Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

2014 American Control Conference (ACC)

June 4-6, 2014. Portland, Oregon, USA

On-Line Estimation of Vehicle Motion and Power Model Parameters


for Skid-Steer Robot Energy Use Prediction
Jesse Pentzer, Sean Brennan, and Karl Reichard

Abstract— This paper presents a method of estimating skid- focused on modeling tire/tread slippage since the pioneering
steer robot power usage using on-line estimation of terrain and terramechanics work of Bekker in the 1960’s [3].
kinematic parameters. For vehicles operating at low speeds on The vast majority of commercially available skid-steer
hard, flat surfaces, kinematic models utilizing the instantaneous
centers of rotation (ICRs) of the tracks or wheels of a skid- vehicles utilize batteries for energy storage and provide little
steer vehicle have been shown to provide accurate motion and to no information on the remaining endurance of a robot in
power use estimation. Previous work has relied on post-process operation. Such endurance information is useful not only for
optimization to learn necessary ICR location and terrain infor- manual teleoperation but also for planning energy-efficient
mation for motion and power modeling. The work presented traversal paths [4]. Two areas of important research affect
here utilizes an extended Kalman filter for learning ICR
locations and the recursive least squares algorithm for learning skid-steer endurance estimation significantly: improving bat-
terrain-related power model parameters. The algorithms have tery models to determine the remaining available energy and
been implemented on a wheeled skid-steer vehicle, and field test developing accurate estimates of the energy required for a
results show good estimation of motion and power usage using robot to complete its mission. The two areas are coupled
no prior terrain information and only knowledge of vehicle in that remaining energy may in fact be a function of the
geometry and mass distribution, intermittent GPS and heading,
and odometry information from the slipping tires/treads. expected discharge rate for the battery, and the choice of
future missions (and hence the discharge rate) may strongly
I. INTRODUCTION depend on the current estimate of remaining battery energy.
The focus of this work is the second area, that of developing
The area of ground robotics has experienced dramatic accurate and adaptable models for predicting the power usage
growth in the past decade. A significant portion of this of skid-steer robots.
growth has been in response to a need for effective impro- Significant work has been performed on modeling the
vised explosive device (IED) countermeasures [1], but robots motion of skid-steer vehicles. While dynamic models of
are also seeing increased use in disaster search and rescue skid-steer motion have been developed and verified ex-
situations [2]. One commonality between robots designed as perimentally [6], [7], implementation on robotic platforms
IED countermeasures and as search and rescue vehicles is is difficult due to the complexity and need for extensive
the use of skid-steer locomotion. knowledge of vehicle and terrain parameters. Research has
Skid-steer locomotion is widely used by ground mobile turned to kinematic models of skid-steer motion to enable
robots due to its simplicity, effective mobility, and durability. easy implementation and reduce the number of parameters to
Rather than using a steering actuation system, skid-steer be identified. The work of [8] presents an extended Kalman
robots turn by adjusting the speed of the left and right filter (EKF) for the estimation of track slip on a skid-steer
side tracks or wheels. At the extreme, when the left and vehicle from trajectory measurements. The coupling of slip
right sides have opposite speeds, skid-steer vehicles are able estimates with a non-linear terrain model is proposed for
to turn in place, which provides useful mobility in tight robust control of skid-steer vehicles. Slip estimation using
spaces and during teleoperation. When skid-steer vehicles EKFs [9], [10] has continued to be a significant area of
turn, all contact points on the tracks/wheels of the vehicle research, but utilizing slip estimates in power use prediction
in front or behind the center of rotation must be dragged for skid-steer vehicles remains an important area of research.
laterally across the surface. The large amounts of slippage An EKF is utilized to estimate traction coefficients of friction
that occur during skid-steer operation make modeling the in [11], but the extension to power use modeling is not made.
movement of skid-steer vehicles extremely difficult. Also, Another approach is to fit exponential functions of turning
the power requirements for skid-steer motion are closely radius to slippage [12]. Interestingly, exponential functions
related to this slippage, and power consumption literature has have also been used for the estimation of power usage with
good results [13], but twelve parameters must be learned
J. Pentzer is with the Department of Mechanical and Nuclear from experimental data for each terrain to be modeled.
Engineering, The Pennsylvania State University, State College, PA An alternative approach to the estimation of slip is to
jlp5573@psu.edu
S. Brennan is with the faculty of mechanical and nuclear engineering, estimate the ICR of each track or wheel-set on a skid-steer
Department of Mechanical and Nuclear Engineering, The Pennsylvania State vehicle. This approach has been applied successfully when
University, State College, PA sbrennan@psu.edu ICR locations are learned in post-processing on recorded data
K. Reichard is with the Applied Research Laboratory and faculty
of acoustics, The Pennsylvania State University, State College, PA [14]. ICR kinematics have also been applied successfully for
kmr5@psu.edu modeling of power usage for tracked and wheeled skid-steer

