Gradient Informed Proximal Policy Optimization

Conference paper

Sanghyun Son, Laura Yu Zheng, Ryan Sullivan, Yi-Ling Qiao, Ming Lin
NeurIPS, 2023

Arxiv Github

Cite

APA Click to copy
Son, S., Zheng, L. Y., Sullivan, R., Qiao, Y.-L., & Lin, M. (2023). Gradient Informed Proximal Policy Optimization. In NeurIPS.

Chicago/Turabian Click to copy
Son, Sanghyun, Laura Yu Zheng, Ryan Sullivan, Yi-Ling Qiao, and Ming Lin. “Gradient Informed Proximal Policy Optimization.” In NeurIPS, 2023.

MLA Click to copy
Son, Sanghyun, et al. “Gradient Informed Proximal Policy Optimization.” NeurIPS, 2023.

BibTeX Click to copy

@inproceedings{son2023a,
  title = {Gradient Informed Proximal Policy Optimization},
  year = {2023},
  author = {Son, Sanghyun and Zheng, Laura Yu and Sullivan, Ryan and Qiao, Yi-Ling and Lin, Ming},
  booktitle = {NeurIPS}
}

Poster at NeurIPS 2023

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 {\alpha}-policy that stands as a locally superior policy. By adaptively modifying the {\alpha} 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.