A Comparison between Different Encoding Strategies for Snake-Like Robot Controllers

In this paper, we present the results of the tests we have performed with different encoding strategies for evolving controllers for a snake-like robot. This study is aimed at finding the best encoding for on-line learning of basic skills, such as locomotion (both free and directed to an objective) and obstacle avoidance. The snake moves in a virtual world, which realistically simulates all the physical conditions of the real world. This is the first step of our research on on-line, embedded and open-ended evolution of robot controllers, where robots have to learn how to survive during their lifetime, and occasionally mate with other robots. A simple (1+1) evolutionary strategy has been adopted for lifetime learning. The results of the tests have shown that the best results, tested on the locomotion skills, is the ’He1Sig’ controller, that uses a different set of parameters for each segment of the snake but only one mutation rate, common to all parameters, that is encoded in the chromosome and therefore undergoes evolution itself.