978-1-4799-3274-0/$31.00 ©2014 AACC 2786

Authorized licensed use limited to: Lodz University of Technology. Downloaded on March 15,2021 at 14:57:57 UTC from IEEE Xplore. Restrictions apply.
vehicles [15], [16], but again, terrain parameters relating to The location of the right side track ICR is yICRr , and the
power usage are learned through post-process optimization. location of the left side track ICR is yICRl . The longitudinal
Recent work presented an approach focused on breaking the location of both the left and right track ICRs along the X
model developed by Wong [7] into two parts to improve axis is xICRv . A detailed development of these equations is
computation speed [17]. Results show accurate estimation provided in [19].
of skid-steer power use, but the model requires extensive
A. Estimation of ICR Locations
knowledge of terrain parameters.
The approach presented here builds upon previous work A kinematic extended Kalman filter, derived in [19], can
in ICR kinematics by identifying ICR locations and power be designed to estimate ICR locations using the kinematic
model parameters on-line during operation of the robot. model (1)-(3) and measurements of position and heading of
Specifically, an EKF is designed to estimate the locations the vehicle. Rotating the local frame velocities produced by
of track/wheel Instantaneous Centers of Rotation (ICRs), (1)-(3) to the global north-east-down frame yields
which provide a kinematic model of vehicle movement. The Ṅ = vx cos ψ − vy sin ψ (4)
estimated ICR locations are then combined with on-board
Ė = vx sin ψ + vy cos ψ (5)
sensor measurements in a recursive least squares algorithm to
estimate power model parameters related to friction between where ψ is the heading angle defined as positive clockwise
the track/wheels and ground and rolling resistance in linear and zero in the north direction. The kinematic model for the
motion. EKF is
The remainder of this paper is organized as follows.   
vx cos ψ − vy sin ψ + wN


Section II provides a brief overview of ICR kinematics  Ė   vx sin ψ + vy cos ψ + wE 
and a description of the EKF algorithm for estimation of  
 ωz   − Vx −Vx
 l r

+ w
 
ω
ICR locations. Section III presents a power model for skid-  =
 ẏICRr  
yICRl −yICRr 
 (6)
steer robots, and Section IV provides a description of the    wr 
 ẏICRl   wl

