Towards Safe Reinforcement Learning with a Safety Editor Policy Haonan Yu, Wei Xu, and Haichao Zhang Horizon Robotics Cupertino, CA95014 {haonan.yu,wei.xu,haichao.zhang }@horizon.ai

Oct-2-2025, 09:59:15 GMT–Neural Information Processing Systems 

Assuming no prior knowledge or pre-training of the environment safety model given a task, an agent has to learn, via exploration, which states and actions are safe.