EVOLUTIONARY ROBOTICS AND THE RADICAL ENVELOPE-OF-NOISE HYPOTHESIS

For several years now, various researchers have endeavored to apply ar tificial evolution to the automatic design of control systems for real robots. One of the major challenges they face concerns the question o f how to assess the fitness of evolving controllers when each evolutio nary run typically involves hundreds of thousands of such assessments. This article outlines new ways of thinking about and building simulat ions upon which such assessments can be performed. it puts forward suf ficient conditions for the successful transfer of evolved controllers from simulation to reality and develops a potential methodology for bu ilding simulations in which evolving controllers are forced To satisfy these conditions if they are to be reliably fit. it is hypothesized t hat as long as simulations are built according to this methodology, it does not matter how inaccurate or incomplete they are: Controllers th at have evolved to be reliably fit in simulation still will transfer i nto reality. Two sets of experiments are reported both of which involv e minimal look-up fable-based simulations built according to these gui delines. in the first set, controllers were evolved that allowed a Khe pera robot to perform a simple memory task in the real world. in the s econd set, controllers were evolved for the Sussex University gantry r obot that were able so distinguish visually a triangle from a square, under extremely noisy real-world conditions, and to steer the robot to ward the triangle. in both cases, controllers that were reliably fit i n simulation displayed extremely robust behavior when downloaded into reality.