Model-Based Reinforcement Learning via Meta-Policy Optimization

TL;DR

This paper introduces MB-MPO, a model-based reinforcement learning method that uses meta-policy optimization with an ensemble of models to achieve data efficiency and robustness, matching the performance of model-free methods.

Contribution

MB-MPO is a novel approach that reduces reliance on accurate dynamics models by meta-learning a policy that adapts quickly to model discrepancies.

Findings

MB-MPO is more robust to model imperfections than previous methods.

It achieves asymptotic performance comparable to model-free algorithms.

Requires significantly less experience to reach high performance.

Abstract

Model-based reinforcement learning approaches carry the promise of being data efficient. However, due to challenges in learning dynamics models that sufficiently match the real-world dynamics, they struggle to achieve the same asymptotic performance as model-free methods. We propose Model-Based Meta-Policy-Optimization (MB-MPO), an approach that foregoes the strong reliance on accurate learned dynamics models. Using an ensemble of learned dynamic models, MB-MPO meta-learns a policy that can quickly adapt to any model in the ensemble with one policy gradient step. This steers the meta-policy towards internalizing consistent dynamics predictions among the ensemble while shifting the burden of behaving optimally w.r.t. the model discrepancies towards the adaptation step. Our experiments show that MB-MPO is more robust to model imperfections than previous model-based approaches. Finally, we demonstrate that our approach is able to match the asymptotic performance of model-free methods while requiring significantly less experience.