attached to it. Each
tamp is achieved by dropping the steel plate from a 5-6 cm height; subsequent tamps overlap
each previous one with approximately 30%.
The steel plate is roughly one third of the width of the soil bin so three side-by-side lanes
were tamped, for a total width of 60 cm. By executing the compaction procedure along these
lanes ten times it was guaranteed each part of the soil surface would be compacted by at least ten
tamps.
Each bucket-wheel and flat-plate test implement was tested at least 3 times, measuring the
force and torque generated during excavation. Test implements were advanced for a distance of
25 cm, using an actuated axis on the soil bin. Before starting each test the test implement was
raised above the prepared soil and the load cell was biased. Load cell data collected commenced
once test implements were set to the correct depth.
Bucket-wheels advanced perpendicularly to the axis of rotation at a fixed depth and speed
through the center of the soil bin. Reaction loads were measured using a 6-DOF Force/Torque
load cell to which the BW motor was mounted. The BW was then set to rotate and gradually
lowered into the soil until the desired depth had been reached. To measure depth a ruler was
fixed to the arm holding the BW (as seen in Figure 4.11, left) and vertical distance from the BW
axle to a 3 mm thick clear plastic plate resting on the soil was measured. Once the BW was set to
the right depth, the BW set to rotate and advance. As the BWs do not have side-walls to contain
the soil they excavate a shop vacuum cleaner was used to remove soil from each blade just as it
fully cleared the soil surface.
Flat plates were also advanced through the center of the soil bin (Figure 4.11, right). Reaction
66

Figure 4.11: Experimental setup for measuring excavation resistance forces of bucket-wheels
(left) and flat-plates (right)
loads were again measured using a 6-DOF Force/Torque load cell to which the angled plate was
mounted. The plate was lowered in a space where soil was cleared away to the required depth.
Depth was measured similarly to the procedure for BWs. Once the FP was set to the right depth,
it was advanced through the soil.

4.4.4 Experimental results


Theory developed above suggests F and dF /dP stay constant when excavation implements are
scaled in cohesionless soil. Bucket-wheel and flat plate excavation resistance was measured at
4 separate scales, with these scales defined in Table 4.2. Each BW and FP was tested at least
3 times, measuring the force and torque generated during excavation. Non-dimensionalized test
data enables the comparison of F and dF /dP at each scale. To non-dimensionalize, Fex test data
was normalized by an assumed Wrobot for each scale, as listed in Table 4.2. The absolute values
67

of these weights are somewhat arbitrary, but are estimated to be reasonable for the given lengths;
the most important thing to note is that the weights are scaled according to the cube of length
dimensions, as described in Section 4.1. The time scale is also non-dimensionalized in the test
data to correspond to a payload ratio accumulation (again normalized by the assumed Wrobot for
each scale).
Figure 4.12 shows experimental results from bucket-wheel and flat-plate tests at each of the
4 scales. Raw data from multiple runs at the same scale are overlaid with a linear regression of
combined test data. The data shows how bucket-wheel excavation resistance is bounded, while
it rises approximately linearly with accumulating payload for flat-plate excavation. Variability
around this linear rise also tends to increase as a flat-plate cut proceeds. Figure 4.13 shows the
same linear regressions all plotted on the same non-dimensionalized axes. Qualitatively, it shows
no obvious trend in F or dF /dP with changes in scale for either bucket-wheels or flat-plates.
Excavation resistance force scaling is also analyzed quantitatively by fitting raw (unnormalized) force data to a power law vs. scale:
Fex = aS
where S is scale and a and are the power law coefficient and exponent, respectively. For
constant F and dF /dP , one expects 3, as Fex needs to scale cubically to keep pace with
cubically increasing weight.
For bucket-wheels, mean force from each test is used for the fit. Figure 4.14 shows the mean
force data and the power law modeled to fit them. The best fit predicts = 2.73, with a 95%
confidence interval for of [2.5, 3.0]. For flat-plates, two seperate fits are calculated for the
excavation resistance force at the start of a cut, denoted F0 and shown in Figure 4.15, and for
the rate of force increase, dF/dt (Figure 4.16). The best fit for F0 is = 2.36 (95% confidence
[1.9, 2.9]), and for dF/dt it is 3.05 (95% confidence [2.7, 3.4]). The fits for bucket-wheel F and
flat-plate dF/dt are close to 3. For flat-plate F0 , is somewhat low and has a wide confidence
interval; this may be due to experimental procedure. As described in Section 4.4.3, soil was
68

0.08

0.08

0.06

0.06

0.04

0.04

0.02

0.02

0
0

Fex/W

Fex/W

0.1

0.02

0.04
Payload ratio

0
0

0.06

0.1

0.1

0.08

0.08

0.06

0.06

Fex/W

Fex/W

0.1

0.04

0.04

0.02

0.02

0
0

0.02

0.04
Payload ratio

0
0

0.06

0.02

0.04
Payload ratio

0.06

0.02

0.04
Payload ratio

0.06

Figure 4.12: Discrete flat-plate excavation resistance (blue) increases without bound, while continuous bucket-wheel excavation resistance (red) remains low and bounded. Plots show excavation resistance coefficient, F , vs. payload ratio accumulated, P , in cohesionless soil at 4 different
scales: 1.0 (top-left), 1.3 (top-right), 1.7 (bottom-left), and 2.3 (bottom-right). Raw data and linear regression shown.

69

0.1

Fex/W

0.08

0.06

0.04

0.02

0
0

0.02

0.04
Payload ratio

0.06

Figure 4.13: Discrete flat-plate excavation resistance (blue) increases without bound, while continuous bucket-wheel excavation resistance (red) remains low and bounded. The trends in excavation resistance coefficient, F , vs. payload ratio accumulated, P , are unaffected by scaling
(ranging from 1.0 up to 2.3) in cohesionless soil. Linear regressions shown.

70

12
10

8
6

2
0

1.2

1.4

1.6
1.8
Length scale

2.2

Figure 4.14: Scaling of mean bucket-wheel excavation resistance force. Best fit power law
exponent = 2.73 (compared to a predicted value of approximately 3)
cleared away to enable lowering the flat-plate to the required depth. Variability in this clearing
could lead to the plate first engaging soil at slightly different times between tests, and this shift
in data along the time axis introduces error into measurement of the intercept, F0 .

4.5

Conclusions regarding thrust and resistance for lightweight


resistance

Excavation is more difficult on planetary surfaces, especially on the Moon, than it is on Earth.
Lightweight operation with small robots and/or in reduced gravity disproportionately reduces
excavation thrust while also increasing excavation resistance disproportionately in cohesive lunar
regolith. For these reasons, excavation in Earth gravity (even with a robot of relevant scale in a
regolith simulant) overpredicts the performance of excavators in reduced gravity.
The ratio P20 /W is an appropriate metric to use for excavation thrust coefficient, T, when
assuming thrust is generated through traction. When slip goes above 20%, the mobility of most
wheels can degrade rapidly; maintaining drawbar pull low enough to keep slip below 20% satis71

9
8
7

F0

6
5
4
3
2
1
0

1.2

1.4

1.6
1.8
Length scale

2.2

Figure 4.15: Scaling of flat-plate excavation resistance force intercept. Best fit power law exponent = 2.36

0.014
0.012

dF/dt

0.01
0.008
0.006
0.004
0.002
0

1.2

1.4

1.6
1.8
Length scale

2.2

Figure 4.16: Scaling of the slope of flat-plate excavation resistance force. Best fit power law
exponent = 3.05

72

fies the assumption that an excavator maintains nominal capability during digging. T = P20 /W
is approximately constant with changing load, for a given wheel with scale and gravity kept constant. On the other hand, T is predicted to decrease for reduced gravity and for reduced mass &
scale. Kobayashis experiments support this prediction for reduced gravity [43].
The excavation thrust coefficient, F , is bounded in continuous excavation, and increases
approximately linearly with payload accumulation in discrete excavation. F and dF /dP for
cutting cohesive soil are predicted to increase for reduced gravity and for reduced mass & scale.
Boles experiments support this prediction for F in reduced gravity [13]. In cohesionless soil, F
and dF /dP are predicted to remain constant; this is supported by experimentation herein.
Further study of mobility and excavation in reduced gravity flights is recommended. Additional scaling experiments studying wheel traction as well and excavation in cohesive soils are
also suggested for future work.

73

74

5 The lightweight threshold


All excavators have a lightweight threshold in the operating space, below which their productivity is limited by virtue of lacking the weight to produce enough thrust to overcome excavation
resistance. This threshold is crossed at lower weights for continuous excavators than discrete
excavators. The lightweight threshold is described by a non-dimensional quantity that relates
payload ratio, excavation resistance, and excavation thrust.
The lightweight threshold arises from the fact that mobility can quickly degrade if excavation
resistance surpasses excavation thrust, as shown in the previous chapter. An excavator that has
crossed this threshold is operating in the lightweight regime.

