Research

My current research mainly falls into developing distributed algorithms for multi-agent systems consisting of multiple unmanned mobile agents, through the combination of theories in control, networks, optimizations, and learning.

Currently, I am building a lab call COL-LAB.

What is Multi-Agent System?

•Multi-agent systems are networks of autonomous agents that collaborate to solve problems that are difficult for an individual agent to solve.

•The agents: robots, sensors, or any other physical devices with on-board processing, sensing and communication capabilities.

•The problem: develop distributed algorithms for computing, estimating, or controlling various quantities across a network consisting of a group of agents.

Distributed Algorithm for Multi-Agent Systems:

Estimation, Control, and Learning

Inspired by nature, it is expected that by cooperating multiple autonomous agents can offer better performance in complicated missions such as exploration of the unknown, disaster rescue, etc.

Distributed algorithms for state estimation, with arbitrary estimation convergence rate.
Distributed algorithms for formation control based on local information with global stability.
Distributed algorithms for clustering with exponential convergence.

Karthik, S.. “Underwater Vehicle for Surveillance with Navigation and Swarm Network Communication.” Indian journal of science and technology 7 (2014): 22-31.

https://www.416th.com/TypicalMission_Formation.html

Generalize clustered model of WSN

Future Vision

All models are wrong, but some are useful: create systematic approaches to achieve distributed estimation for stochastic models.

For the scenario when it is not possible to develop accurate mathematical models for the physical systems, it is necessary to develop distributed model-free estimation.

For most multi-agent systems, it is necessary to assume that all the agents in the network behave or operate reliably. The system becomes vulnerable in the presence of malicious agents. Two problems arise. One is to find a systematic approach to achieve automated resilience against attacks without identification. The other is to design an algorithm to identify the malicious agents through learning, or a well-defined index.

Page updated

Google Sites

Report abuse