Safety-Polarized and Prioritized Reinforcement Learning

Ke Fan, Jinpeng Zhang, Xuefeng Zhang, Yunze Wu, Jingyu Cao, Yuan Zhou, Jianzhu Ma
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:15862-15886, 2025.

Abstract

Motivated by the first priority of safety in many real-world applications, we propose MaxSafe, a chance-constrained bi-level optimization framework for safe reinforcement learning. MaxSafe first minimizes the unsafe probability and then maximizes the return among the safest policies. We provide a tailored Q-learning algorithm for the MaxSafe objective, featuring a novel learning process for optimal action masks with theoretical convergence guarantees. To enable the application of our algorithm to large-scale experiments, we introduce two key techniques: safety polarization and safety prioritized experience replay. Safety polarization generalizes the optimal action masking by polarizing the Q-function, which assigns low values to unsafe state-action pairs, effectively discouraging their selection. In parallel, safety prioritized experience replay enhances the learning of optimal action masks by prioritizing samples based on temporal-difference (TD) errors derived from our proposed state-action reachability estimation functions. This approach efficiently addresses the challenges posed by sparse cost signals. Experiments on diverse autonomous driving and safe control tasks show that our methods achieve near-maximal safety and an optimal reward-safety trade-off.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-fan25i, title = {Safety-Polarized and Prioritized Reinforcement Learning}, author = {Fan, Ke and Zhang, Jinpeng and Zhang, Xuefeng and Wu, Yunze and Cao, Jingyu and Zhou, Yuan and Ma, Jianzhu}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {15862--15886}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/fan25i/fan25i.pdf}, url = {https://proceedings.mlr.press/v267/fan25i.html}, abstract = {Motivated by the first priority of safety in many real-world applications, we propose MaxSafe, a chance-constrained bi-level optimization framework for safe reinforcement learning. MaxSafe first minimizes the unsafe probability and then maximizes the return among the safest policies. We provide a tailored Q-learning algorithm for the MaxSafe objective, featuring a novel learning process for optimal action masks with theoretical convergence guarantees. To enable the application of our algorithm to large-scale experiments, we introduce two key techniques: safety polarization and safety prioritized experience replay. Safety polarization generalizes the optimal action masking by polarizing the Q-function, which assigns low values to unsafe state-action pairs, effectively discouraging their selection. In parallel, safety prioritized experience replay enhances the learning of optimal action masks by prioritizing samples based on temporal-difference (TD) errors derived from our proposed state-action reachability estimation functions. This approach efficiently addresses the challenges posed by sparse cost signals. Experiments on diverse autonomous driving and safe control tasks show that our methods achieve near-maximal safety and an optimal reward-safety trade-off.} }
Endnote
%0 Conference Paper %T Safety-Polarized and Prioritized Reinforcement Learning %A Ke Fan %A Jinpeng Zhang %A Xuefeng Zhang %A Yunze Wu %A Jingyu Cao %A Yuan Zhou %A Jianzhu Ma %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-fan25i %I PMLR %P 15862--15886 %U https://proceedings.mlr.press/v267/fan25i.html %V 267 %X Motivated by the first priority of safety in many real-world applications, we propose MaxSafe, a chance-constrained bi-level optimization framework for safe reinforcement learning. MaxSafe first minimizes the unsafe probability and then maximizes the return among the safest policies. We provide a tailored Q-learning algorithm for the MaxSafe objective, featuring a novel learning process for optimal action masks with theoretical convergence guarantees. To enable the application of our algorithm to large-scale experiments, we introduce two key techniques: safety polarization and safety prioritized experience replay. Safety polarization generalizes the optimal action masking by polarizing the Q-function, which assigns low values to unsafe state-action pairs, effectively discouraging their selection. In parallel, safety prioritized experience replay enhances the learning of optimal action masks by prioritizing samples based on temporal-difference (TD) errors derived from our proposed state-action reachability estimation functions. This approach efficiently addresses the challenges posed by sparse cost signals. Experiments on diverse autonomous driving and safe control tasks show that our methods achieve near-maximal safety and an optimal reward-safety trade-off.
APA
Fan, K., Zhang, J., Zhang, X., Wu, Y., Cao, J., Zhou, Y. & Ma, J.. (2025). Safety-Polarized and Prioritized Reinforcement Learning. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:15862-15886 Available from https://proceedings.mlr.press/v267/fan25i.html.

Related Material