Multi-source Machine Learning for Control

Vision

• We focus on improving data-driven control using data-driven learning. We develop reinforcement learning techniques that leverage knowledge of the physics of the system to solve the two-point boundary optimal control problem of linear time-varying systems with unknown model dynamics. We also extend learning from the control of low-order linear systems to large-scale complex nonlinear systems.

Funded Projects

• Multi-agent Reinforcement Learning for Traffic Control – Siemens 2022

Publications

• Trajectory Generation using Activator-Inhibitor Systems – IEEE CDC 2023
• Near-Optimal Trajectory Generation for Flexible Motion Systems using Two-Boundary Approach – ECC 2023
• Singular Perturbation-based Reinforcement Learning of Two-Point Boundary Optimal Control Systems – American Control Conference (ACC 2022) – code, presentation, poster
• Cooperation Learning in Time-Varying Multi-Agent Networks – AAAI Workshop on Reinforcement Learning in Games (AAA-RLG 2022) – code, presentation, poster