Cooperative Assembly Agents - 2 RMM Agents trying to move the blocks to the upper right corner in the order A, B, C. The symbol . represents agent one and .. represents agent two. The symbols A, B and C represent the blocks.
This video shows the effect of two agents using an RMM structure of four levels. Each level contains two models, a rational agent model and a no-informational model. The models are combined using preset probabilistic values. The rational model returns a definitive answer, whereas the no-informational model returns a uniform distribution on the choice of actions. Since neither a rational or no-informational model is inherently cooperative this example shows the emergent property of cooperation via RMM.
Here are the same two agents with four blocks. Please note the element labeled T is just a stationary object not a block. Next, this video shows the effect of bayesian belief updating for an agent using 3 RMM models of the other agent. The models are a rational, random, and no action agent. These possible models of the other agent are probabilisticly mixed for deciding upon the other agent's action. Bayes formula is used to update this mixing via probability action assignment. This assignment is made via continuous observation of the actions of the other agent. Note: In this example agent two does nothing. Agent one assumes at the start with a high probability that agent two is a rational agent. But after seeing that agent two does nothing agent one modifies its belief about agent two and carries out all the block moving.This work has been sponsored by the Office of Naval Research under contract N00014-95-1-0775.
MISSION · PUBLICATION · MEMBER · RESEARCH · LINKS
© 1998 Multi-Agent Systems Group at UIC