Cohesive networks using delayed self-reinforcement 

Abstract: How a decentralized network gets to the goal (a consensus value) can be as important as reaching the consensus value. While current network theories seek to enable fast synchronization (fast convergence to the steady state), maintaining cohesion, i.e., synchronization, during the transition between consensus values or during tracking of a time varying goal, remains challenging and has not been addressed. Maintaining cohesion is important, e.g., to maintain similar inter-vehicle spacing in connected automated transportation systems, alignment synchronization to help maintain formations during maneuvers of flocks and swarms in nature, to avoid damage due to large deformations when transporting flexible objects and to maintain formation of engineered networks such as satellites, unmanned autonomous vehicles and collaborative robots. 

The challenge to maintain cohesion between agents arises because information about the desired response (such as the desired orientation or speed of the agents) might be available to only a few agents in a decentralized framework. The desired-response information needs to be propagated through the network to other agents, which results in response-time delays between agents that are “close to” the information source and those that are “farther away.”  

The talk will start with a centralized approach where all agents have access to the time varying goal and the use of inversion theory to enable precision tracking for nonlinear heterogeneous agents resulting in an ideal cohesive network. Then, we develop a decentralized implementation of this ideal centralized approach using a delayed self-reinforcement (DSR) approach, where each individual agent augments its neighbor-based information update using its previously available updates, to improve cohesiveness of the network response during transitions. The advantages of the proposed DSR approach are that it only requires already-available information from a given network to improve the cohesion and does not require network-connectivity modifications (which might not be always feasible) nor increases in the system’s overall response speed (which can require larger input). Results are presented that show substantial improvement in cohesion with DSR.

Bio: Santosh Devasia received the B.Tech. (Hons) from the Indian Institute of Technology, Kharagpur, India, in 1988, and the M.S. and Ph.D. degrees in Mechanical Engineering (ME) from the University of California at Santa Barbara in 1990 and 1993 respectively. He is the Minoru Taya Endowed Chair from 2018-2024 at UW, Seattle. He joined the faculty of the UW Mechanical Engineering (ME) Department in 2000 after teaching from 1994 to 2000 in the ME Department at the University of Utah, Salt Lake City. He is a fellow of ASME and IEEE. His current research interests include control of multi-agent systems and precision human-machine systems.