Gradient Informed Proximal Policy Optimization

Sanghyun Son, Laura Yu Zheng, Ryan Sullivan, Yi-Ling Qiao, Ming Lin

Keywords: Reinforcement Learning, Analytic Gradient-Based Policy Learning, Proximal Policy Optimization, Differentiable Programming

TL;DR: In this paper, we introduce a novel policy learning approach that leverages analytical gradients, which may exhibit high variance or bias, in order to improve the performance of the Proximal Policy Optimization algorithm.

Abstract: We introduce a novel policy learning method that integrates analytical gradients from differentiable environments with the Proximal Policy Optimization (PPO) algorithm. To incorporate analytical gradients into the PPO framework, we introduce the concept of an α-policy that stands as a locally superior policy. By adaptively modifying the α value, we can effectively manage the influence of analytical policy gradients during learning. To this end, we suggest metrics for assessing the variance and bias of analytical gradients, reducing dependence on these gradients when high variance or bias is detected. Our proposed approach outperforms baseline algorithms in various scenarios, such as function optimization, physics simulations, and traffic control environments. Our code can be found online: https://github.com/SonSang/gippo.

Submission Number: 5513