Differentiable Trust Region Layers for Deep Reinforcement Learning

TL;DR

This paper introduces differentiable trust region layers for deep reinforcement learning that enforce trust regions via closed-form projections, improving robustness and ease of implementation over existing approximation-based methods.

Contribution

It proposes novel differentiable neural network layers for trust region enforcement in deep RL, formalizing state-specific trust regions with closed-form solutions based on various divergence measures.

Findings

Achieves comparable or better results than existing methods.

Layers are nearly agnostic to implementation choices.

Formalizes trust regions for each state individually.

Abstract

Trust region methods are a popular tool in reinforcement learning as they yield robust policy updates in continuous and discrete action spaces. However, enforcing such trust regions in deep reinforcement learning is difficult. Hence, many approaches, such as Trust Region Policy Optimization (TRPO) and Proximal Policy Optimization (PPO), are based on approximations. Due to those approximations, they violate the constraints or fail to find the optimal solution within the trust region. Moreover, they are difficult to implement, often lack sufficient exploration, and have been shown to depend on seemingly unrelated implementation choices. In this work, we propose differentiable neural network layers to enforce trust regions for deep Gaussian policies via closed-form projections. Unlike existing methods, those layers formalize trust regions for each state individually and can complement existing reinforcement learning algorithms. We derive trust region projections based on the Kullback-Leibler divergence, the Wasserstein L2 distance, and the Frobenius norm for Gaussian distributions. We empirically demonstrate that those projection layers achieve similar or better results than existing methods while being almost agnostic to specific implementation choices. The code is available at https://git.io/Jthb0.