Trust in an Autonomous Guidance System for a Planetary Rover Task

Crewed deep space missions will increasingly rely on collaborations between astronauts and autonomous systems. Autonomous intelligent systems on board the spacecraft may perform some tasks traditionally done by ground control (e.g, planning, information transfer, decision support) without any communication delays. The human crew should be able to establish a functional level of trust in such systems. In order to assess trust in autonomous systems, we performed an experiment using a task that required participants to drive a lunar rover to a series of set locations. Subjects had to navigate the rover around terrain obstacles without getting stuck or running out of battery. Subjects were provided navigational assistance and trust in that guidance was measured via self-reported trust in the guidance in a post experiment questionnaire. A 2×2 between-subjects experimental design was used: 1) The primary independent variable was the source of the navigational assistance, either an autonomous system or a human (analogous to a ground controller). 2) To isolate subjective perceptions of the two systems from the objective guidance quality, the second independent variable was the source of the guidance as briefed to the subject (perceived autonomous vs. perceived human). The autonomous guidance used a Markov decision process based upon an accurate, but reduced resolution, hazard map. When human guidance was provided, the same conducive research assistant sat in the adjacent room (mimicking the role of a ground controller) and provided directions with a thumb-stick while using the same hazard map. Additionally, we tested control subjects who were provided no guidance and could be compared against both guidance systems. Results found that perceived guidance source did not change subject self-reported trust or task performance. These results suggest that, in terms of self-assessment and task performance, subjects are agnostic to guidance source and type on this task.