Evolving locomotion gaits for quadruped walking robots

Abstract

Purpose This paper presents an evolutionary algorithm (EA) for Sony legged robots to learn good walking behaviours with little or no interaction with the designers. Once the learning method is put into place, the module can learn through its interaction with the real world. Design/methodology/approach An EA for developing locomotion gaits of quadruped walking robots is presented in this paper. It is based on a hybrid approach that changes the probability of genetic operators in respect to the performance of the operator's offspring. Findings The mutating and combination behaviours of the genetic algorithms allow the process to develop a useful behaviour over time. The resulting gait from this training proved to be a better solution than the non‐interference training for movements over all types of surfaces, pointing to a local optima being discovered in the non‐environmental interference situation. Research limitations/implications The behaviour of these algorithms is stochastic so that they may potentially present different solutions in different runs of the same algorithm. The mechanism described here has several features that should be noted. It allows rapid parameterisation of operator probabilities across the range of potential genetic algorithms and operator set. It is tailored to a steady state reproduction scheme. It would not be literally applicable to problems with noisy evaluation functions. Originality/value Provides novel application of genetic algorithms to a potentially practical application area.