Reinforcement Learning for Control with Probabilistic Stability Guarantee: A Finite-Sample Approach

Researchers have developed L-REINFORCE, a novel reinforcement learning algorithm that provides probabilistic stability guarantees for control systems using finite data samples. The approach bridges reinforcement learning and control theory by extending classical REINFORCE algorithms with Lyapunov stability methods, demonstrating superior performance in Cartpole simulations.

• →L-REINFORCE algorithm extends classical REINFORCE to provide probabilistic stability guarantees in control systems.
• →The method uses Lyapunov theory to ensure mean square stability with only finite sampled trajectories.
• →Stability probability increases with more trajectory data, converging to certainty as dataset grows.
• →Simulations on Cartpole tasks show L-REINFORCE outperforms baseline methods in ensuring stability.
• →The work enables model-free stability analysis and controller design in reinforcement learning frameworks.