Gait Parameter Tuning Using Bayesian Optimization for an Alligator Inspired Amphibious Robot

Abstract

This paper reports a sample-efficient Bayesian optimization approach for tuning the locomotion parameters of an in-house developed twelve degrees of freedom alligator-inspired amphibious robot. An optimization framework is used wherein the objective is to maximize the mean robot speed obtained via physical experiments performed on terrains with varying friction and inclinations and in the aquatic environment for swimming locomotion. We obtained an improvement in the mean robot speed by a factor of up to 6.38 using the developed approach over randomly generated locomotion parameters in 15 iterations.