In this thesis, we explore an alternative approach to producing a meaningful hexapod gait. In particular, we successfully distribute control of the hexapod to the legs themselves, and allow the legs to cooperate in creating an efficient gait by communicating their status to one another. This is done by devising a neural network controller for each of the legs that all receive inputs from the sensors from the other legs, and evolving the thresholds and connection weights of these neural networks with six genetic algorithms (one algorithm for each of the six hexapod legs,) simultaneously evolving both the individual leg cycles and the leg coordination all at once. Thus, there is no central controller, and the legs all learn to create a hexapod gait with swarm intelligence. This results in emergent behaviors of the individual legs, as they evolve with the purpose of improving the performance of the entire hexapod.
Angliss, Cameron, "Using Multi-Agent Learning to Achieve Emergent Decentralized Hexapod Gait" (2022). Computer Science Honors Papers. 12.
The views expressed in this paper are solely those of the author.