Safe Approximate Dynamic Programming Via Kernelized Lipschitz Estimation
This work addresses the challenge of ensuring safety during policy initialization in reinforcement learning for uncertain systems, which is an incremental improvement in safe control methods.
The paper tackles the problem of safe initialization for reinforcement learning in uncertain systems by developing a method using kernelized Lipschitz estimation and semidefinite programming to compute admissible initial control policies with provably high probability, resulting in safe initialization, constraint enforcement, and exponential stability of the closed-loop system.
We develop a method for obtaining safe initial policies for reinforcement learning via approximate dynamic programming (ADP) techniques for uncertain systems evolving with discrete-time dynamics. We employ kernelized Lipschitz estimation and semidefinite programming for computing admissible initial control policies with provably high probability. Such admissible controllers enable safe initialization and constraint enforcement while providing exponential stability of the equilibrium of the closed-loop system.