5.1

A non-dimensional Lightweight number

An excavator is operating in the lightweight regime if excavation thrust is less than excavation
resistance at any point during digging; i.e., an excavator that produces thrust through traction is
in the lightweight regime if at any point:

P20 < Fex

(5.1)

Both thrust and resistance depend on the amount of payload collected. The ratio P20 /W
can be assumed to be constant, as discussed in Section 4.2, and equates to the excavation thrust
coefficient, T, so:
75

P20 = TW
Note that W is total weight:

W = Wrobot + Wpayload
Inequality (5.1) thus becomes:

T(Wrobot + Wpayload ) < Fex


Normalizing both sides by empty robot weight, Wrobot , leads to:

T(1 + P ) < F
Defining a lightweight number, L, distinguishes the lightweight operating regime when:

T(1 + P )
<1
F

(5.2)

The dependence of F on P has been discussed in Chapters 2 and 4, and can be represented
by a linear approximation. Figure 5.1 shows the straight-line approximations of F (P ) for both
continuous and discrete excavation. Agui et al. also found that, equivalently, Fex (x) can be
approximated by straight lines for discrete excavation [2]).
Denoting this linear approximation by:

F = F0 + F P
with F shorthand for dF /dP . Factoring out F0 and substituting into (5.2) leads to:
76

Figure 5.1: F 0.9 for discrete excavation, which is substantially higher than both F0,disc
and F0,cont ; as a result, continuous excavators achieve higher lightweight numbers, L, and are
thus more suitable for lightweight operation. Plot shows excavation resistance coefficient, F ,
vs. payload ratio accumulated, P , for continuous (red) and discrete (blue) excavation at equal
production rate. Raw data and linear regression shown.

L=

T(1 + P )

F0 (1 + F P )

(5.3)

F0

5.1.1 L for continuous and discrete excavation


The accumulation of payload has opposite effects on L for continuous and discrete excavators.
The most challenging operating point for continuous excavators is at the start of a cut, after which
digging gets easier. On the other hand, for discrete excavators digging becomes more difficult as
cutting proceeds.

For continuous excavation, F is bounded so a linear approximation yields Fcont


= 0. Substi-

tuting into (5.3) gives:

Lcont =

T(1 + P )
F0,cont

In this case, adding payload (i.e. increasing P ) increases L. Additional payload pushes
continuous excavators away from the lightweight regime.
77

min(Lcont ) =

T
F0,cont

(5.4)

For discrete excavation, L decreases with increasing P as long as F /F0 > 1. In reality,

Fdisc
is greater than F0,disc by between 1 and 2 orders of magnitude (again, see Figure 5.1 and

Agui [2]). This means that additional payload pushes excavators towards the lightweight threshold. The theoretical lower bound (as P ) for discrete L approaches:

min(Ldisc ) =

T
F

Even with an assumption of P = 0.5 (which is more realistic, as discussed in Section 1.4.2):

min(Ldisc ) <

3T
F

(5.5)

Finally, another observation that can be drawn from Figure 5.1 is that Fdisc
is orders of mag-

nitude greater than F0,cont , at equivalent production ratios. Meanwhile, a factor of just 3 ensures a higher lightweight number, L, for the continuous configuration (for the practical case of
P = 0.5). This higher margin on L enables continuous excavator weight to be reduced further
than for discrete excavators, before crossing the lightweight threshold.

The lower lightweight threshold for continuous excavators hinges on Fdisc


being larger than

F0,cont . In the experiments discussed in Section 4.4, as well as in those discussed below in

Section 5.2, Fdisc


is orders of magnitude larger than F0,cont . Although these two distinct examples

do not prove the result generally, they strongly suggest that it likely holds in all but the most
degenerate of cases.
78

5.2

Gravity offloaded excavation experiments

Novel experimentats are described here that for the first time subject excavators to gravity offload (a cable pulls up on the robot with 5/6 its weight, to simulate lunar gravity) while they
dig. Although not fully representative of excavation on planetary surfaces (where the regolith is
also subject to reduced gravity), these experiments are better tests of planetary excavation performance than testing in Earth gravity. The experiments demonstrate the disproportionate effects
of reduced gravity on discrete excavation, compared to continuous excavation, predicted in the
preceding section.
Testing in Earth gravity is an inadequate evaluation of planetary excavators, as it overpredicts
excavation performance relative to reduced gravity. Operation in reduced gravity reduces excavation thrust coefficient while also increasing excavation resistance coefficient in cohesive lunar
regolith. The most representative test environment is a reduced gravity flight, where excavator
and regolith are both subject to reduced g [13, 43]. Future research on lightweight excavation
would benefit from testing in reduced gravity flights. Opportunities for such tests are infrequent,
though, and their scale (both spatially and temporally) is severely constrained by the logistics of
the flights.
Another class of tests reduces the weight of the robot, but not the regolith. NASA JPL runs
mobility tests for the Curiosity rover using a full geometric scale 3/8th mass SCARECROW
rover [71]. SCARECROW is comprised of the chassis and mobility subsystems and preserves
center of gravity location. SCARECROWs 3/8th mass loads the wheels with an equivalent
weight to the full mass Curiosity rover in Mars gravity. For equivalent testing for lunar conditions, a full geometric scale 1/6th mass rover would be required; this is very little mass even for
only the suspension and mobility subsystems. Another way to achieve equivalent results is to
use a full mass robot, but to offload gravity (see Section 2.2.2); this is the approach used in this
work.
Testing with reduced robot weight in Earth gravity may not exhibit the same mobility per79

formance as planetary driving, where the regolith is also subject to reduced gravity [79]. It
may in fact overpredict traction for scenarios governed by the ratio P20 /W , such as pulling and
slope climbing. Terramechanics models and experiments both suggest that P20 /W is approximately constant with changing load (i.e. changing W but keep scale and gravity constant, as
with SCARECROW or gravity offload). Meahwhile, they suggest that changing W by reducing
gravity reduces P20 /W , as discussed in Section 4.2.
On the other hand, reducing robot weight but not regolith weight makes excavation more difficult than is to be expected in reduced gravity. Longitudinal soil-tool interactions are not directly
affected by reduced robot weight, so Fex remains constant. Reducing weight to 1/6 thus directly
inceases Fex /W sixfold. For planetary excavation, this corresponds to the worst possible case
of purely cohesive regolith. As neither lunar nor Martian regolith is purely cohesive, excavation
resistance on these planetary surfaces in not expected to scale so poorly.
Excavating with gravity offload thus underestimates the detrimental effects of gravity on
traction, but overestimates the detrimental effects on excavation resistance. Though not ideal,
this is a more balanced test than excavating in Earth gravity, which underestimates detrimental
effects on both traction and resistance.

5.2.1 Experimental setup


To explore the differences in lightweight thresholds for continuous and discrete excavation, gravity offloaded excavation experiments were conducted at NASA Glenn Research Centers (GRC)
Simulated Lunar OPErations (SLOPE) lab. The facility contains a large soil bin with GRC1 [56] lunar simulant. Continuous bucket-wheel and discrete bucket excavation was performed
using the Scarab robot (for a detailed description of the robot, see [5, 70]). With Scarabs shell
removed, excavation tools were mounted to the robots structural chassis.
For continuous excavation, a bucket-wheel was mounted with its axis of rotation aligned with
Scarabs driving direction. The bucket wheel is 80 cm diameter with 12 buckets, and each bucket
80

Figure 5.2: Gravity offload testing with bucket-wheel (left) and front-loader bucket (right) on the
Scarab robot. A cable pulls up on the robot, tensioned by weights acting through a 2:1 lever arm.
The weights and lever assembly hang from a hoist that is pulled along a passive rail by a separate
winch-driven cable.
has a width of 15 cm The bucket used for discrete excavation is 66 cm wide, and was mounted
behind Scarabs front wheels at a cutting angle, , of 15 degrees from horizontal (see Fig. 3.3 for
a definition of ).
This research developed an experimental apparatus for achieving gravity offload in the SLOPE
lab. The main aspects of the apparatus are shown in Figure 5.2. A cable pulls up on the robot,
tensioned by weights acting through a 2:1 lever arm. The weights and lever assembly hang from
a hoist that is pulled along a passive rail by a separate winch-driven cable. All tests are conducted
in a straight line below the hoist rail. The winch speed is controlled so that the hoist is pulled
along at the same speed as the robot is driving, keeping the cable vertical. For tests where excavator speed remains constant, winch speed is set open loop. For tests where the excavator enters
into high slip, winch speed has to be manually reduced to match the robots decreasing speed.
Scarab has a mass of 312 kg (weight of 3060 N in Earth gravity) in the configuration used for
these experiments. The connection point for the gravity offload cable was adjusted to preserve the
robots weight distribution (54% on the rear wheels). This was confirmed by weighing Scarab on
4 scales (one under each wheel) before and after being connected to the gravity offload apparatus.
81

The offloading cable was equipped with a 2-axis inclinometer and a single-axis load cell to
measure cable angle and tension, respectively.
Between each test run, soil conditions were reset using a technique developed at NASA
GRC. First, the GRC-1 simulant is fully loosened by plunging a shovel approximately 30 cm
deep and then levering the shovel to fluff the regolith to the surface; this is repeated every 1520 cm in overlapping rows. Next, the regolith is leveled with a sand rake (first with tines, then
the flat back edge). The regolith is then compacted by dropping a 10 kg tamper from a height of
approximately 15 cm; each spot of soil is tamped 3 times. Finally, the regolith is lightly leveled
again for a smooth flat finish. A cone penetrometer was used to verify that the soil preparation
consistently achieved bulk density between 1700 kg/m3 and 1740 kg/m3 .
Continuous and discrete excavation experiments were conducted at equivalent nominal production rates of approximatly 0.5 kg/s, and at equal speeds of 2.7 cm/s. To account for the
differing geometry of the excavation tools, the rectangular discrete bucket cut at a depth of 2 cm,
and the bucket-wheel cut at a depth of 5 cm. Depth was set using Scarabs active suspension,
which raises and lowers the central chassis. Regolith picked up by the bucket-wheel was collected into 5-gallon buckets, as shown in Figure 5.3, and weighed. The discrete bucket collected
regolith directly, and after a test that regolith was transfered into 5-gallon buckets and weighed.
The lightweight threshold is characterized by a degradation of mobility. To capture mobility
information, the excavators position was tracked at a data rate of 1 Hz using a laser total station.

