Director of studies: Dr Alan Millard
2nd supervisor: Dr David Walker
3rd supervisor: Dr Sanjay Sharma
Applications are invited for a three-year MPhil/PhD studentship. The studentship will start on 1 October 2019.
Swarm robotics is an approach to the coordination of large numbers of robots that places emphasis on local sensing and decentralised control to create flexible, scalable, and robust systems. Engineering these systems is non-trivial, as the designer must determine how individual robots should behave such that the desired collective behaviour emerges from their interactions. Automated approaches typically attempt to solve this problem with metaheuristics by optimising neural networks / finite state machines that control robot behaviour. However, these approaches often rely upon offline learning in simulation prior to deployment on real robots, resulting in solutions tailored to a particular robot platform, task, or environment, which cannot cope with uncertainty and fail to generalise to unseen scenarios.
PhD aims to overcome these issues through the application of hyper-heuristics,
which identify heuristics for generating solutions to optimisation problems, by
combining machine learning with metaheuristics. They are either selection-based
(identifying good heuristics from a predefined pool), or generative-based
(creating novel heuristics as the optimisation proceeds), and are generally
trained online, making them ideally suited to use in dynamic environments where
the problem landscape is constantly changing.
Research will initially be undertaken in our dedicated Swarm Robotics Laboratory, using a swarm of 40 e-puck robots. This work will then be extended to small autonomous vessels in our COAST Laboratory, to assess the swarm’s ability to perform inspection/maintenance tasks on a scale model of an offshore wind farm, while simultaneously adapting to changing wave conditions representative of a real marine environment.