DMPC: A data-and model-driven approach to predictive control

This work presents DMPC (Data-and Model-Driven Predictive Control) to solve control problems in which some of the constraints or parts objective function are known, while others entirely unknown controller. It is assumed that there an exogenous “black box” system, e.g. a machine learning technique, predicts value functions for given trajectory. (1) provides approach merge both model-based and black-box systems; (2) can cope with very little data sample efficient, building its solutions based on recently generated trajectories; (3) improves cost each iteration until converging optimal trajectory, typically needing only few trials even nonlinear dynamics objectives. Theoretical analysis algorithm presented, proving quality trajectory does not worsen new iteration. We apply motion planning autonomous vehicle dynamics.