Actor-critic reinforcement learning methods are reformulated as mirror descent on the occupancy manifold, and a Hessian-based update is proposed for nonlinear and constrained objectives.
Embedding Safety into RL: A New Take on Trust Region Methods
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
abstract
Reinforcement Learning (RL) agents can solve diverse tasks but often exhibit unsafe behavior. Constrained Markov Decision Processes (CMDPs) address this by enforcing safety constraints, yet existing methods either sacrifice reward maximization or allow unsafe training. We introduce Constrained Trust Region Policy Optimization (C-TRPO), which reshapes the policy space geometry to ensure trust regions contain only safe policies, guaranteeing constraint satisfaction throughout training. We analyze its theoretical properties and connections to TRPO, Natural Policy Gradient (NPG), and Constrained Policy Optimization (CPO). Experiments show that C-TRPO reduces constraint violations while maintaining competitive returns.
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
The Geometry of Nonlinear Reinforcement Learning
Actor-critic reinforcement learning methods are reformulated as mirror descent on the occupancy manifold, and a Hessian-based update is proposed for nonlinear and constrained objectives.