Safe Distributional Reinforcement Learning
This work addresses safety concerns in RL for domains like autonomous driving or finance, but it is incremental as it builds on existing safe policy optimization methods.
The paper tackles the problem of ensuring safety in reinforcement learning by formalizing it as a constrained RL problem within a distributional RL setting, resulting in a more efficient algorithm that handles various safety definitions and is validated empirically against state-of-the-art methods.
Safety in reinforcement learning (RL) is a key property in both training and execution in many domains such as autonomous driving or finance. In this paper, we formalize it with a constrained RL formulation in the distributional RL setting. Our general model accepts various definitions of safety(e.g., bounds on expected performance, CVaR, variance, or probability of reaching bad states). To ensure safety during learning, we extend a safe policy optimization method to solve our problem. The distributional RL perspective leads to a more efficient algorithm while additionally catering for natural safe constraints. We empirically validate our propositions on artificial and real domains against appropriate state-of-the-art safe RL algorithms.