Reactions to a Multi-Agent Planning System Using Behaviour-Based Intelligence

Student: Rex Heathwood

Supervisor: Dr. Gordon Wyeth

Category: Computer Systems Engineering Thesis Project

The subject of this thesis is an implemented design for the action and reaction to a Multi Agent Planning System in the context of a simulated soccer match. The Multi Agent Planning System (MAPS) assesses the player’s situation by looking at its local vision and then decides upon an appropriate course of action. While MAPS is able to react to change, it is more a strategy-based intelligence, which is not always the best solution when reacting to a real time environment. The aim of this thesis is to take the suggested action and either implement the suggested action or react to its own local vision by implementing a different course of action. This is necessary as the game is played in a real-time dynamic environment.

Using a form of behaviour intelligence and basic subsumption architecture, the player is able to react to the real-time dynamic environment. The subsumption allows the player to override or subsume the MAPS suggested action when situations arise where the best action is not necessarily the suggested one. While reaction was a major part of this thesis, implementing the MAPS suggested actions or ‘Commands’ was just as important if not more so. Without a decent implementation of the MAPS commands, players would react slowly or even incorrectly, which in turn would reduce the usefulness of MAPS.

In the final solution the MAPS commands were implemented relatively efficiently, although the reactive capabilities were limited to the interception of the ball at defined distances.

Poster Presentation (PDF)

Thesis Document (PDF)