Keywords: multi-agent reinforcement learning, human-machine teaming, multi-agent Cooperation, Visual Search
Abstract: Human-machine teaming is a challenging and important problem that requires designing autonomous agents that can effectively cooperate with humans in complex and dynamic scenarios. In this paper, we explore multi-agent dynamics in a reinforcement learning (RL) framework in a Stag Hunt scenario, where interactions can be either cooperative or independent. We design a system involving two agents: Agent A, which follows a mixed strategy based on predefined probability distributions, and Agent B, an RL agent with human-like visual constraint that learns an adaptive strategy to selectively sample information from the environment to infer Agent A’s intent. Our results indicate that Agent B adapts its policy effectively, exhibiting adaptive gaze strategies tailored to the policy of Agent A. We discuss the implications of our findings and the design of RL agents capable of interacting with human-like agents.
Submission Number: 7
Loading