🤖 AI Summary
To address the high computational cost of nonlinear model predictive control (NMPC) hindering real-time deployment, this paper proposes a safety-enhanced neural network controller. Methodologically, we introduce a novel verifiable neural architecture that integrates online feasibility verification and forward-integration dynamics; upon detecting output infeasibility or performance degradation, the controller automatically reverts to a precomputed safe candidate solution—thereby rigorously enforcing state and input constraints while guaranteeing closed-loop stability and convergence. Evaluated on three standard nonlinear NMPC benchmarks, the approach achieves sub-0.2 ms average inference latency, accelerating computation by several orders of magnitude over conventional NMPC solvers. It significantly outperforms unsafe “naïve” neural controllers and, for the first time, enables high-speed NMPC approximation with deterministic safety guarantees.
📝 Abstract
Model predictive control (MPC) achieves stability and constraint satisfaction for general nonlinear systems but requires computationally expensive online optimization. This brief studies approximations of such MPC controllers via neural networks (NNs) to achieve fast online evaluation. We propose safety augmentation that yields deterministic guarantees for convergence and constraint satisfaction despite approximation inaccuracies. We approximate the entire input sequence of the MPC with NNs, which allows us to verify online if it is a feasible solution to the MPC problem. We replace the NN solution by a safe candidate based on standard MPC techniques whenever it is infeasible or has worse cost. Our method requires a single evaluation of the NN and forward integration of the input sequence online, which is fast to compute on resource-constrained systems, typically within 0.2 ms. The proposed control framework is illustrated using three numerical nonlinear MPC benchmarks of different complexities, demonstrating computational speedups that are orders of magnitude higher than online optimization. In the examples, we achieve deterministic safety through the safety-augmented NNs, where a naive NN implementation fails.
