About the project
The aim of this PhD project is to explain the behaviour of a swarm using generative imitation learning.
Understanding large complex systems allows us to closely mirror numerous natural and man-made phenomena, from the flocking behavior of birds to the flow of traffic in dense urban areas.
Our cells are also large complex systems that respond to their local environment and fulfil specific roles. Understanding the underlying principles of these swarm systems can provide insights into optimizing and managing such phenomena in real life.
Many natural systems, from ant colonies to human crowds, rely on swarm-like behaviors, and deciphering these can offer valuable insights into their optimisation and management. In the technological domain, as we transition towards a more interconnected world, swarm intelligence can drive advancements in fields like robotics, traffic management, and automated systems, enhancing efficiency and adaptability.
Collective behavior in swarms emerges from local interactions between neighboring agents and their environment. Understanding these interactions is vital for controlling swarm behavior.
In this project, you will use machine learning techniques, specifically generative imitation learning, to replicate the behavior of the swarm and extract reward functions that lead to such collective behavior.
By studying agent-level policies and multi-agent interactions, the focus is on reproducing an observed collective behavior. This will help us understand how behaviors emerge from local interactions and how certain actions can guide the swarm towards desired outcomes. For example, by feeding a set of videos from a traffic flow, the algorithm will be able to extract micro-level actions from the drivers or certain properties of the road that are causing a certain flow and can suggest interventions that can lead to more desirable collective behavior.
