KnowledgeBased Development of Planetary Rovers

1
Knowledge-Based Development of Planetary Rovers

• CPE 481 KB Nugget
• Jay Holcomb
• Spring 2003 TT 300-430

2
Rover Goals

• Must be able to traverse challenging terrain
  with
• not exposing rover to high risks
• little human help

3
Sensory Hardware

• CPU
• Ability to process camera and other sensory data
• Cameras
• Triple pair of stereo cameras giving 180
  viewable arc for up to 5 meters
• Provide data to determine best traversable
  direction using Fuzzy Logic

4
Software

• Fuzzy Logic
• Roughness
• Slope
• Discontinuity
• Terrain Hardness
• Overall Traversability

5
Fuzzy Logic

• Roughness
• Horizon-Line Extraction
• Rocks identified from ground
• Number of pixels determines rock size
• Average distance between rocks is calculated

6
Fuzzy Logic

• Slope
• Cartesian (x,y,z) found examining points of
  largest rock along the horizon plane
• An average slope is then calculated.

7
Fuzzy Logic

• Discontinuity
• Horizon Difference
• Difference in multiple horizontal planes alerts
  rover of a drop off
• Difference in planes is set as a magnitude and
  entered as a Fuzzy Logic variable

8
Fuzzy Logic

• Terrain Hardness
• How well rover will have traction
• Image Block Extraction
• Take random pictures of ground
• Data Dimensionality Reduction
• Take geometric features from pictures
• Neural Network Classification
• Link geometric features to known terrain
  geometric feature database

9
Fuzzy Logic

• Traversability Index
• Roughness, Slope, Discontinuity, and Hardness
  entered into another Fuzzy Logic equation for
  movement difficulty.
• Low Dangerous
• High Safe

10
The Human Factor

• Rover trained by Human Expert
• Expert used as a supervisor while rover learns
• Fine-tuning done by modifying
• membership functions
• rule-base

11
Trial and Error

• Design Optimization
• Procedure (DOP)
• Compares rover performance to the Experts
  response
• Simplex Optimization
• Adjusts the fuzzy system towards human performance

12
Ready for the Real World

• Performance Index (PI)
• Error between rover and Expert
• When reaches a steady and non-decreasing value
  the rover is then placed on real-time terrain.

13

14
References

• 1) A. Howard, E. Tunstel, D. Edwards, A. Carlson,
  Enhancing Fuzzy Robot Navigation Systems by
  Mimicking Human Visual Perception of Natural
  Terrain Traversability, Joint 9th IFSA World
  Congress and 20th NAFIPS International
  Conference, Vancouver, Canada, July 2001.
• 2) A. Howard, H. Seraji, A Real-Time Autonomous
  Rover Navigation System, World Automation
  Congress, June 2000.
• 3) http//telerobotics.jpl.nasa.gov/
• 4) http//mars.jpl.nasa.gov/