skid-steer vehicle utilized for testing in this work. Results 
of experimental testing are presented in Section V, and ẋICRv wx
conclusions are given in Section VI. where wN -wx are additive zero-mean Gaussian process
noises. An assumption that the three estimated ICR coordi-
II. ICR KINEMATICS OF SKID-STEER ROBOTS
nates are constant produces the last three rows in the motion
The use of ICR kinematics in motion estimation for skid- model. ICR locations have been shown to remain within
steer vehicles is well presented in [14], [19]. A brief overview small, bounded regions for skid-steer vehicles operating on
is presented here due to the importance of ICR kinematics in hard, flat terrain [14]. The experimental testing presented in
the EKF, presented later in this section, and the power usage this work falls under these assumptions. A detailed presen-
model presented in Section III. The ICR of a rigid body in tation of the EKF algorithm with experimental results for
planar motion is defined as the location at which the relative tracked and wheeled skid-steer robots is given in [19].
velocity of the body to the ICR point is zero. Thus, all other
III. ESTIMATION OF TERRAIN PARAMETERS
points on the rigid body exhibit pure rotation about the ICR
FOR POWER MODELING
[18]. The ICRs of interest in skid-steer kinematics are those
between the spinning tracks or wheels of the vehicle and the Following the work of [14], a framework for estimating
ground plane on which it operates. For simplicity, a tracked the power usage of skid-steer tracked [15] and wheeled [16]
robot is assumed here but the equations are equally valid for robots was presented. These works calculated ICR locations
a wheeled vehicle. and optimal terrain parameters on recorded data, whereas the
Assume a body-fixed frame (X, Y, Z) with the X axis method presented here focuses on the identification of ICR
pointing forward, Y axis to the right, and Z axis downward. locations and terrain parameters during vehicle operation.
The relationship between vehicle movement, ICR locations, This section will present the ICR power model of [16] and
and track speed inputs is adapt it to enable estimation of model coefficients.
The first type of power loss to be modeled is that of the
Vxr yICRl − Vxl yICRr dynamic friction between the tracks or wheels on the ground
vx = (1)
yICRl − yICRr as a result of slippage when turning a robot. The slip velocity
(V l − Vxr )xICRv of an element on the track can in fact be calculated once
vy = x (2) the ICR locations are known. A diagram of the right side
yICRl − yICRr
Vxl − Vxr track of a skid-steer robot is shown in Fig. 1. The vector ~a
wz = − . (3) describes the position of the point of interest on the track
yICRl − yICRr
relative to the geometric center of the robot, and the vector
where vx is the velocity of the vehicle in the X direction, C~ r describes the position of the ICR for the right hand track
vy is the velocity of the vehicle in the Y direction, and wz relative to the geometric center of the robot. The subtraction
is the angular velocity of the vehicle around the Z axis. of C ~ r from ~a yields the position of the point of interest on
The left and right track speeds are Vxl and Vxr , respectively. the track relative to the ICR location. The track is in pure

2787

Authorized licensed use limited to: Lodz University of Technology. Downloaded on March 15,2021 at 14:57:57 UTC from IEEE Xplore. Restrictions apply.
rotation about the ICR which allows the velocity of the track resistances from drive belts, gears, and drive tracks them-
point of interest relative to ground, ~va to be calculated as selves. For low speeds, this power has been shown to be
proportional to the speeds of the drive wheels or tracks as
~v (a) = ω ~ r,l ).
~ z × (~a − C (7)
PR = G(|Vxr | + |Vxl |) (13)
The power lost to dynamic friction, PS can be modeled
as Z where G denotes the proportional resistance constant [15],
[13].
PSl,r = F~ (a) · ~v (a)dA (8)
Al,r Equations (12) and (13) provide a power model of the
skid-steer vehicle, and if the ICR locations, normal force,
where dA is the differential of the surface integral, F~ (a) is angular rate, and track or wheel speeds of the vehicle are
the friction force resisting slipping, and the superscript l, r known, the power model is linear in the parameters µ and
denotes the left or right track, respectively. G as
Assuming a coulomb friction model, the friction force at 4
the point of interest, F~a , is
X
PM = µ|ωz | ~ r,l k+G(|V r | + |V l |)
p(n)k~a − C (14)
x x
~va n=1
F~a = −µp(a) (9)
k~va k where PM is the total power output by the drive motors.
A recursive least squares (RLS) algorithm is implemented
where µ is the coefficient of friction and p(a) is the normal
to estimate the values of µ and G during vehicle operation
force between the track and ground at the point of interest.
using the standard equations [20]
The total power loss is found by substituting (7) and (9)
T T
into (8), yielding Ksk = Psk−1 Hsk (Hsk Psk−1 Hsk + Rs )−1 (15)
qk = qk−1 + Ksk (Pmeas − Hs qk−1 ) (16)
Z
PSl,r = µp(a)k~va kdA (10)
Al,r Z

