Learning from Limited Data
We are interested in techniques to enable robots' learning from limited demonstrations. Towards this end, we have developed an Inverse Optimal Control (IOC) approach for the case of incomplete observations and also the case when objective functions are with multiple phases. Currently we are working on a generalization of this approach to cooperative IOC in multi-agent networks, and also Inverse Reinforcement Learning.
Scalable Algorithms for Distributed Fusion
We have developed a distributed algorithm for fusing linear observations, which converges exponentially fast, operates asynchronously, work for time-varying networks and involve no step-size. Further progress along this direction includes elimination of initialization, and decrease of state dimension by the sparsity. We have also recently developed a discrete-time distributed algorithm for least-square solution, and solutions with minimum L1 norm. This serves as a foundation for pre-processing sensing information instead of sending a huge amount of raw data in UAV networks.
Consensus - based Distributed Optimization
We are also interested in consensus-based distributed optimizations for multi-agent networks, in which each agent has an objective function and controls a state subject to its own local constraint. By coordination among nearby neighbors, all agents in the networks reach a consensus value which minimize the sum of all agents' cost functions. Applications include Distributed Resource Allocation, Distributed Coordination of Heterogeneous Vehicles and so on.
Resilience of Distributed Algorithms in Large Networks
We are also interested in resilience for distributed algorithms in large networks, in which each agent is only with locally available information while the attack is very sophisticated. The underlying network is time-varying and the malicious agent indicated in red is also with high mobility. We aim to develop systematic approaches to achieve automated resilience against Byzantine attacks without identification or isolation.
Experimental Research in Multi-Vehicle Coordination
We have also been interested in implementations of advanced control algorithms into multi-vehicles coordination.