Opportunistic robot control for interactive multiobjective optimization under human performance limitations

Abstract

This paper proposes opportunistic state-triggered strategies for solving convex multiobjective optimization problems that involve human–robot interaction. The robot is aware of the multiple objective functions defining the problem, but requires human input to find the most desirable Pareto solution. In order to avoid overloading the human with queries, we view her as a limited resource to the robot, and design event-triggered controllers that opportunistically prescribe the information exchanges among them. We consider various models of human performance, starting with an ideal one where queries are responded instantaneously, and later considering constraints on the response time and the interaction frequency. For each model, we formally establish the asymptotic convergence to the desired optimizer and rule out the existence of Zeno behavior.