Near Optimal Policy Optimization via REPS
This work addresses a theoretical gap for a widely used reinforcement learning algorithm, offering guarantees that could improve reliability in robotic and simulated domains.
The paper provides the first performance guarantees and convergence rates for the Relative Entropy Policy Search (REPS) algorithm when using first-order optimization methods, showing how near-optimality in the objective translates to near-optimal policies in both exact and stochastic gradient settings.
Since its introduction a decade ago, \emph{relative entropy policy search} (REPS) has demonstrated successful policy learning on a number of simulated and real-world robotic domains, not to mention providing algorithmic components used by many recently proposed reinforcement learning (RL) algorithms. While REPS is commonly known in the community, there exist no guarantees on its performance when using stochastic and gradient-based solvers. In this paper we aim to fill this gap by providing guarantees and convergence rates for the sub-optimality of a policy learned using first-order optimization methods applied to the REPS objective. We first consider the setting in which we are given access to exact gradients and demonstrate how near-optimality of the objective translates to near-optimality of the policy. We then consider the practical setting of stochastic gradients, and introduce a technique that uses \emph{generative} access to the underlying Markov decision process to compute parameter updates that maintain favorable convergence to the optimal regularized policy.