Data-Driven MPC with Adaptively Sampled Non-Expert Demonstrations: Performance Guarantees and Sample Complexity

This arXiv paper presents a novel approach to data-driven policy construction for dynamical systems using trajectories generated by surrogate finite-horizon Model Predictive Control. The method interprets these trajectories as non-expert demonstrations and proposes a memory-based, nonparametric policy that constructs a Lipschitz-regularized upper envelope of the finite-horizon surrogate value function. The research provides theoretical performance guarantees and sample complexity analysis for approximating infinite-horizon optimal control policies.