Towards Practical Planning to Predict and Exploit Intentions for Interacting with Self-Interested Agents

A key challenge in non-cooperative multi-agent systems is that of developing effi-cient planning algorithms for intelligent agents to interact and perform effectivelyamong boundedly rational, self-interested agents (e.g., humans). The practicalityof existing works addressing this challenge is being undermined due to either therestrictive assumptions of the other agents’ behavior, the failure in accounting fortheir rationality, or the prohibitively expensive cost of modeling and predictingtheir intentions. To boost the practicality of research in this field, we investigatehow intention prediction can be efficiently exploited and made practical in plan-ning, thereby leading to efficient intention-aware planning frameworks capable ofpredicting the intentions of other agents and acting optimally with respect to theirpredicted intentions. We show that the performance losses incurred by the result-ing planning policies are linearly bounded by the error of intention prediction.Empirical evaluations through a series of stochastic games demonstrate that ourpolicies can achieve better and more robust performance than the state-of-the-artalgorithms.