Safe Approximate Dynamic Programming Via Kernelized Lipschitz Estimation
Systems and Control
2019-07-05 v1 Machine Learning
Systems and Control
Dynamical Systems
Optimization and Control
Abstract
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.
Cite
@article{arxiv.1907.02151,
title = {Safe Approximate Dynamic Programming Via Kernelized Lipschitz Estimation},
author = {Ankush Chakrabarty and Devesh K. Jha and Gregery T. Buzzard and Yebin Wang and Kyriakos Vamvoudakis},
journal= {arXiv preprint arXiv:1907.02151},
year = {2019}
}