Embedding Safety into RL: A New Take on Trust Region Methods
This addresses safety concerns in RL for applications like robotics and autonomous systems, but it is incremental as it builds on existing trust region methods.
The paper tackles the problem of unsafe behavior in reinforcement learning agents by introducing Constrained Trust Region Policy Optimization (C-TRPO), which ensures safety constraints are satisfied throughout training while maintaining competitive returns.
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.