= µ|ωz | ~ r,l kdA


p(a)k~a − C (11) Psk = (I2x2 − Ksk Hsk )Psk−1 (I2x2 − Ksk Hsk )T
T
Al,r
+ Ksk Rs Ksk (17)
where µ is assumed to be constant over the entire track
surface. where Ks is the gain, Ps is the error covariance matrix,
For a four-wheeled skid-steer robot, assuming contact Hs is the measurement equation vector, Rs is the measure-
between the vehicle and the ground is assumed to occur at ment error covariance matrix, q is the vector of estimated
only the center of each wheel, the integral of (11) simplifies coefficients, and Pmeas is the measurement of power usage
to a sum as from sensors on the robot. The Hs vector is populated at
each time step using current estimates of ICR location and
4
X
~ r,l k measurements of vehicle angular rate as
PS = µ|ωz | p(n)k~a − C (12) " ! #
n=1 4
X
Hs = ~ r,l k |Vxr | + |Vxl | .

|ωz | p(n)k~a − C
where n is the wheel number and Cr or Cl is selected as n=1
appropriate for the right or left wheel of interest. (18)
The second power loss to be modeled includes rolling
resistance due to wheel and soil deformation and internal IV. DESCRIPTION OF SKID-STEER TEST
PLATFORM S
Two skid steer vehicles, one using wheels and the other
tracks, were used during experimental testing. The RMP400
is a wheeled robot developed by Segway Robotics with
a width of 0.61 m and a mass of 118 kg. The tracked
skid-steer vehicle was developed at The Pennsylvania State
University and has a width of 0.52 m and a mass of
37 kg. An embedded PC computer running the open-source
Robot Operating System (ROS) [21] software under Linux is
utilized to interface with motor drivers on each vehicle and
additional sensors.
Both vehicles utilize a Hemisphere A325 real time kine-
matic (RTK) GPS system with an accuracy of 1 cm RMS.
An Xsens MTi-G attitude heading reference system (AHRS)
provides measurements of vehicle heading, pitch, and roll at
Fig. 1: Diagram of right side track showing differential 100 Hz. Both vehicles are equipped to measure wheel speed
element of track contact area. and motor current among other system parameters.

2788

Authorized licensed use limited to: Lodz University of Technology. Downloaded on March 15,2021 at 14:57:57 UTC from IEEE Xplore. Restrictions apply.
The mass distribution of the RMP400 is very symmetrical,
resulting in a center of mass at the geometric center of the
300 Measurement (a)
vehicle. The normal force is assumed to be evenly distributed

Power (W)
Model
among the four wheels, giving a force of 290 N at each 200
wheel. The normal force for the tracked vehicle is assumed
100
to be evenly distributed among the eight bogie wheels that
contact the ground, This results in a force of 92 N at each 0
0 20 40 60 80 100 120 140
bogie wheel. Time (s)

and Modeled Power (W)


Difference in Measured
V. DESCRIPTION OF TESTING AND RESULTS (b)
50
This section presents the results of field testing the RLS
terrain parameter estimation algorithm of Section III. The 0
terrain parameter estimation algorithm and the ICR EKF
−50
algorithm were implemented to run concurrently on the
robots while driven manually over flat grass, asphalt, and 0 20 40 60 80 100 120 140
frozen snow terrains. Time (s)