5.2.2 Predicted lightweight numbers


Estimates of lightweight numbers, L, for the various test conditions predict that Scarab will
not enter the lightweight regime for either mode of excavation at 1 g, but that it will cross the
lightweight threshold when performing discrete excavation under gravity offload.
A maximum payload ratio of P = 0.5 is assumed for the excavation configurations of interest. In this case, Equation 5.5 is applicable for discrete excavation, and Equation 5.4 applies
82

Figure 5.3: Regolith collection during an offloaded bucket-wheel test

to continuous excavation (independent of the choice of P ). Excavation thrust, T is the same for
both modes of excavation, as the excavator uses the same spring tires for all the experiments.

For these tires, T 0.25, as seen in Figure 4.2. The value of Fdisc
, between 0.2 and 0.3, is

estimated from experiments with a similar discrete bucket [2]. Force measurements from preliminary tests with the bucket-wheel suggest a F0,cont between 0.002 and 0.004. These values
lead to estimates of L between 60 and 130 for continuous excavation, and between 2 and 4 for
discrete excavation, in 1 g.

Gravity offload reduces drawbar pull proportionally to the reduction in weight, without changing excavation resistance. This effectively reduces L in step with the weight reduction, leading
to L values between 10 and 20 for continuous excavation, and between 0.3 and 0.7 for discrete
excavation when operating with the excavator weight offloaded to 1/6. An L below 1 for the discrete excavation case represents crossing the lightweight threshold and thus an inability to safely
collect a payload ratio of 0.5. Table 5.1 summarizes these predictions.
83

Continuous excavation
Discrete excavation

1g
60-130
2-4

Offloaded to 1/6 g
10-20
0.3-0.7

Table 5.1: Predicted values for L for the relevant experimental conditions. The excavator is
predicted to cross the lightweight threshold (L < 1) when performing discrete excavation under
gravity offload.
Excavation type
Driving only
Continuous
Discrete
Driving only
Continuous
Discrete

Gravity
1g
1g
1g
1/6g
1/6g
1/6g

Average v
2.6 cm/s
2.6 cm/s
2.6 cm/s
2.7 cm/s
2.7 cm/s
no S/S

v
0.2 cm/s
0.3 cm/s
0.4 cm/s
0.3 cm/s
0.3 cm/s
n/a

Table 5.2: Discrete excavation offloaded to 1/6 g is the only test condition that does not maintain
constant steady state (S/S) velocity throughout. Note that v represents the mean of the 3 tests
values, not the of the 3 tests mean velocities (which showed negligible variation between
tests of any single set)

5.2.3 Experimental results


Experiments show that in 1 g neither continuous nor discrete excavation crosses the lightweight
threshold. On the other hand, in gravity offloaded 1/6 g, discrete excavation crosses the lightweight
threshold, while continuous excavation still does not.
Three or four runs were conducted at each of the test conditions, including baseline runs of
driving without digging. Total station data were analyzed to calculate excavator speed during
each test, as shown in Figure 5.4. The excavator maintains constant forward progress in all cases
except discrete excavation with gravity offload. Average speed (as well as standard deviation)
for the various test cases, is summarized in Table 5.2.
Tests in 1 g exhibit a slightly slower speed, because the higher weight compresses the compliant spring tires and reduces their radius. Excavation and gravity offload both introduce a
small amount of additional variability in speed compared to driving without digging in 1 g.
Continuous and discrete excavation in 1 g, as well as continuous excavation in gravity of84

0.04

0.035

0.035

0.03

0.03
Speed (m/s)

Speed (m/s)

0.04

0.025
0.02
0.015

0.02
0.015

0.01

0.01

0.005

0.005

0
0

20

40
Time (s)

60

0
0

80

40
Time (s)

60

80

20

40
Time(s)

60

80

0.035

0.035

0.03
Speed(m/s)

0.03
0.025
0.02
0.015

0.025
0.02
0.015

0.01

0.01

0.005

0.005

0
0

20

0.04

0.04

Speed (m/s)

0.025

20

40
Time (s)

60

80

Figure 5.4: Excavator forward driving speed during continuous excavation in 1 g (top left),
discrete excavation in 1 g (top right), continuous excavation in gravity offloaded 1/6 g (bottom
left), and discrete excavation in gravity offloaded 1/6 g (bottom right; time axes aligned at stall
point). The excavator maintains constant forward progress in all cases except discrete excavation
with gravity offload.

85

Figure 5.5: Discrete bucket at end of excavation tests without gravity offload (left; full with
approx. 45 kg of GRC-1) and with gravity offload (right; only 15-20 kg of GRC-1 collected)
floaded 1/6 g, all collected approximately 45 kg during each 2.5 m test run. Discrete excavation
in gravity offloaded 1/6 g collected only 15-20 kg, in contrast. Figure 5.5 shows the state of fill
of the discrete bucket after 1 g and 1/6 g experiments.
Gravity offload was controlled with sufficient precision to avoid pulling the excavator forward or backward. Figure 5.6 shows longitudinal cable angle and cable tension for a continuous
excavation test. Cable angle was unbiased about vertical, with a mean value of just 0.1 degrees.
Transient motions of the cable did not exceed 0.8 degrees from vertical for more than a fraction
of a second; with a cable tension of 2600 N, this corresponds to brief transients of 35 N, or 7%
of offloaded excavator weight. Cable tension varies just 1% which, amplified by the offloading
ratio, corresponds to 5% variation in the offloaded excavator weight. Figure 5.7 shows longitudinal cable angle and cable tension for a discrete excavation test. Variability in angle and tension
were again small.

5.3

Conclusions regarding the lightweight threshold

All excavators have a lightweight threshold in the operating space, below which their productivity is limited. This threshold is crossed at lower weights for continuous excavators than
discrete excavators. The lightweight threshold is described by a non-dimensional quantity, the
86

1.5

2700

2680
Cable tension (N)

Cable angle (degrees)

1
0.5
0
0.5

2640

2620

1
1.5
0

2660

20

40
60
Time (s)

80

2600
0

100

20

40
60
Time (s)

80

100

1.5

2580

2560
Cable tension (N)

Cable angle (degrees)

Figure 5.6: Longitudinal angle (left) and tension (right) of the gravity offloading cable during a
continuous excavation experiment, showing minimal variation.

0.5
0
0.5
1
1.5
0

2540
2520
2500
2480
2460

20

40
60
Time (s)

80

2440
0

100

20

40
60
Time (s)

80

100

Figure 5.7: Longitudinal angle (left) and tension (right) of the gravity offloading cable during a
discrete excavation experiment, showing minimal variation.

87

lightweight number L, that relates payload ratio, excavation resistance, and excavation thrust.
For continuous excavation, L depends on the initial excavation resistance coefficient, F0,cont .
For discrete excavation it depends on the rate of increasing resistance with respect to increasing

payload, Fdisc
. The lower lightweight threshold for continuous excavators hinges on Fdisc
being

larger than F0,cont ; two distinct experiments show that Fdisc


is orders of magnitude larger than

F0,cont . Although this does not prove the result generally, it strongly suggests that it likely holds
in all but the most degenerate of cases.
An excavators lightweight threshold can be estimated using laboratory experiments. Drawbar pull tests with a full-scale wheel and soil bin excavation tests with a full-scale tool predict
T and F , respectively, for terrestrial gravity. The dependence of L on reduced gravity is not
fully characterized, but assuming direct scaling of L with g provides a balanced estimate. Future
research involving reduced gravity flights to directly investigate the effects of g on drawbar pull
coefficient and excavation resistance would provide more refined estimates for L(g).
Testing in Earth gravity is an inadequate evaluation of planetary excavators, as it underestimates detrimental effects of reduced gravity on both traction and excavation resistance. Gravity
offloaded testing effectively makes the assumption of direct scaling of L with g in an experimental context. This research developed a novel experimental apparatus for achieving gravity
offload, and was the first to subject excavators to offload while they dig.

88

6 Lightweight excavator development


This research developed a novel robotic bucket-wheel excavator prototype, based upon the principles identified in prior chapters of this thesis. It is distinguished by high payload ratio and
high driving speed, features shown to govern lightweight excavator productivity in Chapter 3.
It is a continuous excavator, the preferable tooling configuration according to the theory and
experiments presented in Chapter 5. Its wheels were developed for high drawbar pull, another
important characteristic discussed in Chapters 4 and 5. The excavator prototype is shown in
Figure 6.1.
Lightweight excavator productivity is governed by productivity during the haul stage of the
load-haul-dump cycle, as discussed in Section 3.4. This depends directly on payload ratio and
driving speed, treated here, but also on operational efficiency (percentage of time spent actually performing work, as opposed to waiting for commands or performing computations). High
operational efficiency relies upon high performance autonomy or teleoperation algorithms, recommended as a direction for related future work. Combining high payload ratio and high speed
driving is an important improvment upon the state of the art. Past lightweight excavator prototypes were too slow or carried too little regolith payload, as summarized in Table 6.1.

