A comparison of three evolved controllers used for robotic navigation

Abstract

This paper compares three evolved controllers including, an evolvable hardware controller, an artificial neural network and a lookup table. The comparison made between these controllers looks at relative evolutionary efficiency, controller performance and scalability. The controllers were evolved for three navigational behaviours including light following, obstacle avoidance, and the combined behaviours of light following while avoiding obstacles. Both monolithic and subsumption techniques were used to evolve the combined behaviours to evaluate scalability. It was found that all three evolved controllers performed the assigned tasks equally well. The evolutionary efficiency and scalability of the evolvable hardware and artificial neural network were similar, whereas the lookup table had an acceptable result but was subjective to scalability. The virtual-FPGA can be implemented in a fault tolerant system using a hybrid FGPAs with a hard-core processor for continuous evolution.