Neural approximation can inherit robust MPC safety and stability guarantees
Original title: Certified Neural Approximation of Robust Nonlinear Model Predictive Control
Design conditions treat network error as a bounded disturbance so feasibility and stability hold before deployment.
A new framework builds neural controllers that are meant to keep the safety and stability of a robust nonlinear model predictive policy. Unlike post-verification or extra online safety filters, it sets constructive conditions before deployment and treats approximation error as a bounded input disturbance inside the MPC design. The offline synthesis needs specified Lipschitz bounds. Simulation shows the learned controller matches closed-loop behavior while cutting online computation.
Unconfirmed: Reported behavior comes from simulation; no hardware deployment is described.
- Approximation error is handled as a bounded input disturbance.
- Recursive feasibility and stability are designed in before deployment.
- The pipeline runs offline once Lipschitz bounds are specified.
- Simulation reports similar closed-loop behavior at lower compute.
From the original article
1 min read at arxiv.orgModel predictive control is a high-performance control method, but it incurs comparatively high online computational costs across many control tasks and is challenging to realize in low-level hardware settings. To address this, we present a framework for generating neural network controllers that provably inherit the safety and stability guarantees of a robust model predictive policy.
By Hendrik Alsmeier · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.