Coalition formation and trust in collaborative control

We consider the collaboration of autonomous agents as a coalitional game subject to constraints, like communication, trust or reputation. We analyze the tradeoff between the benefits of collaboration and these constraints (or costs of collaboration) via dynamic, iterative stochastic games. The agents can observe locally the actions of other agents in their neighborhood and decide whether to collaborate or not in a distributed asynchronous manner. Trust values are taken into consideration when agents act and collaborate. Specifically they affect the selection of agents to collaborate with. We consider only problems where the information used for either estimating reputations or deciding on strategies is mostly local; i.e. from the neighbors of each agent. We show that randomized algorithms emerge as the agents try to reach the maximum payoff. We also investigate the topology of the formed network by studying the second largest eigenvalue of the corresponding graph, and describe its effects on control performance.