Fig. 3: Comparison of measured and modeled power usage


A. Skid-Steer Power Modeling on Grass Terrain for wheeled robot on grass using values of µ and G estimated
The wheeled vehicle was first driven on grass over a during previous run on grass. Plot (a) shows the measured
curved path at 0.4 m/s. Using the output of the ICR EKF and modeled power usage throughout the run, plot (b) shows
and power measurements provided by sensors on the vehicle, the difference between the measured and modeled power
the RLS terrain parameter estimator produced the estimates usage.
of µ and G shown in Fig. 2. The estimated values of µ
and G converged to values of 1.12 and 57.2 N, respectively, integration is shown in Fig. 4. Throughout the run, the mean
for the grass terrain. These values of terrain parameters error between the modeled and measured power usage was
were then utilized to estimate the power usage during a 173 J, and the error at the end of the run was 46 J, or 0.4%.
subsequent run on grass without updates from power usage
measurements on the vehicle. A comparison of the measured
and model power is shown in Fig. 3. The measured and B. Skid-Steer Power Modeling on Asphalt Terrain
modeled power signals are noisy and were filtered with a The preceding analysis was repeated while driving the
3rd order Butterworth filter with a cutoff frequency of 10 Hz wheeled robot on asphalt at 0.4 m/s. Using the output of
to make distinguishing between the two easier in Fig. 3. All the ICR EKF and power measurements provided by sensors
statistics on error and energy calculations were performed on the vehicle, the RLS terrain parameter estimator produced
on unfiltered data. The mean and standard deviation of the estimates of µ and G which converged to values of 1.15 and
error between the measured and model power was 13.3 W 47.7 N, respectively, for the asphalt terrain. These values
and 11.7 W, respectively. of terrain parameters were then utilized to estimate the
The measured and modeled power can easily be integrated
to show energy usage throughout the run. The result of this 10

2
8
1
µ

0
6
Energy (kJ)

−1
0 50 100 150
Time (s) 4
150
Measurement
100 2
Model
G (N)

50
0
0 0 20 40 60 80 100 120 140
0 50 100 150 Time (s)
Time (s)
Fig. 4: Comparison of measured and modeled energy usage
Fig. 2: Estimate of µ and G terrain parameters during for wheeled robot on grass using values of µ and G estimated
wheeled robot run. during previous run on asphalt.

2789

Authorized licensed use limited to: Lodz University of Technology. Downloaded on March 15,2021 at 14:57:57 UTC from IEEE Xplore. Restrictions apply.
power usage during a subsequent run on asphalt without
updates from power usage measurements on the vehicle.
300 Measurement (a)
The measured and modeled power for this run are shown

Power (W)
Model
in Fig. 5. For plotting, the measured and modeled signals 200
were again filtered as described in Section V-A. The modeled
100
power usage matches the measured well with the mean and
standard deviation of the error being 12.2 W and 10.8 W, 0
0 50 100 150
respectively. Time (s)
The measured and modeled energy usage for the run are

and Modeled Power (W)


Difference in Measured
100
shown in Fig. 6. For the asphalt test run, the mean error (b)
between the modeled and measured energy usage was 120 J, 50
and the error at the end of the run was 91 J, or 0.6%.
0
C. Terrain Change Detection
−50
An important aspect of power usage modeling is the 0 50 100 150
detection of terrain changes and adaptation of the model pa- Time (s)
rameters. Terrain change detection was tested experimentally
Fig. 5: Comparison of measured and modeled power usage
by manually driving the tracked skid-steer robot from asphalt
for wheeled robot on asphalt using values of µ and G
to frozen snow. A camera on board the tracked vehicle was
estimated during previous run. Plot (a) shows the measured
used to monitor the color of the terrain directly in front of
and modeled power usage throughout the run, plot (b) shows
the robot. When a large change in terrain color was detected,
the difference between the measured and modeled power
the error covariance estimates of the power model parameters
usage.
were reset to enable new parameters to be learned. The
estimated coeffients for the run are shown in Fig. 7 and
the time of terrain change is labeled at approximately 270 15
seconds into the run. The coefficient of friction µ is fairly
constant for both terrain surfaces with a value near 0.4.
The proportional resistance coefficient G shows a significant
change between the two terrains and rises from a value of
10
92 N on asphalt to 130 N for the frozen snow terrain. This
Energy (kJ)

increase in power usage for the snow terrain is apparent in


the modeled and measured power usage, shown in Fig. 8.
The increased variance in the modeled power usage error is
likely due to variations in the frozen snow terrain. Overall, 5
Measurement
the algorithm shows the ability to learn power modeling
coefficients for varying terrains when informed of a terrain Model
change through other sensors.

