Multi-Constraint Safe RL with Objective Suppression for Safety-Critical Applications

arXiv:2402.15650v1 Announce Type: new
Abstract: Safe reinforcement learning tasks with multiple constraints are a challenging domain despite being very common in the real world. To address this challenge, we propose Objective Suppression, a novel method that adaptively suppresses the task reward maximizing objectives according to a safety critic. We benchmark Objective Suppression in two multi-constraint safety domains, including an autonomous driving domain where any incorrect behavior can lead to disastrous consequences. Empirically, we demonstrate that our proposed method, when combined …

abstract applications arxiv benchmark challenge constraints cs.ai cs.lg domain multiple novel reinforcement reinforcement learning safety safety-critical tasks type world