6.1

Excavation tooling configuration

Bucket-wheel excavators have been shown to produce low resistance forces suitable for lightweight
operation [36]. Bucket-wheels, and any other continuous excavators such as bucket-ladders, also
89

Robot

Mass

Payload
Ratio

Driving
Speed

Bucket wheel
excavator

< 100 kg

n/a

[52]

Bucket drum
excavator

< 100 kg

Mod.

< 5 cm/s

[17]

Bucket ladder
excavators

< 100 kg

High

Various

[NASA]

NASA Cratos
scraper

< 100 kg

High

5 cm/s

[16]

Juno
loadhaul-dump

> 300 kg

Low

> 1 m/s

[67]

NASA Chariot
w/
LANCE
blade

> 1000 kg

Low

> 1 m/s

[39]

NASA Centaur II w/
bucket

> 500 kg

Low

> 1 m/s

[NASA]

Image

Table 6.1: Lunar and planetary excavation robot prototypes

90

Ref.

Figure 6.1: Lightweight robotic excavator prototype featuring central bucket-wheel, high payload dump-bed, high traction wheels, and high-speed wheel actuation

do not suffer from increasing resistance from soil accumulation described in the previous chapters. Prototype bucket-wheel excavators have had difficulty transferring regolith from bucketwheel to collection bin in the past, and as a result bucket-ladders have gained favor [37].
Bucket-ladders use chains to move buckets along easily shapeable paths, making transfer to a
collection bin easy. Winners of the NASA Regolith Excavation Challenge and subsequent Lunabotics mining competitions (which require digging in lunar regolith simulant for 30 minutes)
all employed bucket-ladders driven by exposed chains [49]. However, bucket-ladder chains are
exposed directly to the soil surface and these would degrade very quickly in harsh lunar regolith
and vacuum. Exposed bucket-ladder chains are thus not relevant to operation in lunar conditions.
A novel excavator configuration, with bucket-wheel mounted centrally and transverse to driving direction, achieves direct regolith transfer into a dump-bed. The bucket-wheel is a single
moving part, with no need for chains or conveyors. This reduces complexity and risk from regolith and dust. Once regolith has been carried to the top of the wheel in an individual bucket, it
91

Figure 6.2: Robotic excavator configuration with transverse bucket-wheel and large dump-bed
(left). Close-up of unique direct regolith transfer into dump-bed enabled by transverse orientation
(right)
drops down out the back of the bucket and into a dump-bed. This configuration offers a simple
solution to the transfer problem for bucket-wheels identified in past literature. The dump-bed
transfer concept is shown in Figure 6.2, and implemented in practice in Figure 6.3.
The large dump-bed achieves a high payload ratio, enabling productive execution of excavation tasks.

6.1.1 Testing transverse bucket-wheels


The novel bucket-wheel excavator configuration simplifies regolith transfer into a dump-bed, but
it is important to establish if that does not come at a cost, such as higher excavation resistance. A
transverse bucket-wheel configuration must not lose the low resistance that makes bucket-wheels
desirable in the first place.
Excavation forces and production rates of bucket-wheels digging in lunar simulant are measured experimentally. Experiments compare resistance forces encountered by bucket wheels
92

Figure 6.3: Sand in dump-bed, having been transfered directly by gravity from bucket-wheel

advancing through GRC-1 lunar simulant in a transverse configuration (axis of rotation along
direction of travel) and in a forward configuration (axis of rotation lateral to direction of travel).
An experimental apparatus pushes a bucket-wheel along a direction of travel while rotating
it; the bucket-wheel orientation can be set either transverse or forward. A load cell measures the
horizontal force opposing travel.
Excavation resistance for a transverse bucket-wheel is shown to depend on rotation speed (as
a ratio to forward advance rate). Once a sufficiently high rotation speed is achieved, there is little
difference in excavation resistance between transverse and forward bucket-wheel configurations.
To further reduce excavation resistance for the excavator prototypes transverse bucket-wheel,
cutting faces are angled outward (see Figure 6.6). This prevents bulldozing by the cutting face.
During gravity offloaded bucket-wheel excavation tests, excavation resistance forces were
low enough that mobility was unaffected. Force data were collected using a 6-axis force/torque
sensor mounted between the bucket-wheel actuator and the robot chassis. Lateral and longitudinal forces acting on the bucket-wheel were on the order of 10-20 N, and 5-10 N, respectively (as
shown in Figure 6.7). Lateral forces of this magnitude are less than 5% of the vehicles offloaded
weight, and did not induce discernable slew or yaw in the robots trajectory.
93

Figure 6.4: Experimental apparatus for bucket-wheel orientation testing

Figure 6.5: Transverse bucket-wheels do not exhibit significantly higher excavation resistance
once bucket rotation speed is sufficient

94

30

30

25

25
Longitudinal force (N)

Lateral force (N)

Figure 6.6: Side view of bucket-wheel, noting the cutting face angle. The angle is 14 degrees on
both faces of the transverse bucket-wheel to enable excavation during both forward and backward
driving.

20
15
10
5
0
0

20
15
10
5
0

20

40

60
Time (s)

80

100

120

20

40

60
Time (s)

80

100

120

Figure 6.7: Lateral (lower left) and longitudinal (lower right) excavation resistance forces are
low enough that excavator mobility is unaffected, even with gravity offload. Lateral forces, due
to the transverse bucket-wheel orientation, do not cause the excavator to slew or yaw. Lateral
(blue) and longitudinal (red) directions relative to bucket-wheel are indicated in the upper image.
95

6.2

Excavator mobility system

High driving speed, while shuttling between dig and dump sites, governs lightweight excavator
productivity, as shown in Chapter 3. The maximum driving speed of the excavator developed in
this work is 0.41 m/s, as measured during field testing. For comparison, the Lunakhod rovers
(the fastest planetary mobile robots deployed to date) typically operated at 0.26 m/s. They had a
top speed approaching 0.56 m/s, although this speed was used quite infrequently [38].
The excavators wheels are made of lightweight composite materials, and are rigid. To
achieve high drawbar pull with rigid wheels, grousers are employed. Grouser spacing and height
are selected to mitigate resistive forward soil flow, in compliance with the grouser spacing equation derived in Appendix B. Nominal operating conditions of 20% slip and 10% sinkage (measured as a percentage of wheel radius) are assumed. The selection of a 60 cm diameter wheel
with 36 grousers fixes the appropriate grouser height at 2 cm. The excavators wheels can be
seen prominently in Figure 6.10.
Soil flow imaging tests confirm that the selected grouser geometry appropriately mitigates
resistive forward flow. Figure 6.8 shows experimental data of soil flowing beneath the wheel.
Forward flow induced by a wheel with insufficient grousers is shown in Figure 6.9 for comparison.

6.3

Conclusions regarding lightweight robotic excavator development

Continuous excavator configurations, such as bucket-wheels and bucket-ladders, are preferable


to discrete wide bucket excavators, such as scrapers and front-loaders, for lightweight lunar
and planetary excavation. Bucket-wheel excavators produce low resistance forces that enable
lightweight operation, but in the past have had difficulty transferring regolith from bucket-wheel
to collection bin or dump-bed (necessitating impractical conveyor systems). As a result, bucket96

Figure 6.8: Grouser spacing tests for excavator wheel. Processed imagery of soil motion induced
by driving at 20% slip, shown by pairs of processed images at 6 instances. The top image of each
pair shows soil velocity magnitude (colors range from blue for stationary soil to red for maximum
observed speed); the bottom image shows soil velocity direction (colors correspond to directions
indicated by the color wheel in the bottom right). Red dots are added to aid in following the
progress of three individual grousers (each dot indicates the base of a grouser). Periodic effects
of grouser interactions are observed, with little to no resistive forward flow (which would appear
as large yellow areas in direction plots) seen. Compare to Figure 6.9

Figure 6.9: Processed imagery of soil motion with a wheel with insufficient grousers. The (lower)
direction plot shows a large region of resistive forward flow, indicated by yellow and orange.
97

Figure 6.10: Excavator wheels with proper grouser spacing ensure high traction performance
ladders have gained favor, but their exposed chains would fare poorly in harsh lunar regolith and
vacuum.
A centrally mounted and transverse bucket-wheel configuration, developed in this work,
achieves simplified transfer of regolith into a dump-bed with no significant increase in excavation resistance. The dump-bed is designed to collect substantial payload ratio. The excavator
also has a high driving speed, as payload ratio and driving speed govern lightweight excavator productivity. The design and testing of the excavators rigid wheels demonstrate successful
application of a grouser spacing equation derived as part of this work.

98

7 Conclusions and future work


7.1

Conclusions

This thesis shows that there is a quantifiable, non-dimensional threshold that distinguishes lightweight
from heavy excavation. This threshold is crossed at lower weights for continuous excavators than
discrete excavators. The lightweight threshold relates payload ratio (weight of regolith payload
collected to empty robot weight), excavation resistance (force imparted on an excavator by cutting and collecting soil), and excavation thrust (force supplied by an excavator that is available
for cutting soil).

