Integrating human and robot decision-making dynamics

Humans and robots each have strengths and weaknesses associated with making good decisions to address complex tasks in uncertain, changing environments. We investigate how humans and robots can best jointly contribute to decision making so that strengths are exploited and weaknesses compensated. Our approach to this integration problem is to leverage experimental and modeling work of psychologists on human decision making. We seek commonality between the kinds of decisions humans make in complex tasks and the kinds of decisions humans make in psychology experiments; when commonality conditions are met, the psychology results can be used to predict how humans will behave in the complex task. A problem well studied in the psychology literature is the two-alternative forced-choice task, in which the human subject chooses between two options at regular time intervals and receives a reward after each choice. Interestingly, experiments show convergence of the aggregate behavior to rewards that are often suboptimal. We introduce a decision-making problem associated with a complex task that integrates human and robotic decision-making dynamics with feedback. The setting is a human-supervised collective robotic foraging problem, where the human decision-making takes the form of a two-alternative forced-choice task and the reward report is a feedback from the robots. Using a popular, experimentally verified, decision-making model, we prove convergence of the human behavior to the observed aggregate decision making for reward structures with matching points. Since behavior converges to suboptimal performance, we show how adaptive laws for the robot feedback, which use only local information, can be applied to help the human make optimal decisions.