Rational Coordination in Cognitive Agents: A Decision-Theoretic Approach Using ERMM

Over the years, research in multi-agent systems has become increasingly popular. Agents evolve by interacting with their environment and must communicate with other agents in order to do various cooperative tasks. The research aims to provide efficient coordination among cooperative cognitive agents in unpredictable multi-agent situations. Xiang’s rational agent model addresses scenarios when no social conventions or predefined communication protocols exist for the agents’ interaction and then makes decisions by recursive modeling. We address the deficiencies of the loosely coupled framework and the problem of mispredictions in Xiang’s architecture. It is based on Lawniczak’s Architecture for generic cognitive agents and an enhanced model of Xiang’s Recursive Modeling Method for coordinated decision-making in multi-agent situations. We instruct the cognitive agent to learn about other agents from past mispredictions and then consider its best choice. The feedback module is incorporated so agents can learn to maximize their joint expected reward. It filters the mispredictions and evaluates the error rate. We compare the enhanced method with the Recursive Modeling Method. The results show that mispredictions are corrected from 33% to 10.9% and errors in perception get reduced 22% to 0.097%, as the system progresses. Overall, the approach demonstrates superior performance. It significantly lowers the rate of mispredictions about other agents’ actions and takes 30% to 42% less time and 55.4 % fewer moves than RMM.