Stephen Williams and Ayanna M. Howard

1
Evaluation of Visual Navigation Methods for Lunar
Polar Rovers in Analogous Environments

• Stephen Williams and Ayanna M. Howard
• Human-Automation Systems Lab
• Georgia Institute of Technology

2
Motivation

• Recent lunar missions have focused on the search
  for water ice near the poles
• LCROSS
• Clementine
• Lunar Prospector
• On-orbit sensing still requires sensor
  measurement validation
• The only direct, unambiguous way is through
  ground-based data
• However, there has never been a landed mission to
  this region

3
Motivation

• Earth-centric satellite sensing suffers from
  similar issues
• Surface data validation even more critical due to
  complex atmospheric effects
• Glacial regions are of particular importance
• More sensitive to climate change mechanisms
• Poorly modeled by other observed surfaces
• Remote, harsh climate leads to few validation
  trials

4
SnoMote Project

• Fixed weather stations do exist in Greenland and
  Antarctica
• Sparse coverage of glacial surface
• Scientists interested in the ability to collect
  dense measurements in targeted locations
• This project focused on developing enabling
  technologies for a glacial mobile weather station

5
SnoMote Sensor Node
6
Mendenhall Glacier Tests
7
Environment

• Low contrast
• Slope-based hazards
• Exposed mountain peaks
• Crevasse
• Blue ice

8
Multi-agent Research

• Distributed Task Allocation (Antidio Viguria)
• Fault tolerant
• Closer to the global optimal than standard
  algorithms
• Embedded Graph Grammars (Brian Smith)
• Enables nodes to self-assemble and maintain
  desired network topology
• Accounts for the dynamics of the platform

9
Visual Navigation

• Terrain Assessment and Localization
• Low contrast, snow-covered terrain
• Methods for testing and evaluating through
  simulation

10
Glacial Terrain Assessment

• Estimate the directionality of small-scale
  surface features within a small region
• Based on methods used for Fingerprint Analysis
• Determines the Least Squares Estimate of the
  dominate 2-D Fourier Spectrum direction

11
Glacial Terrain Assessment
12
Glacial Visual SLAM

• Standard Visual SLAM system
• Major Challenge Feature Extraction
• Region Extraction
• Contrast Enhancement
• SIFT Features

13
Glacial Visual SLAM
14
Evaluation Using Simulation

• Ground truth unknown for real environment
• Local scale terrain topology is unavailable
• Commodity GPS has significant uncertainty
• Simulation system solves these issues
• Visual quality of simulation impacts the results
  of visual algorithms

15
Simulation Design

• Presented work based on Gazebo open source
  simulation system
• Real surface topologies used
• SRTM data for terrestrial sites
• LRO data for lunar sites
• Satellite imaging used to paint the terrain

16
Simulation Design

• Photo-realistic background applied using a
  skybox
• Background images created from panorama of test
  site
• Local scale texture blended with main terrain
  coloration
• High frequency components extracted from
  photography of test site

17
Simulation Design
18
Assessing Simulation Quality

• Easy to perform a qualitative comparison
• That looks good
• A method is needed to compare qualitatively

19
Assessing Simulation Quality

• Any visual algorithm may be applied to images
  from either simulation or the real terrain
• A performance metric can be used to evaluate the
  effectiveness a specific algorithm
• If the difference in performance results are not
  statistically significant, then the simulation
  may be viewed as sufficient
• Must be evaluated on each algorithm-metric pair

20
Assessing Simulation Quality
Region Extraction
Feature Count
21
Conclusions

• Real-time vision-based processing techniques were
  presented
• Implemented to cope with image characteristics of
  glacial terrain
• Time and expense of field deployments in remote
  regions prevent frequent trials
• Ground truth data difficult to obtain in real
  environments
• Visually faithful simulation system developed to
  test and validate vision-based algorithms
• Analysis conducted to assess the visual quality
  of the simulation

22
Future Work

• Further testing and validation of simulation
  assessment method
• Development of a simulated science sensor to
  enable testing of science-driven control
  behaviors
• Spatial and temporal data interpolation
• Include noise models
• Opportunistic SnoMote testing in Anchorage,
  Alaska

23
Acknowledgments

• NASA Earth Science Technology Office provided
  funding for this work under the Applied
  Information Systems Technology Program
• Dr. Magnus Egerstedt, Georgia Institute of
  Technology, provided his experience in
  multi-agent formations
• Dr. Matt Heavner, Associate Professor of Physics,
  University of Alaska Southeast, provided his
  expertise in glacial field work

24
References

• B. P. Gerkey, R. T. Vaughan, and A. Howard, The
  Player/Stage projectTools for Multi-Robot and
  distributed sensor systems, in International
  Conference on Advanced Robotics, ICAR, Coimbra,
  Portugal, July 2003, pp. 317323.
• L. Hong, Y. Wan, and A. Jain, Fingerprint image
  enhancement Algorithm and performance
  evaluation, IEEE Transactions on Pattern
  Analysis and Machine Intelligence, vol. 20, no.
  8, pp. 777789, 1998.
• A. M. Reza, Realization of the contrast limited
  adaptive histogram equalization (CLAHE) for
  Real-Time image enhancement, The Journal of VLSI
  Signal Processing, vol. 38, no. 1, pp. 3544,
  2004.
• S. Williams and A. M. Howard, A single camera
  terrain slope estimation technique for natural
  arctic environments, in IEEE International
  Conference on Robotics and Automation, ICRA,
  Pasadena, CA, May 2008, pp. 27292734.
• , Developing monocular visual pose estimation
  for arctic environments, Journal of Field
  Robotics, vol. 27, no. 2, pp. 145157, 2009.
• , Towards visual arctic terrain assessment,
  in International Conference on Field and Service
  Robotics, FSR, Cambridge, MA, July 2009
• S. Williams, S. Remy, and A. M. Howard, , in
  American Institute of Aeronautics and
  Astronautics Conference Infotech _at_ Aerospace,
  Atlanta, GA, April 2010,