Chris Adams MOBILE ROBOTICS RESEARCH GROUP reinforcement learning
Contact: Gillian Hayes
|
There has a been a good deal of interest in recent years in the role
that learning mechanisms might have to play in robotics. This work
aims to understand how
Reinforcement Learning techniques can be used in the automatic
programming of behaviour-based controllers for mobile robots. We
are currently investigating exploration mechanisms which traverse
the state space maximising the gain in information each step
(ie. take the action which will tell us most about how to do
well). To date an exploration technique for choosing actions in a
bandit problem has been developed. This is now being implemented
and tested for multi-state problems. The aim of the project is to
develop a principled algorithm for integrating teaching and
autonomous exploration.
Work on biological learning mechanisms and on connectionist approaches to environment modelling being carried out in the department is also related to RL. MSc projects: |