Payload ratio governs lightweight excavator productivity


Payload ratio, driving speed, and operational efficiency govern productivity of excavation
tasks by small robots. This has been shown using simulation and sensitivity analysis. Experiments using Lysander validate these sensitivity analysis results quantitatively, by varying driving
speed, payload ratio, and number of wheels and measuring output productivity. The task had
a relatively short travel distance (3 m) between dig and dump. Longer distances would be expected to only further increase the importance of payload ratio and driving speed, as an even
higher percentage of the task would be devoted to hauling regolith, as opposed to digging or
dumping.
Experiments lend credence to simulated results and highlight the reality of some potentially
counter-intuitive findings. It may not be surprising that payload ratio and driving speed should
govern productivity, as making each load-haul-dump cycle faster and carrying more load each
99

time directly speed up the task. The relatively negligible effect of changing the number of wheels
is not necessarily obvious though; one might have thought that added traction could significantly
speed up the digging process and thus overall task productivity. Both experiment and simulation
suggest that optimizing a traction system is not the best place to expend resources that should
rather be used maximizing speed and payload ratio.
The prominence of the three hauling productivity parameters shows that hauling is the important part of load-haul-dump excavation tasks, in terms of productivity. An important underlying
assumption in this result is that the excavators in question can achieve a basic level of excavation
capability. If an excavator gets stuck while digging, or must regularly correct and adjust its cuts,
the time spent digging would grow relative to time spent hauling.

Continuous excavation resistance remains constant as payload accumulates, while discrete excavation resistance increases without bound
The excavation thrust coefficient, F , is bounded in continuous excavation, and increases
approximately linearly with payload accumulation in discrete excavation. F and dF /dP for
cutting cohesive soil are predicted to increase for reduced gravity and for reduced mass & scale.
Boles experiments support this prediction for F in reduced gravity [13]. In cohesionless soil, F
and dF /dP are predicted to remain constant; this is supported by experimentation herein.
The ratio P20 /W is an appropriate metric to use for excavation thrust coefficient, T, when
assuming thrust is generated through traction. When slip goes above 20%, the mobility of most
wheels can degrade rapidly; maintaining drawbar pull low enough to keep slip below 20% satisfies the assumption that an excavator maintains nominal capability during digging. T = P20 /W
is approximately constant with changing load, for a given wheel with scale and gravity kept constant. On the other hand, T is predicted to decrease for reduced gravity and for reduced mass &
scale. Kobayashis experiments support this prediction for reduced gravity [43].
Excavation is thus more difficult on planetary surfaces, especially on the Moon, than it is
on Earth. Lightweight operation with small robots and/or in reduced gravity disproportionately
100

reduces excavation thrust while also increasing excavation resistance disproportionately in cohesive lunar regolith. For these reasons, excavation in Earth gravity (even with a robot of relevant
scale in a regolith simulant) overpredicts the performance of excavators in reduced gravity.

Continuous excavators can operate productively at lower weights than


discrete excavators
All excavators have a lightweight threshold in the operating space, below which their productivity is limited. This threshold is crossed at lower weights for continuous excavators than
discrete excavators. The lightweight threshold is described by a non-dimensional quantity, the
lightweight number L, that relates payload ratio, excavation resistance, and excavation thrust.
For continuous excavation, L depends on the initial excavation resistance coefficient, F0,cont .
For discrete excavation it depends on the rate of increasing resistance with respect to increasing

payload, Fdisc
. The lower lightweight threshold for continuous excavators hinges on Fdisc
being

larger than F0,cont ; two distinct experiments show that Fdisc


is orders of magnitude larger than

F0,cont . Although this does not prove the result generally, it strongly suggests that it likely holds
in all but the most degenerate of cases.
An excavators lightweight threshold can be estimated using laboratory experiments. Drawbar pull tests with a full-scale wheel and soil bin excavation tests with a full-scale tool predict
T and F , respectively, for terrestrial gravity. The dependence of L on reduced gravity is not
fully characterized, but assuming direct scaling of L with g provides a balanced estimate. Future
research involving reduced gravity flights to directly investigate the effects of g on drawbar pull
coefficient and excavation resistance would provide more refined estimates for L(g).

Excavating in Earth gravity does not adequately test planetary excavators


Testing in Earth gravity is an inadequate evaluation of planetary excavators, as it underestimates detrimental effects of reduced gravity on both traction and excavation resistance. Gravity
offloaded testing effectively makes the assumption of direct scaling of L with g in an experimen101

tal context. This research develops a novel experimental apparatus for achieving gravity offload,
and was the first to subject excavators to offload while they dig. A 300 kg excavator offloaded to
1/6 g successfully collects 0.5 kg/s using a bucket-wheel, with no discernable effect on mobility.
For a discrete excavator of the same weight, production rapidly declines as rising excavation
resistance stalls the robot; in total the discrete bucket collects less than 20 kg of regolith.

Prior study of lightweight excavation had not identified the core principles
There is no consensus on appropriate excavation force modeling for lunar excavation. However, it is instructive to rise above the fray of contrasting models and focus on their commonly
shared features. Any estimate of excavation resistance must take into account soil weight (and
thus friction) and cohesion. Surcharge is also very important, particularly for discrete excavation; the weight, and perhaps cohesion, of the accumulating soil comprise this surcharge. These
common features provide a theoretical framework for broadly predicting dependence on key
variables such as soil density and cohesion as well as gravity and cut depth.
Past experiments have studied the effects of many excavation parameters, and have shown
that bucket-wheel excavators produce low resistance forces suitable for lightweight operation.
Only preliminary efforts had been made to study excavation forces in reduced gravity. Experiments with excavator prototypes simulating low gravity constitute a novel contribution to the
field of study.
The wide variability in configurations resulting from lunar excavation trade studies and prototype developments highlight the lack of consensus on appropriate configurations for lightweight
excavators. The result that most closely resembles consensus is the fact that bucket-ladder
trenchers have won the Regolith Excavation Challenge and Lunabotics mining competitions each
of the 4 times such competitions were held.

A transverse bucket-wheel excavator offers a unique solution for lightweight


excavation
102

Continuous excavator configurations, such as bucket-wheels and bucket-ladders, are preferable to discrete wide bucket excavators, such as scrapers and front-loaders, for lightweight lunar
and planetary excavation. Bucket-wheel excavators produce low resistance forces that enable
lightweight operation, but in the past have had difficulty transferring regolith from bucket-wheel
to collection bin or dump-bed (necessitating impractical conveyor systems). As a result, bucketladders have gained favor, but their exposed chains would fare poorly in harsh lunar regolith and
vacuum.
A centrally mounted and transverse bucket-wheel configuration, developed in this work,
achieves simplified transfer of regolith into a dump-bed with no significant increase in excavation resistance. The dump-bed is designed to collect substantial payload ratio. The excavator
also has a high driving speed, as payload ratio and driving speed govern lightweight excavator productivity. The design and testing of the excavators rigid wheels demonstrate successful
application of a grouser spacing equation derived as part of this work.

7.2

Contributions

The major contribution of this thesis is the reduction of risk related to future planetary excavation
missions. It also establishes directions and resources to continue refining the results.

7.2.1 Bringing planetary excavation missions forward


Identifying the core principles of lightweight excavation increases the probability of successful
excavation operations. At the same time, distinguishing features that enable lower mass excavators reduces mission cost. Taken together, these factors increase the feasibility of near-term
planetary excavation missions. Successful small scale missions could then be scaled up, following the approach used for the Mars rovers.
The key developments that lead to this contribution are a quantifiable lightweight thresh103

old and a methodology for estimating it for planetary conditions. An excavators lightweight
number, L, relates to its operational risk. Determining that a proposed excavators predicted L
(in planetary conditions) is near or below 1, for example, could prevent haphazardly sending a
discrete excavator into dangerous operating conditions. On the other hand, a very high L means
that a lower mass machines could just as well meet excavation requirements.
The insight that continuous excavation has inherently higher L, and thus is more applicable to lightweight excavation, reveals excavator designs that increase mission confidence while
lowering cost. This brings in-situ resource utilization (ISRU) missions forward, by enabling the
production of consumables from native regolith and the building of earthwork infrastructure, at
lower risk and cost.

7.2.2 Establishing resources and direction for future work


This work developed a novel bucket-wheel excavator. This robotic platform is a resource that
enables further research into planetary excavation tasks. For example, it can be used to study operational issues of deep digging. For safe excavation to depth, the sides of a pit must be terraced
or gently sloping; this prevents collapse and cave in of the pit walls. The bucket-wheel excavator
enables testing of operation plans for excavating such a pit. Another example application is using
the robots sensing and actuation to test control strategies for dealing with buried rocks.
This research also developed a load-haul-dump model, Regolith Excavation, MObility &
Tooling Environment (REMOTE). The model could be a useful tool for mission planners and
excavator designers alike.

