The performance Impact of Communication Failure in BlocksWorld for Teams
J.Z. van den Oever (TU Delft - Electrical Engineering, Mathematics and Computer Science)
W.P. Brinkman – Mentor (TU Delft - Interactive Intelligence)
Matthijs Spaan – Mentor (TU Delft - Algorithmics)
Koen V. Hindriks – Graduation committee member (Vrije Universiteit Amsterdam)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Multi-agent systems often communicate to perform better at their tasks. However, messages sometimes get lost after sending, and the communication is disrupted. This paper looks into the effects of messages having a chance of not arriving with an agent using the Blocks World for Teams simulation environment. The environment uses agent teams written in the GOAL programming language. First, this thesis details a pilot using existing agent teams with different communication failure models. Each agent team is run multiple times per failure model to get preliminary data. Based on this data, the thesis goes into a followup experiment. The experiment starts with the design of four agent teams, which only differ in the communication strategy they use to share information. What follows is the configuration of Blocks World for Teams, agent team size selection, the communication failure model, how the simulations are are run in a consistent manner, and finally, how the data generated by the simulations is processed. After modifying the agent teams, so they no longer waste limited resources, the experiment is repeated. The different communication strategies are affected to different degrees by communication failure. While the teams perform better than a single agent without failure, slowdown with communication failure can make the teams perform worse than the single agent. How often agents obstruct each other does not appear to be affected by how often communication fails as long as it does fail.