VI. CONCLUSIONS 0
50 100 150 200
This paper has presented a method of estimating the move- Time (s)
ment and power usage of a skid-steer robot using parameters Fig. 6: Comparison of measured and modeled energy usage
learned during vehicle operation. An EKF designed for the for wheeled robot on asphalt using values of µ and G
estimation of track or wheel ICR locations was presented, estimated during previous run.
followed by a RLS algorithm for estimating terrain-related
power model coefficients. Continued work in this area will involve utilizing the
Implementation results for the ICR-based power model presented power model for energy efficient path planning
show slightly different terrain parameters for asphalt and and retro-traversal energy use estimation. Also of interest is
grass with a wheeled skid-steer robot. Utilizing power model work on improving the ICR EKF estimates by integrating ad-
coefficients identified on previous runs, power usage esti- ditional sensors such as cameras and LIDAR and continued
mation for both surfaces matched well with a mean error testing of the power model at changing operational speeds.
of 13.3 W during grass terrain testing and 12.2 W during
asphalt terrain testing. Energy usage is within 0.4% for 46 m VII. ACKNOWLEDGMENTS
of travel on grass and within 0.6% over 66 m of travel
on asphalt. Implementation of a terrain detection algorithm This work was supported by the Exploratory and Foun-
through simple image processing showed that the power dational Research Program, administered by the Applied
model coefficients can be identified for varying terrains while Research Laboratory, The Pennsylvania State University. No
the vehicle is in operation. official endorsement should be inferred.

2790

Authorized licensed use limited to: Lodz University of Technology. Downloaded on March 15,2021 at 14:57:57 UTC from IEEE Xplore. Restrictions apply.
executing coverage tasks,” in Proceedings of SPIE, vol. 8387, May
1 2012.
Terrain Change [5] J. Morales, J. L. Martı́nez, A. Mandow, A. J. Garcı́a-Cerezo, J. M.
Gómez-Gabriel, and S. Pedraza, “Power analysis for a skid-steered
tracked mobile robot,” in Proceedings of IEEE 3rd International
0.5
µ

Conference on Mechatronics, 2006, pp. 420–425.


[6] J. Wong, Theory of Ground Vehicles, 4th ed. John Wiley and Sons,
2008.
0
0 100 200 300 400 [7] J. Y. Wong and C. F. Chiang, “A general theory for skid steering of
Time (s) tracked vehicles on firm ground,” Proceedings of the Institution of
Mechanical Engineers, Part D: Journal of Automobile Engineering,
200 vol. 215, no. 3, pp. 343–355, Jan. 2001.
[8] A. T. Le, D. C. Rye, and H. F. Durrant-Whyte, “Estimation of track-
soil interactions for autonomous tracked vehicles,” in Proceedings of
G (N)

the 1997 IEEE International Conference on Robotics and Automation,


100
1997, pp. 1388–1393.
[9] J. Yi, H. Wang, J. Zhang, D. Song, S. Jayasuriya, and J. Liu,
“Kinematic modeling and analysis of skid-steered mobile robots
0 with applications to low-cost inertial-measurement-unit-based motion
0 100 200 300 400 estimation,” IEEE Transactions on Robotics, vol. 25, no. 5, pp. 1087–
Time (s) 1097, 2009.
[10] T. M. Dar and R. G. Longoria, “Slip estimation for small-scale
Fig. 7: Estimate of µ and G terrain parameters during tracked robotic tracked vehicles,” in Proceedings of 2010 American Control
robot run on asphalt and frozen snow. Time of terrain change Conference, 2010, pp. 6816–6821.
is labeled. [11] ——, “Estimating traction coefficients of friction for small-scale
robotic tracked vehicles,” in Proceedings of ASME 2010 Dynamic
Systems and Control Conference, 2010.
[12] S. A. A. Moosavian and A. Kalantari, “Experimental slip estimation
300 for exact kinematics modeling and control of a tracked mobile robot,”
Measurement in 2008 IEEE/RSJ International Conference on Intelligent Robots and
Power (W)

