Constrained Policy Optimization

TL;DR

This paper introduces Constrained Policy Optimization (CPO), a novel reinforcement learning algorithm that ensures safety constraints are satisfied during training, enabling high-dimensional control with safety guarantees.

Contribution

CPO is the first policy search method providing theoretical guarantees for constraint satisfaction throughout training in reinforcement learning.

Findings

Successfully trained neural policies for robot locomotion with safety constraints.

Provided theoretical bounds relating policy performance to divergence measures.

Demonstrated effectiveness on simulated safety-critical control tasks.

Abstract

For many applications of reinforcement learning it can be more convenient to specify both a reward function and constraints, rather than trying to design behavior through the reward function. For example, systems that physically interact with or around humans should satisfy safety constraints. Recent advances in policy search algorithms (Mnih et al., 2016, Schulman et al., 2015, Lillicrap et al., 2016, Levine et al., 2016) have enabled new capabilities in high-dimensional control, but do not consider the constrained setting. We propose Constrained Policy Optimization (CPO), the first general-purpose policy search algorithm for constrained reinforcement learning with guarantees for near-constraint satisfaction at each iteration. Our method allows us to train neural network policies for high-dimensional control while making guarantees about policy behavior all throughout training. Our guarantees are based on a new theoretical result, which is of independent interest: we prove a bound relating the expected returns of two policies to an average divergence between them. We demonstrate the effectiveness of our approach on simulated robot locomotion tasks where the agent must satisfy constraints motivated by safety.