Developments in Computational Approaches for Model Predictive Control
Saturday, April 25, 2020 - 8:30am - 9:00am
Model Predictive Control (MPC) leads to algorithmically defined nonlinear feedback laws for systems with pointwise-in-time state and control constraints. These feedback laws are defined by solutions to appropriately posed dynamic optimization problems that are (typically) solved online. The talk will provide an overview of recent research by the presenter and his students and collaborators into several computational strategies for MPC solutions. These strategies include Newton-Kantorovich inexact methods, sensitivity-based warmstarting which exploits predicting changes to a parameterized optimal control problem based on semiderivative of the solution mapping, improvements to semismooth methods for solving convex quadratic programs, semismooth predictor-correct methods for suboptimal MPC, and strategies for handling large numbers of constraints. The talk will also touch upon the integration of game theoretic models based on cognitive hierarchy theory into model predictive control to facilitate decision-making in dynamic and interactive environments such as for self-driving cars operating in traffic.