Towards the Principled Study of Variable Autonomy in Mobile Robots

Abstract

Safety critical and demanding tasks (e.g. Search and rescue or hazardous environments inspection), can benefit from robotic systems that offer a spectrum of control modes. These can range from direct teleoperation to full autonomy. This paper describes a pilot-study experiment in which a variable autonomy robot completes a navigation task. It explores the comparative performances of the human-robot system at different autonomy levels under different sets of conditions. This is done from a Mixed-Initiative system investigation perspective. Sensor noise was added to degrade robot performance, while a secondary task induced varying degrees of additional workload on the human operator. Carrying out these experiments and analyzing the initial results, has highlighted the profound complexities of designing tasks, conditions, and performance metrics which are: principled, eliminate confounding factors, and yield scientifically rigorous insights into the intricacies of a collaborative system that combines both human and robot intelligences. A key contribution of this paper is to describe the lessons learned from attempting these experiments, and to suggest a variety of guidelines for other researchers to consider when designing experiments in this context.