渡邊 紀文

Human cooperative behavior includes passive action strategies based on others and active action strategies that prioritize one’s own objective. Therefore, for cooperation with humans, it is necessary to realize a robot that uses these strategies to communicate as a human would. In this research, we aim to realize robots that evaluate the actions of their opponents in comparison with their own action strategies. In our previous work, we obtained a Meta-Strategy with two action strategies through the simulation of learning between agents. However, humans’ Meta-Strategies may have different characteristics depending on the individual in question. In this study, we conducted a collision avoidance experiment in a grid space with agents with active and passive strategies for giving way. In addition, we analyzed whether a subject’s action changes when the agent’s strategy changes. The results showed that some subjects changed their actions in response to changes in the agent’s strategy, as well as subjects who behaved in a certain way regardless of the agent’s strategy and subjects who did not divide their actions. We considered that these types could be expressed in terms of differences in Meta-Strategies, such as active or passive Meta-Strategies for estimating an opponent’s strategy. Assuming a human Meta-Strategy, we discuss the action strategies of agents who can switch between active and passive strategies.

In human cooperative behavior, there are some strategies: a passive behavioral strategy based on others' behaviors and an active behavioral strategy based on the objective-first. However, it is unclear how to acquire a meta-strategy to switch those strategies. In this study, we conduct a collision-avoidance experiment with agents taking multiple strategies in a grid-like corridor to see whether the subject's behavior changes when the agent's strategy changes. Furthermore, we compare the behavior selected by the subjects with the behavior of the agents acquired by reinforcement learning. The experimental results show that subjects can read the change in strategy from the behavior of the opposing agent.