Research profile A. (Aliriza) Zolanvari

A. (Alireza) Zolanvari

Project: Interactive Adaptive Multi-agent Imitation Learning

Due to the increase in the number of modern engineering systems comprising of a large number of interacting components, dealing with complex real-world multi-agent problems is quite important. Multi-agent reinforcement learning and imitation learning are among the most powerful approaches in solving such problems. However, each of these approaches suffers from undeniable drawbacks. For example, imitation learning-based approaches need a large number of near-optimal demonstrations to perform appropriately.

In this project, we propose a novel multi-agent imitation learning algorithm by combining the underlying idea of imitation learning, generative deep learning, and control theory techniques. Our goal is to deal with safety-critical complex multi-agent problems more effectively than the present approaches. Furthermore, the developed solution strategy will have better sample guarantees and will be able to tackle sparse-reward environments.

Keywords: Multiagent networks, Imitation learning, Deep generative networks, Control theory

Fields of expertise involved: Artificial Intelligence, Complex Systems, Control systems