200 Model Systems, Sep. 2008, pp. 95–100.


[13] O. Chuy Jr., E. Collins Jr., W. Yu, and C. Ordonez, “Power modeling
of a skid steered wheeled robotic ground vehicle,” in Proceedings of
100
the 2009 IEEE International Conference on Robotics and Automation,
May 2009, pp. 4118–4123.
0 [14] J. L. Martı́nez, A. Mandow, J. Morales, S. Pedraza, and A. J. Garcı́a-
0 100 200 300 400 Cerezo, “Approximating kinematics for tracked mobile robots,” The
Time (s) International Journal of Robotics Research, vol. 24, no. 10, pp. 867–
and Modeled Power (W)
Difference in Measured

100 878, Oct. 2005.


Terrain Change [15] J. Morales, J. L. Martı́nez, A. Mandow, A. J. Garcı́a-Cerezo, and
S. Pedraza, “Power consumption modeling of skid-steer tracked mobile
robots on rigid terrain,” IEEE Transactions on Robotics, vol. 25, no. 5,
0 pp. 1098–1108, 2009.
[16] J. Morales, J. L. Martı́nez, A. Mandow, A. Pequeno-Boter, and
A. J. Garcı́a-Cerezo, “Simplified power consumption modeling and
−100 identification for wheeled skid-steer robotic vehicles on hard hori-
0 100 200 300 400 zontal ground,” in Proceedings of the 2010 IEEE/RSJ International
Time (s) Conference on Intelligent Robots and Systems, 2010, pp. 4769–4774.
[17] T. Guo and H. Peng, “A simplified skid - steering model for torque
Fig. 8: Comparison of measured and modeled power usage and power analysis of tracked small unmanned ground vehicles *,” in
for tracked robot on asphalt and frozen snow. Plot (a) shows Proceedings of 2013 American Control Conference, 2013, pp. 1106–
1111.
the measured and modeled power usage throughout the run, [18] J. J. Uicker, Jr, G. R. Pennock, and J. E. Shigley, Theory of Machines
plot (b) shows the difference between the measured and and Mechanisms, 3rd ed. New York: Oxford University Press, 2003.
modeled power usage. Time of terrain change is labeled. [19] J. Pentzer, S. Brennan, and K. Reichard, “Model-based prediction of
skid-steer robot kinematics using on-line estimation of track instanta-
neous centers of rotation,” Journal of Field Robotics, vol. in press.
R EFERENCES [20] D. Simon, Optimal State Estimation: Kalman, H Infinity, and Nonlin-
ear Approaches, 2006.
[1] Joint IDE Defeat Organization, “Counter-improvised explosive device [21] M. Quigley, B. Gerkey, K. Conley, J. Faust, T. Foote, J. Leibs,
strategic plan,” 2012. [Online]. Available: www.jieddo.mil E. Berger, R. Wheeler, and A. Ng, “ROS : an open-source robot operat-
[2] A. Jacoff, H.-M. Huang, A. Virts, A. Downs, and R. Sheh, “Emergency ing system,” in Proceedings of 2009 IEEE International Conference on
response robot evaluation exercise,” in Proceedings of Workshop on Robotics and Automation Workshop on Open Source Robotics, 2009.
Performance Metrics for Intelligent Systems, no. c, 2012, pp. 145–154. [22] G. Anousaki and K. J. Kyriakopoulos, “A dead-reckoning scheme for
[3] M. G. Bekker, Theory of Land Locomotion. Ann Arbor: University skid-steered vehicles in outdoor environments,” in Proceedings of 2004
of Michigan Press, 1962. IEEE International Conference on Robotics and Automation, 2004, pp.
[4] J. Broderick, D. Tilbury, and E. Atkins, “Energy usage for UGVs 580–585.

2791

Authorized licensed use limited to: Lodz University of Technology. Downloaded on March 15,2021 at 14:57:57 UTC from IEEE Xplore. Restrictions apply.

You might also like