Learning Stochastic Binary Tasks using Bayesian Optimization with Shared Task Knowledge

Abstract

Robotic systems often have tunable parameters which can aect performance; Bayesian optimization methods provide for ecient parameter optimization, reducing required tests on the robot. This paper addresses Bayesian optimization in the setting where performance is only observed through a stochastic binary outcome { success or failure. We dene the stochastic binary optimization problem, present a Bayesian framework using Gaussian processes for classication, adapt the existing expected improvement metric for the binary case, and benchmark its performance. We also exploit problem structure and task similarity to generate principled task priors allowing ecient search for dicult tasks. This method is used to create an adaptive policy for climbing over obstacles of varying heights.