Predictive Control
Model Predictive Control (MPC) is an advanced control technique that optimizes a system's trajectory over a future time horizon by iteratively solving optimization problems. Current research emphasizes improving MPC's efficiency and robustness, particularly through integrating machine learning methods like reinforcement learning and neural networks (e.g., using transformers for faster computation or neural networks to approximate MPC solutions), and addressing challenges posed by model uncertainties and safety constraints using techniques such as control barrier functions and Bayesian optimization. These advancements are significantly impacting various fields, including robotics, autonomous vehicles, and energy systems, by enabling more efficient, safe, and adaptable control of complex systems.
Papers
First, Learn What You Don't Know: Active Information Gathering for Driving at the Limits of Handling
Alexander Davydov, Franck Djeumou, Marcus Greiff, Makoto Suminaka, Michael Thompson, John Subosits, Thomas Lew
Transformer-based Model Predictive Control: Trajectory Optimization via Sequence Modeling
Davide Celestini, Daniele Gammelli, Tommaso Guffanti, Simone D'Amico, Elisa Capello, Marco Pavone
Tiny Learning-Based MPC for Multirotors: Solver-Aware Learning for Efficient Embedded Predictive Control
Babak Akbari, Justin Frank, Melissa Greeff
Model Predictive Control is Almost Optimal for Restless Bandit
Nicolas Gast, Dheeraj Narasimha
Construction of Musculoskeletal Simulation for Shoulder Complex with Ligaments and Its Validation via Model Predictive Control
Yuta Sahara, Akihiro Miki, Yoshimoto Ribayashi, Shunnosuke Yoshimura, Kento Kawaharazuka, Kei Okada, Masayuki Inaba