A. Katriniok | Searcharxiv

arXiv2025

Value Function Approximation for Nonlinear MPC: Learning a Terminal Cost Function with a Descent Property

We present a novel method to synthesize a terminal cost function for a nonlinear model predictive controller (MPC) through value function approximation using supervised learning. Existing methods enforce a descent property on the terminal cost function by construction, thereby restricting the class of terminal cost functions, which in turn can limit the performance and applicability of the MPC. We present a method to approximate the true cost-to-go with a general function approximator that is convex in its parameters, and impose the descent condition on a finite number of states. Through the scenario approach, we provide probabilistic guarantees on the descent condition of the terminal cost function over the continuous state space. We demonstrate and empirically verify our method in a numerical example. By learning a terminal cost function, the prediction horizon of the MPC can be significantly reduced, resulting in reduced online computational complexity while maintaining good closed-loop performance.

math.OC↗

arXiv2025

Online Learning of Interaction Dynamics with Dual Model Predictive Control for Multi-Agent Systems Using Gaussian Processes

The control of a single agent in complex and uncertain multi-agent environments requires careful consideration of the interactions between the agents. In this context, this paper proposes a dual model predictive control (MPC) method using Gaussian process (GP) models for multi-agent systems. While Gaussian process MPC (GP-MPC) has been shown to be effective in predicting the dynamics of other agents, current methods do not consider the influence of the control input on the covariance of the predictions, and hence lack the dual control effect. Therefore, we propose a dual MPC that directly optimizes the actions of the ego agent, and the belief of the other agents by jointly optimizing their state trajectories as well as the associated covariance while considering their interactions through a GP. We demonstrate our GP-MPC method in a simulation study on autonomous driving, showing improved prediction quality compared to a baseline stochastic MPC. The results show that GP-MPC can learn the interactions between the agents online, demonstrating the potential of GPs for dual MPC in uncertain and unseen scenarios.

math.OC↗