Automatic gait optimization with Gaussian process regression

摘要: Gait optimization is a basic yet challenging problem for both quadrupedal and bipedal robots. Although techniques automating the process exist, most involve local function procedures that suffer from three key drawbacks. Local are naturally plagued by optima, make no use of expensive gait evaluations once step taken, do not explicitly model noise in evaluation. These drawbacks increase need large number evaluations, making slow, data inefficient, manually intensive. We present Bayesian approach based on Gaussian regression addresses all It uses global search strategy posterior inferred individual noisy evaluations. demonstrate technique quadruped robot, using it to optimize two different criteria: speed smoothness. show cases our requires dramatically fewer than state-of-the-art gradient approaches.