Evolutionary algorithm for the promotion of evolvable hardware

Abstract

Evolvable hardware offers much for the future of complex system design. Evolutionary techniques not only give the potential of a larger solution space, but also when implemented on hardware allow system designs to adapt to changes in the environment, including failures in system components. An evolutionary algorithm is developed that is solely based on a fitness-dependent mutation rate. One major objective for the authors is to produce an evolutionary system that is easily implementable on custom-off-the-shelf hardware. As an example of the system, an FPGA-based controller for a mobile robot is presented. The controller is made up of look up tables which perform the mapping from sensor data to actuator (commands to the motors) and it is evolved using the developed evolutionary algorithm. Experimental results obtained on a Khepera robot show that the method can successfully evolve a robot controller for autonomous navigation to avoid collisions in an unknown/changing environment even if sensor faults occur prior to the evolution or after a successful member of a population has been evolved.