Co-evolving controller and sensing abilities in a simulated Mars Rover explorer
21st May, 2009 Trondheim, Norway
Introduction: Mars Robots
Fundamental requirement of all planetary robots is a high degree of autonomy and safety It takes 4.3 to 21 minutes for radio
signal to reach Mars from Earth Planetary robotics mission are extremely
expensive, e.g. NASA spent $800 million to build and launch Spirit and Opportunity rovers to Mars
Introduction: Mars Robots
Spirit and Opportunity rovers travel slower than 1cm/s when in autonomous mode
Obstacle avoidance is based on stereo cameras producing 3D representation of surrounding environment This solution is effective but rather slow as lot of
processing needs to be done
Rover model: predecessors
Rover model: overview
Rocker-bogie suspension Open dynamics engine Open GL
MSL rover 2.9 x 2.7 x 2.2m 775kg
Rover model: closer look
Rover model: “brain” & “senses”
Rover model: threshold
40cm
Rover model: threshold
Rover model: sensors in action
Experimental setup:
Elitism Fitness function:
Genotype
...
connection weights threshold
w1 w2 w3 w4 w5 w6 w7 w8 w9 w10 w11 w80 t
Environment
Results
Simulations results show that the robot, at the end of the evolutionary process, is able to avoid rocks, holes and steep slopes based purely on the information provided by the infrared sensors
Co-evolving controller and sensing ability Fitness starts to increase once the threshold
value settles
Co-evolving controller and sensing ability Fitness starts to increase after sudden change
of threshold value
Co-evolving controller and sensing ability Early presence of good obstacle avoidance
behaviour with unsuitable threshold
Tests: evaluating robustness
Best rover from last generation was tested on two other terrains. Terrain with more obstacles Terrain with increased surface
roughness
Tests: evaluating robustness
Exploration ability depends on fitness and on type of the terrain
Current research on active vision
pan
tilt
zoom
speed
steering
pan
tilt
speed
steering
visual neurons
Preliminary results
Conclusion
Evolutionary robotics applied in space research domain
A model of a Mars rover robot autonomously avoiding obstacles in different environments
Different co-evolutionary scenarios Tests of robustness of the evolved
controllers Current research on active vision