Stochastic Search and Surveillance Strategies for Mixed Human-Robot Teams

Stochastic Search and Surveillance Strategies for Mixed Human-Robot Teams Vaibhav Srivastava Mixed human-robot teams are becoming increasingly important in complex and information rich systems. The purpose of the mixed teams is to exploit the human cognitive abilities in complex missions. It has been evident that the information overload in these complex missions has a detrimental effect on the human performance. The focus of this dissertation is design of efficient human-robot teams. It is imperative for an efficient human-robot team to handle information overload and to this end, we propose a two-pronged strategy: (i) for the robots, we propose strategies for efficient information aggregation; and (ii) for the operator, we propose strategies for efficient information processing. The proposed strategies rely on team objective as well as cognitive performance of the human operator. In the context of information aggregation, we consider two particular missions. First, we consider information aggregation for a multiple alternative decision making task and pose it as a sensor selection problem in sequential multiple hypothesis testing. We design efficient information aggregation policies that enable the human operator to decide in minimum time. Second, we consider a surveillance problem and design efficient information aggregation policies that enable the human operator detect a change in the environment in minimum time. We study the surveillance problem in a decision-theoretic framework and rely on statistical quickest change detection algorithms to achieve a guaranteed surveillance performance. In the context of information processing, we consider two particular scenarios. First, we consider the time-constrained human operator and study optimal resource allocation problems for the operator. We pose these resource allocation problems as knapsack problems with sigmoid utility and develop constant factor algorithms for them. Second, we consider the human operator serving a queue of decision making tasks and determine optimal information processing policies. We pose this problem in a Markov decision process framework and determine approximate solution using certainty-equivalent receding horizon framework.