A research conducted by Ryuta Moriyasu et al. in collaboration with Toyota Industries Corporation and Kyoto University has been published in Automatica.
Model Predictive Control (MPC) is a widely adopted control methodology in industry, including automotive applications, where system behavior is predicted and optimized to compute control inputs. However, MPC requires solving an optimization problem at every control step, which leads to substantial computational burden. As a promising approach to this challenge, fast MPC has recently attracted attention, where the control input is applied before the optimization fully converges. Despite its potential, fast MPC may lose closed-loop stability once discretized for digital execution, which is essential for practical control systems.
In this study, we propose a discrete-time implementation of the Primal-Dual Gradient Dynamics - a core algorithm in fast MPC - that guarantees stability. By introducing modified update rules and step-size conditions, the proposed method achieves stability and high control performance while keeping computational complexity extremely low.
These results are expected to serve as foundational technology for enhanced reliability and advanced control performance in industrial machinery and other applications requiring real-time capabilities.
Title : Sampled-Data Primal-Dual Gradient Dynamics in Model Predictive Control
Authors: Moriyasu, R., Kawaguchi, S., Kashima, K.
Journal Name: Automatica
Published: September 22, 2025
https://doi.org/10.1016/j.automatica.2025.112621