Human adaptation to latency in teleoperated multi-robot human-agent search and rescue teams

Abstract

Abstract Teleoperation of unmanned vehicles in high stress environments has been a subject of research in many domains, which focus primarily on system and operator performance. Unmanned ground vehicles for rescue, also known as search and rescue robots, serve as extensions of responders in a disaster, providing real-time video and other relevant information about the situation. However, physically separating responder and robot introduces latency between the human input provided to the unmanned vehicle to execute an operation and the subsequent response provided by the system. This latency (lag or time delay) is determined by the distance and the bandwidth of the connection between the operator and the unmanned vehicle. Automating these systems may mitigate the effect of latency to an extent; however, this has its own consequences, such as leaving the responder out of the loop, which subsequently leads to detrimental effects on situational awareness. This research investigates the relationship between latency and the performance of the human operator of a teleoperated robot at different levels of system complexity and the effect of different levels of automation on this relationship. Eighty participants operated one or two unmanned teleoperated robots to complete two search and rescue tasks. The study utilized a 2 × 2 × 2 mixed-subjects experimental design with the automation level and latency level being the between-subjects factors and the system complexity (controlling one or two robots) being the within-subjects factor. The dependent variables were operator performance, perceived workload, and the subjective rating of trust with automation. A latency of 500 ms showed a significant decrease in performance in time to complete the task and a significant increase in the perceived physical workload. Both the automation level and latency level moderated the system complexity effect on the subjective rating of trust in the robotic system. The level of trust decreased over time in the one-robot condition as opposed to no change in the two-robot condition. The error rate decreased over time at different rates based on the number of robots or the latency level. Based on the results of the study, several design implications are suggested for improving performance including adding features to the automation that will allow the operator to use common strategies and providing necessary information using multiple sensory channels. Future research directions are also proposed.