Adaptive attention allocation in human-robot systems

Abstract

We propose an optimization framework to study two fundamental attention control aspects in human-robot systems: Where and how much attention should the operator allocate? In other words, which information source should be observed by the operator, and how much time duration should be allocated to the information feed in order to optimize the overall performance of the human-robot system? The proposed framework incorporates (i) operator performance constraints, such as error rates and service times based utilization history, (ii) sensor constraints, such as processing/travel time, and (iii) task constraints, such as prioritization. We use a receding horizon approach to solve the resulting dynamic program, leading to efficient policies for operator time duration allocation and sensor selection. We demonstrate our methodology in a distributed surveillance problem.