7.3

Future work

Future research on lightweight excavation would benefit from testing in reduced gravity flights.
These provide the most representative test environment short of actually operating on a planetary
104

Figure 7.1: Rocks produce large spikes in excavation resistance, posing a challenge for
lightweight continuous and discrete excavators alike.

surface, as excavator and regolith are both subject to reduced gravity. The dependence of the
lightweight number, L, on reduced gravity is not yet fully characterized. Reduced gravity flights
to directly investigate the effects of g on drawbar pull coefficient and excavation resistance would
provide more refined estimates for L(g).
Another important direction for future study is deep excavation in the presence of submerged
rocks, which pose challenges for lighweight continuous and discrete excavators alike. Initial tests
show that rocks produce large spikes in excavation resistance, as shown in Figure 7.1. Rocks are
especially important to consider during deep excavation. For safe excavation to depth, the sides
of a pit must be terraced or gently sloping; this prevents collapse and cave in of the pit walls.
Shallow slopes translate into large excavated volumes, though, and thus higher probability of
encountering rocks of any given size.
Continuous excavators have a lower lightweight threshold than discrete excavators; this re
sult hinges on Fdisc
being at least 3 times larger than F0,cont . Two distinct experiments herein

show that Fdisc


is orders of magnitude larger than F0,cont . This strongly suggests that continuous

excavators will have a lower lightweight threshold in all but the most degenerate of cases, though
105

Figure 7.2: Bucket soil flow imaged and processed. The top image shows soil velocity magnitude
(colors range from blue for stationary soil to red for maximum observed speed); the bottom image
shows soil velocity direction (colors correspond to directions indicated by the color wheel in the
bottom right)
it does not directly prove the result generally. Experiments to confirm the generality of these
results could explore a wide assortment of continuous and discrete excavation tools.
Additional experiments to study the effects of scaling on traction and excavation resistance
would provide further insight into the dependence of excavation on changes in weight. Such
experiments include (continuous) bucket-wheel and (discrete) flat-plate tests in cohesive soil, as
well as rigid and compliant wheel scaling in relevant soils.
Excavation modeling is fruitful ground for future research. Some aspects of excavation tasks
are ignored within REMOTE, including wheel slip sinkage and soil accumulation at the bucket
cutting edge. Soil accumulation and slip sinkage are appropriate directions for future modeling
development. Excavation resistance varies significantly during a cut as soil accumulates in the
bucket, and classical models only approximate this effect. They fail to capture excavation soil
flows. Novel soil flow imaging techniques developed as part of this work (see Appendix B) shed
light on soil flow during excavation, as shown in Figures 7.2 and 7.3. Discrete Element Modeling
(DEM) shows promise in modeling such excavation soil flows.
106

Figure 7.3: Bucket-wheel soil flow imaged and processed.

107

108

8 Bibliography
[1] Nader Abu El Samid.

Infrastructure robotics: A trade-off study examining both au-

tonomously and manually controlled approaches to lunar excavation and construction. Masters thesis, University of Toronto, 2008. 2.4
[2] Juan H. Agui and Allen Wilkinson. Granular flow and dynamics of lunar simulants in
excavating implements. In ASCE Earth & Space 2010 Proceeding, Honolulu, HI, March
2010. 2.2, 2.2.1, 4.3.1, 5.1, 5.1.1, 5.2.2
[3] Maximilian Apfelbeck, Sebastian Ku, Bernhard Rebele, and Bernd Schafer. A systematic
approach to reliably characterize soils based on bevameter testing. Journal of Terramechanics, 48(5):360371, 2011. 4.4.2
[4] Vladilen Ivanovich Balovnev. New methods for calculating resistance to cutting of soil.
Amerind Publishing Company, 1983. 3.1.2, 3.1.2
[5] Paul Bartlett, David Wettergreen, and William Whittaker. Design of the scarab rover for
mobility and drilling in the lunar cold traps. In International Symposium on Artificial Intelligence, Robotics and Automation in Space, pages 36. Citeseer, 2008. 5.2.1
[6] Mieczyslaw Gregory Bekker. Photographic method of determining the soil action beneath
footings. In Proc. 2nd Int. Conf. on Soil Mechanics and Foundation Engineering, pages
193194, 1948. B
[7] Mieczyslaw Gregory Bekker. Introduction to terrain-vehicle systems. University of Michigan Press Ann Arbor, MI, 1969. 3.1.1
109

[8] Leonhard E. Bernold. Experimental studies on mechanics of lunar excavation. Journal of


Aerospace Engineering, 4(1):922, January 1991. 2.6
[9] Jeffrey K. Berry. System model for lunar regolith excavation machine optimization. Mechanical engineering, Bradley University, Peoria, Illinois, 2006. 2.1.4
[10] Michael J Black and Paul Anandan. The robust estimation of multiple motions: Parametric
and piecewise-smooth flow fields. Computer vision and image understanding, 63(1):75
104, 1996. B
[11] Stephane Blouin, Ahmad Hemami, and Mike Lipsett. Review of resistive force models
for earthmoving processes. Journal of Aerospace Engineering, 14(3):102111, July 2001.
1.4.4
[12] Walter W Boles, David B Ashley, and Richard L Tucker. Lunar-base construction equipment and methods evaluation. Journal of Aerospace Engineering, 6(3):217235, 1993. 1.3,
2.4
[13] Walter W Boles, Wesley D Scott, and John F Connolly. Excavation forces in reduced
gravity environment. Journal of Aerospace Engineering, 10(2):99103, 1997. 2.2.2, 4.3,
4.4, 4.5, 5.2, 7.1
[14] Ha H Bui, Taizo Kobayashi, Ryoichi Fukagawa, and John C Wells. Numerical and experimental studies of gravity effect on the mechanism of lunar excavations. Journal of
Terramechanics, 46(3):115124, 2009. 2.1.4
[15] D Carrier. The four things you need to know about the geotechnical properties of lunar soil.
Lunar Geotechnical Institute, P.O. Box 5056, Lakeland, FL 33807, September 2005. 4.4.2
[16] John J Caruso, Dan C Spina, Lawrence C Greer, Wentworth T John, Clem Michele, Mike J
Krasowski, and Norman F Prokop. Excavation on the moon: Regolith collection for oxygen
production and outpost site preparation. Technical Report 20080012503, NASA Glenn
Research Center, Cleveland, Ohio 44135, 2008. 2.5, 6

110

[17] D Larry Clark, R Patterson, and D Wurts. A novel approach to planetary regolith collection:
the bucket drum soil excavator. In AIAA Space 2009 Conference & Exposition, September
2009. 2.5, 6
[18] J. Craft, J. Wilson, P. Chu, K. Zacny, and K. Davis. Percussive digging systems for robotic
exploration and excavation of planetary and lunar regolith. In Aerospace conference, 2009
IEEE, Big Sky, MT, March 2009. 2.2.2, 2.6
[19] GF Diaz Lankenau, K Skonieczny, WL Whittaker, and DS Wettergreen. Effect of bucketwheel scale on excavation forces and soil motion. Journal of Terramechanics, 49(6):341
348, 2012. 4.4.1
[20] L. Ding, H. Gao, Z. Deng, and J. Tao. Wheel slip-sinkage and its prediction model of lunar
rover. Journal of Central South University of Technology, 17(1):129135, 2010. 3.1.1
[21] Liang Ding, Haibo Gao, Zongquan Deng, Keiji Nagatani, and Kazuya Yoshida. Experimental study and analysis on driving wheels performance for planetary exploration rovers
moving in deformable soil. Journal of Terramechanics, 48(1):2745, 2011. 4.2
[22] Matthew Dunbabin, Peter Corke, Graeme Winstanley, and Jonathan Roberts. Off-world
robotic excavation for large-scale habitat construction and resource extraction. In To Boldly
Go Where No Human-Robot Team Has Gone Before: Papers from the 2006 AAAI Spring
Symposium, Technical Report SS-06-07, pages 95103. AAAI, March 2006. 2.7
[23] R. C. Elphic, D. A. Paige, M. A. Siegler, A. R. Vasavada, V. R. Eke, L. F. A. Teodoro, and
D. J. Lawrence. South pole hydrogen distribution for present lunar conditions: Implications
for past impacts. In 41st Lunar and Planetary Science Conference, page 2732, Woodlands,
TX, March 2010. 1.1, 1.1
[24] D.R. Freitag, A.J. Green, and K.J. Melzer. Performance evaluation of wheels for lunar
vehicles. Technical report, DTIC Document, 1970. 1.4.3, 4.2
[25] Philippe Gachet, Georg Klubertanz, Laurent Vulliet, and Lyesse Laloui. Interfacial behav-

111

ior of unsaturated soil with small-scale models and use of image processing techniques.
Geotechnical Testing Journal, 26:1221, 2003. B
[26] C. A. Gallo, R. A. Wilkinson, R. P. Mueller, J. M. Schuler, and A. J. Nick. Comparison of
isru excavation system model blade force methodology and experimental results. Technical
Report TM2010-215591, NASA, 2010. 2.1
[27] Leslie Gertsch, Robert Gustafson, and Richard Gertsch. Effect of water ice content on
excavatability of lunar regolith. In AIP Conference Proceedings, volume 813, page 1093,
2006. 2.6
[28] Alex Green, Kris Zacny, Juan Pestana, Dennis Lieu, and Robert Mueller. Investigating
the effects of percussion on excavation forces. Journal of Aerospace Engineering, 26(1):
8796, 2013. 2.6
[29] Robert Gustafson and Leslie Gertsch. Cutterhead development for the low-energy planetary
excavator. In Space Technology and Applications International Forum, Albuquerque, NM,
February 2008. American Institute of Physics. 2.6
[30] Grant H Heiken, David T Vaniman, and Bevan M French, editors. The Lunar sourcebook:
A users guide to the Moon. Cambridge University Press, 1991. 2.6, 3.1.1, 3.2.1
[31] DRP Hettiaratchi and AR Reece. The calculation of passive soil resistance. Geotechnique,
24(3):289310, 1974. 4.4.2
[32] DRP Hettiaratchi, BD Witney, and AR Reece. The calculation of passive pressure in twodimensional soil failure. Journal of Agricultural Engineering Research, 11(2):89107,
1966. 2.1, 2.1, 2.1.2
[33] Karl Iagnemma and Steven Dubowsky. Traction control of wheeled robotic vehicles in
rough terrain with application to planetary rovers. The International Journal of Robotics
Research, 23(10-11):10291040, 2004. 3.2.2
[34] M Iai. Scaled experimental study on excavation of lunar regolith with rakes/rippers and
112

