Distributed Control and Optimization
Multi-agent systems find applications in diverse areas, including engineering, economics, social and computer science. We have achieved several important results on the consensus of multi-agent systems with various dynamics and under diverse constraints such as intermittent communication, time delay, input saturation and measurement noises.
Resilient Reinforcement Learning
Multi-agent reinforcement learning (MARL) endows large-scale systems with the intelligence of learning the optimal behavior by interacting with an unknown and complex environment. We reveal that traditional consensus-based reinforcement learning algorithms are vulnerable to adversary attacks. Based on this fact, we propose a resilient reinforcement learning algorithm that can effectively mitigate the influence of adversary attacks.
Event-Triggered Control Design for Constrained Systems
Event-triggered control has the advantage of reducing unnecessary samplings and requiring fewer control updates, which helps to conserve the resources of a control system. To facilitate the implementation of event-triggered control algorithms in real-world systems, we take input saturation and output saturation into consideration and design control algorithms for systems with different dynamics.