Bayesian Optimization with Automatic Prior Selection for Data-Efficient Direct Policy Search

TL;DR

This paper introduces MLEI, a new acquisition function for Bayesian optimization that automatically selects the most relevant prior, improving data-efficient policy learning for robots in transfer and damage scenarios.

Contribution

The paper proposes MLEI, a novel acquisition function that combines prior likelihood with expected improvement to automatically select the best prior in Bayesian optimization.

Findings

MLEI effectively identifies relevant priors in transfer learning tasks.

The method accelerates learning in robot damage and transfer scenarios.

Results demonstrate improved data efficiency over baseline methods.

Abstract

One of the most interesting features of Bayesian optimization for direct policy search is that it can leverage priors (e.g., from simulation or from previous tasks) to accelerate learning on a robot. In this paper, we are interested in situations for which several priors exist but we do not know in advance which one fits best the current situation. We tackle this problem by introducing a novel acquisition function, called Most Likely Expected Improvement (MLEI), that combines the likelihood of the priors and the expected improvement. We evaluate this new acquisition function on a transfer learning task for a 5-DOF planar arm and on a possibly damaged, 6-legged robot that has to learn to walk on flat ground and on stairs, with priors corresponding to different stairs and different kinds of damages. Our results show that MLEI effectively identifies and exploits the priors, even when there is no obvious match between the current situations and the priors.