flat blade. PhD thesis, Missouri University of Science and Technology, 2010. 2.6
[35] Caterpillar Inc., editor. Caterpillar performance handbook. Caterpillar Inc., Peoria, IL, 33
edition, 2002. 1.4.2, 4.1
[36] LL Johnson and RH King. Measurement of force to excavate extraterrestrial regolith with
a small bucket-wheel device. Journal of Terramechanics, 47(2):8795, 2010. ISSN 00224898. 2.2, 2.5, 6.1
[37] LL Johnson and PJ Van Susante. Excavation system comparison: Bucket wheel vs. bucket
ladder. In Space Resources Roundtable VIII, Golden, CO, November 2006. 2.5, 2.5, 6.1
[38] Simon Kassel. Lunokhod-1 soviet lunar surface vehicle. Technical Report R-802-ARPA,
RAND Corporation, 1971. 2.7, 6.2
[39] R.H. King, P.J. van Susante, and R.P. Mueller. Comparison of lance blade force measurements with analytical model results. In Space Resources Roundtable XI / Planetary &
Terrestrial Mining Sciences Symposium Proceedings, 2010. 2.1, 2.2.1, 2.3, 6
[40] J Ledlie Klosky, Stein Sture, Hon-Yim Ko, and Frank Barnes. Vibratory excavation and
anchoring tools for the lunar surface. In Engineering, Construction, and Operations in
Space V. ASCE, 1996. 2.6
[41] J Ledlie Klosky, Stein Sture, Hon-Yim Ko, and Frank Barnes. Geotechnical behavior of
jsc-1 lunar soil simulant. Journal of Aerospace Engineering, 13(4):133138, 2000. 4.3
[42] T Kobayashi, H Ochiai, R Fukagawa, S Aoki, and K Tamoi. A proposal for estimating
strength parameters of lunar surface from soil cutting resistances. In Earth & Space, 2006.
2.1.3
[43] Taizo Kobayashi, Yoichiro Fujiwara, Junya Yamakawa, Noriyuki Yasufuku, and Kiyoshi
Omine. Mobility performance of a rigid wheel in low gravity environments. Journal of
Terramechanics, 47(4):261274, 2010. 4.2, 4.5, 5.2, 7.1
[44] Takashi Kubota, Kenji Nagaoka, Satoru Tanaka, and Taro Nakamura. Earth-worm typed
113

drilling robot for subsurface planetary exploration. In Robotics and Biomimetics, 2007.
ROBIO 2007. IEEE International Conference on, pages 13941399. IEEE, 2007. 2.5
[45] O Luengo, Sanjiv Singh, and Howard Cannon. Modeling and identification of soil-tool
interaction in automated excavation. In Intelligent Robots and Systems, 1998. Proceedings.,
1998 IEEE/RSJ International Conference on, volume 3, pages 19001906. IEEE, October
1998. 2.7
[46] H.J. Luth and R.D. Wismer. Performance of plane soil cutting blades in sand. Transactions
of the ASAE, 14(2):255262, Mar-Apr 1971. 3.1.2
[47] Scott Moreland, Krzysztof Skonieczny, David Wettergreen, Colin Creager, and Vivake Asnani. Soil motion analysis system for examining wheel-soil shearing. In 17th International
Conference for the Society of Terrain-Vehicle Systems, 2011. B
[48] Scott Moreland, Krzysztof Skonieczny, Hiroaki Inotsume, and David Wettergreen. Soil
behavior of wheels with grousers for planetary rovers. In Aerospace Conference, 2012
IEEE, pages 18. IEEE, 2012. B
[49] Robert P Mueller and Paul J Van Susante. A review of lunar regolith excavation robotic
device prototypes. In AIAA Space, 2011. 2.8, 6.1
[50] Robert P Mueller, Jonathan D Smith, Rachel E Cox, Jason M Schuler, Tom Ebert, and
Andrew J Nick. Regolith advanced surface systems operations robot (rassor). In IEEE
Aerospace, 2013. 2.5
[51] R.P. Mueller and R.H. King. Trade study of excavation tools and equipment for lunar
outpost development and isru. In AIP Conference Proceedings, volume 969, pages 237
244, January 2008. 2.4
[52] T Muff, L Johnson, R King, and MB Duke. A prototype bucket wheel excavator for the
moon, mars and phobos. In AIP Conference Proceedings, volume 699, page 967, 2004.
2.5, 6

114

[53] Tim Muff, RH King, and MB Duke. Analysis of a small robot for martian regolith excavation. In AIAA Space 2001 Conference & Exposition, Albequerque, NM, August 2001.
AIAA. 2.3, 3.1.2
[54] Tejas G Murthy, Ebenezer P Gnanamanickam, Christopher Saldana, and Srinivasan Chandrasekar. Deformation field in indentation of granular materials. In AIP Conference Proceedings, volume 1145, page 263, 2009. B
[55] H. Nayar, A. Jain, J. Balaram, J. Cameron, C. Lim, R. Mukherjee, M. Pomerantz, L. Reder,
S. Myint, N. Serrano, and S. Wall. Recent developments on a simulator for lunar surface
operations. In AIAA SPACE 2009 Conference & Exposition, 2009. 3.1.4
[56] H.A. Oravec, X. Zeng, and V.M. Asnani. Design and characterization of grc-1: A soil for
lunar terramechanics testing in earth-ambient conditions. Journal of Terramechanics, 47
(6):361377, 2010. 2.2.2, 4.4.1, 4.4.2, 5.2.1
[57] Yang Qinsen and Sun Shuren. A soil-tool interaction model for bulldozer blades. Journal
of terramechanics, 31(2):5565, 1994. 2.1.3, 4.3.1
[58] G. Sanders and W. Larson. Progress made in lunar in situ resource utilization under nasas
exploration technology and development program. Journal of Aerospace Engineering, 26
(1):517, 2013. 1.1, 1.2
[59] A Sanin, I Mitrofanov, W Boynton, G Chin, N Demidov, J Garvin, D Golovin, L Evans,
K Harshman, A Kozyrev, et al. Mapping of lunar hydrogen according to the lend neutron
measurements onboard the nasa lro. In Lunar and Planetary Institute Science Conference
Abstracts, volume 41, page 2437, 2010. 1.1
[60] Thomas B. Sheridan. Space teleoperation through time delay: Review and prognosis. IEEE
Transactions on Robotics and Automation, 9(5):592606, October 1993. 2.7
[61] I Shmulevich, Z Asaf, and D Rubinstein. Interaction between soil and a wide cutting blade
using the discrete element method. Soil and Tillage Research, 97(1):3750, 2007. 2.1.3,

115

4.3.1, 4.3.1, 4.6


[62] Sanjiv Singh. The state of the art in automation of earthmoving. Journal of Aerospace
Engineering, 10(4):179188, October 1997. 2.7
[63] Krzysztof Skonieczny, David S. Wettergreen, and W. L. Red Whittaker. Parameters governing regolith site work by small robots. In ASCE Earth & Space 2010 Proceedings,
Honolulu, HI, March 2010. 3.1.5, 3.3
[64] Krzysztof Skonieczny, Scott J Moreland, and David S Wettergreen. A grouser spacing
equation for determining appropriate geometry of planetary rover wheels. In Intelligent
Robots and Systems (IROS), 2012 IEEE/RSJ International Conference on, pages 5065
5070. IEEE, 2012. B
[65] Krzysztof Skonieczny, Matthew Delaney, David S Wettergreen, and William LRed Whittaker. Productive lightweight robotic excavation for the moon and mars. Journal of
Aerospace Engineering, in press, 2013. 3.1, 3.10, 3.12
[66] B. Szabo, F. Barnes, S. Sture, and H-Y. Ko. Effectiveness of vibrating bulldozer and plow
blades on draft force reduction. Transactions of the ASAE, 41(2):283290, 1998. 2.6
[67] Ronny Theiss, Dale Boucher, Marcel Viel, Dave Roberts, and James Kutchaw. Interchangeable payloads for isru mobility chassis. In Space Resources Roundtable XI / Planetary &
Terrestrial Mining Sciences Symposium Proceedings, 2010. 2.5, 6
[68] G.W. Turnage and D.C. Banks. Lunar surface mobility studies past and future. Technical
report, DTIC Document, 1989. 1.4.3, 4.2
[69] Otis R. Walton and Scott M. Johnson. Dem simulations of the effects of particle shape,
interparticle cohesion and gravity on rotating drum flows of lunar regolith. In ASCE Earth
& Space 2010 Proceedings, Honolulu, HI, March 2010. 2.1.4
[70] D. Wettergreen, S. Moreland, K. Skonieczny, D. Jonak, D. Kohanbash, and J. Teza. Design
and field experimentation of a prototype lunar prospector. The International Journal of
116

