Title: A Methodology for Training a team of Agents and its Application to a Problem in Air Traffic Control
Speaker: Kuang-Yuan (KY, Steven) Chen
(PhD Confirmation seminar)
Place: Room 621, GP South (Building 78)
Time: 1600 Thurs 6 Sept 2007
The project will consider the general problem of how machine learning techniques can be adapted to multi-agent team situations in which individuals attempt to optimise their individual performance while contributing to the best overall team performance.
The optimization problem is a non-classical Multi-Objective Problem (MOP), because there can be conflicts between the objectives of different agents or between those of the agent and the team. This project attempts to develop a methodology to decompose and optimize the MOP. The focus is on developing a methodology for finding near-optimal solutions with limited computing resources, rather than on finding fully optimal solutions.
The methodology is being developed on a case study from Air Traffic Management. The XCS machine learning technique is being used to develop rules for determining suitable trajectories for aircraft to fly. The team of aircraft needs to decide on a set of trajectories that is safe and which optimises team objectives, while optimising their own individual objectives.
Host: Peter Lindsay (UQ) & Hussein Abbass (UNSW@ADFA)