A safety governor for learning explicit MPC controllers from data

2025-07-21

AI Summary

This paper addresses the challenge of simultaneously ensuring constraint satisfaction and maintaining accurate linearization near the origin in learning-based explicit model predictive control (MPC). To this end, we propose a novel dual-mode neural network controller architecture. Methodologically, explicit MPC is reformulated as a “safety governor + local linear feedback” structure: a neural network learns the global nonlinear control policy, while the safety governor—based on the maximum constraint-admissible set—performs real-time input validation and correction to guarantee recursive feasibility and strict satisfaction of state and input constraints; within a neighborhood of the origin, the controller automatically reverts to linear state feedback. Our key contribution is the first integration of a safety governor into a dual-mode learning-based explicit MPC framework, markedly improving real-time performance and robustness for high-dimensional systems. Numerical experiments demonstrate that the proposed method ensures strict constraint satisfaction and closed-loop stability while achieving superior computational efficiency compared to existing learning-based MPC approaches.

Technology Category

• Natural Language Processing: Safety and Robustness
• Search and Optimization: Learning to Search
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

We tackle neural networks (NNs) to approximate model predictive control (MPC) laws. We propose a novel learning-based explicit MPC structure, which is reformulated into a dual-mode scheme over maximal constrained feasible set. The scheme ensuring the learning-based explicit MPC reduces to linear feedback control while entering the neighborhood of origin. We construct a safety governor to ensure that learning-based explicit MPC satisfies all the state and input constraints. Compare to the existing approach, our approach is computationally easier to implement even in high-dimensional system. The proof of recursive feasibility for the safety governor is given. Our approach is demonstrated on numerical examples.

Problem

• Approximating MPC laws using neural networks
• Ensuring safety and constraints in learning-based MPC
• Simplifying implementation in high-dimensional systems

Innovation

• Dual-mode scheme for learning-based explicit MPC
• Safety governor ensures constraint satisfaction
• Computationally easier in high-dimensional systems