Robotics Research, 29(12):15501564, 2010. 1.4.3, 4.2, 5.2.1


[71] Christopher V White, John K Frankovich, Philip Yates, George Wells Jr, and Losey Robert.
A capable and temporary test facility on a shoestring budget: the msl touchdown test facility. In 24th Aerospace Testing Seminar, April 8, 2008, Manhattan Beach, California.
Pasadena, CA: Jet Propulsion Laboratory, National Aeronautics and Space Administration,
2008., 2008. 5.2
[72] DJ White, WA Take, and MD Bolton. Soil deformation measurement using particle image
velocimetry (piv) and photogrammetry. Geotechnique, 53(7):619631, 2003. B
[73] A. Wilkinson and A. DeGennaro. Digging and pushing lunar regolith: Classical soil mechanics and the forces needed for excavation and traction. Journal of Terramechanics, 44:
133152, 2007. 2.1, 2.1.1, 2.3, 3.1
[74] Brian M. Willman. Modeling Soil-Tool Interaction in a Lunar Soil Simulant. PhD thesis,
Texas A&M University, 1994. 2.1, 2.2.2, 2.3
[75] R.D. Wismer and H.J. Luth. Performance of plane soil cutting blades in clay. Transactions
of the ASAE, 15:211216, 1972. 3.1.2
[76] J. Y. Wong. Theory of Ground Vehicles. John Wiley & Sons, Inc., 3rd edition, 2001. 3.1.1,
3.2.2
[77] Jo-Yung Wong. Behaviour of soil beneath rigid wheels. Journal of Agricultural Engineering Research, 12(4):257269, 1967. B
[78] Jo-Yung Wong and AR Reece. Soil failure beneath rigid wheels. In Proceedings of the
Second International Conference of the International Society for Terrain-Vehicle Systems,
JN Siddall, and PH Southwell, pages 425445, 1966. B
[79] JY Wong. Predicting the performances of rigid rover wheels on extraterrestrial surfaces
based on test results obtained on earth. Journal of Terramechanics, 49(1):4961, 2012. 5.2
[80] JY Wong and VM Asnani. Study of the correlation between the performances of lunar
117

vehicle wheels predicted by the nepean wheeled vehicle performance model and test data.
Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile
Engineering, 222(11):19391954, 2008. 1.4.3, 4.2
[81] Kris Zacny, Robert P. Mueller, Jack Craft, Jack Wilson, Magnus Hedlund, and Joanna
Cohen. Five-step parametric prediction and optimization tool for lunar surface systems
excavation tasks. In ASCE Earth & Space 2010 Proceedings, Honolulu, HI, March 2010.
4.3
[82] A Nikolaevich Zelenin, Vladilen Ivanovich Balovnev, Igor Petrovich Kerov, et al. Machines
for moving the earth. AA Balkema, 1986. 4.4.2
[83] Xiangwu Zeng, Chunmei He, Heather Oravec, Allen Wilkinson, Juan Agui, and Vivake
Asnani. Geotechnical properties of jsc-1a lunar soil simulant. Jou, 23(2):111116, April
2010. 2.2.2

118

A Extensions
This appendix discusses closely related topics that lie outside the scope of this thesis.

A.1

Regolith shaping

This thesis focuses on the load-haul-dump cycle that is central to all relevant planetary excavation
tasks. However, as described in Section 1.2, there is a class of tasks that also requires the regolith
to be shaped, molded, and perhaps compacted after being dumped. This class of tasks includes
building berms and covering habitats.
Bucket-wheels are not conducive to regolith shaping the way that wide flat (discrete) buckets
are. Discrete buckets, such as front-end loaders, can be used to flatten, smooth, and compress
regolith. This versatility is an important feature to trade off against continuous excavators productivity and mobility advantages. Another advantage of discrete buckets is the lower chance of
clogging, due to their inherently open geometry. For these reasons, it may sometimes be desirable to use a discrete excavator even in conditions where it technically falls below the lightweight
threshold.
If a discrete excavator is operating below the lightweight threshold, it will not be able to
safely collect its desired payload ratio in a single cut. However, for operations with a lightweight
number, L, just below 1, the desired payload ratio can be achieved with 2 cuts as long as regolith
from the first cut is cleared from the cutting edge of the bucket. In practice this clearing can
be performed by overshot dumping into a secondary dump-bed, or perhaps by tilting the bucket
119

back until soil shifts to the back of the bucket (as was done in the Lysander experiments described
in Section 3.2). An even smaller L may require 3 cuts, and so on. At some point the operation
begins to resemble continuous excavation, except for the fact that cutting and clearing are performed serially by a single bucket rather than in parallel by several; which is less productive and
less efficient. There will be a point in the trade space where L is so low as to make multi-cut
discrete excavation impractical.
The additional sensing and control complexity must also be taken into account when considering multi-cut operations with a discrete excavator. To ensure slip does not exceed some
safety threshold (of 20%, for example), slip or drawbar load needs to be estimated. Setting an
appropriate cut depth (for a subsequent cut) is also not trivial in soil disturbed by the pushing and
breakout of the previous cut.

A.2

Non-tractive excavation thrust

This thesis assumes that excavation thrust is provided by traction, as excavator configurations
typically considered for space application cut by driving forward. However, configurations exist
that do no rely on traction for cutting. RASSOR (Figure 2.3) has counter-rotating bucket-drums
that can theoretically achieve zero net horizontal excavation resistance. Clamshell excavators are
another example of machines that can achieve zero net horizontal excavation resistance. In these
cases, vertical excavation resistance dominates.
The concepts presented in this thesis can be extended to these non-tractive excavation modes,
with a few minor adjustments. Vertical forces replace longitudinal forces when considering
excavation resistance coefficient, F . Excavation thrust is provided directly by the robot weight,
so T = 1. For discrete vertical excavation (e.g. clamshell), depth increases with cut progress.
Equation 2.1 suggests that excavation resistance increases with the square of depth. Compared
to the linear increase in resistance with cut length observed for horizontal cutting, the advantages
of continuous excavation may thus be even more pronounced in this mode of operation. Further
120

study of this hypothesis is recommended if considering vertical thrust for lightweight excavation.

121

122

B Soil flow imaging and grouser spacing


A novel analysis technique has been developed to enable detailed investigation of robot interactions with granular regolith. This technique provides visualization and analysis capability of soil
shearing and flow as it is influenced by a wheel or excavation tool. During controlled motion a
test implement up against a glass sidewall, images are taken of the sub-surface soil, and are processed with optical flow software. Analysis of the resulting displacement field identifies clusters
of soil motion and shear interfaces. This enables analysis of robot-soil interactions in richer detail than possible before. Prior art relied on long-exposure images that provided only qualitative
insight, while the new processing technique identifies sub-millimeter gradations in motion and
can do so even for high frequency changes in motion (several Hz).
Direct observation of soil motion through glass sidewalls has been utilized in soil mechanics
and terramechanics research for over half a century [6]. Wong concluded experimentally that
as long as shear stress between glass and soil is negligible, the glass surface acts as a plane of
symmetry and the soil behaves as it would directly below an implement twice as wide [77, 78].
One archetypical photographic method for observing soil motion uses long-exposure photos
and distinguishes sharp and streaking soil grains as stationary and moving, respectively. Streaks
in the photos also provide information about the directions of soil motion. With advances in
digital camera technology and computer vision processing techniques, new methods providing
much richer data have become possible [25, 72]. The image processing technique discussed
in this work is similar in its implementation to Murthys, who performed a preliminary study
of sand displacement under a footing-like indenter [54]. This work applies these newest soil
123

observation techniques to wheel and tool interactions pertinent for planetary robots.
A description of experimental apparatus hardware is found in [47]. A digital SLR camera
with a 50mm macro lens is used to image the soil where it interfaces with the test implement,
logging frames simultaneously with the rest of the telemetry. A frame rate of 8 frames-persecond is used and is sufficiently fast for the slow implement speeds. The camera is mounted
perpendicular to the soil bin glass wall and travels with the implement in the horizontal direction
as the carriage moves. External halogen flood lights at a high angle (from the normal) to the
glass illuminate the soil particles.
Image processing comprises of optical flow and clustering techniques. An overview of the
process described herein is presented in Figure B.1. The optical flow algorithm [10] tracks displacement of soil regions relative to a prior frame and calculates a motion vector at each pixel.
Initial clustering separates each image into soil and not soil regions. Additional processing
and output is continued only for soil regions. The magnitude of flow at each pixel of the soil
regions is calculated from the optical flow vector fields. Soil flow is clustered into significant
and insignificant magnitudes of motion. No explicit threshold is used to demarcate these clusters, but rather automatically adaptive k-means clustering is used. The shear interface is derived
from the boundary between significant and insignificant motions. Soil flow direction is calculated from the optical flow vector fields, for soil regions exhibiting significant soil flow. Soil flow
in any direction (360 degrees) is visualized, and an additional boundary is identified at points
where the soil transitions between forward and rear flow. Figure B.2 is a sample output of the
process, showing soil flow magnitude, shear interface between significant and insignificant flow,
soil flow direction (within region of significant flow), and boundary between forward and rear
flow.
This soil imaging technique was used to derive a quantitative expression for determining
appropriate grouser spacing for rigid wheels. The intuition guiding the search for a grouser
spacing equation is an endeavor to ensure grousers encounter soil ahead of a wheel before the
124

Figure B.1: Overview of image processing and output


wheel rim does. When a wheel rim encounters soil it bulldozes it forward and compacts it,
producing resistance, in addition to shearing it to produce thrust [48]. Grousers, on the other
hand, have a net rearward motion near the bottom of a rotating wheel, and thus pull soil back and
constrain it from undergoing resistive forward flow. The grouser spacing equation is discussed
in detail in [64], and the derivation is provided here for reference.

125

Figure B.2: Sample processed output for driven wheel. Soil flow speed (upper) is denoted from
blue (static) to red (max. speed). Soil flow direction (lower) within the shear interface is denoted
according to the color wheel is the bottom right.

126

127

128

129