ROBOTRONIKA - hypermatic:automagic
19 - 23 June 1998 Museumsquartier, Vienna/Austria
| University of Sussex/Centre for Computational Neuroscience and Robotics |
|
Maggie Maggie is an eight-legged insect-like walking robot Oct 1-a model built by Applied AI systems of Canada (see http://fox.nstn.ca/~aai/oct.htm for their current similar Oct 1-b). Body around 50 cm long, width with legs at full extension about 40 cm, height when walking is 16 cm. Each of the 8 legs has 2 servo motors (forward-back, up-down). Sensors include whiskers, bumpers, 4 infra-red sensors and various ambient light sensors. There is an onboard 68000-family microcontroller to which control systems can be downloaded. It can then operate completely autonomously for 20 to 30 minutes using its onboard batteries. The tricky bit is: how do you design the control systems which act as an artificial `nervous system', connecting motors and sensors so as to produce sensible behaviour? This robot is used at the CCNR (Centre for Computational Neuroscience and Robotics) at the University of Sussex to demonstrate Evolutionary Robotics, the use of artificial Darwinian evolution to design `robot brains'. It will be shown working with a control system (an artificial neural network with 48 `neurons') evolved by Nick Jakobi. Evolution took around 3500 generations, 14 hours in simulation on a Sun workstation, and then works on the real robot. It will walk forward when the way is clear, turning away or backing away from any obstacles, yet this `nervous system' was designed by evolution, not by a human, so we are not quite sure how it works. But in fact on testing it, it works much better than the comparable control system which was designed by a human!
|