Formalizing Motion Plan Legibility Using Empirical Manual Takeover Data in Autonomous Spacecraft Docking

Spacecraft rendezvous and docking maneuvers are becoming highly automated in an attempt to decrease astronaut workload, but still require a human supervisor to continually monitor the system and manually take over control when systems are not performing as expected. This requirement shifts astronaut workload to a monitoring and failure-mitigation task, which must be characterized to assess the influence of this task shift for automated rendezvous and docking (ARD). The physical performance of manual takeover maneuvers are well studied in other fields, such as automated vehicles, but less is known about the factors that influence the decision leading to takeover. This study operationalizes the concept of automation legibility (i.e., intent-expression) to gain insight into when and where supervisors initiate manual takeover. We hypothesized that fundamental aspects of autonomous agent path planning of initial condition and path curvature influence path legibility and takeover decision-making. The study had $N=33$ participants who performed an ARD monitoring task. Metrics for legibility were defined using the positions, where the human initiated a manual takeover along the ARD path. Results support that initial condition, path curvature, and autonomous agent heading were interacting predictors of path legibility and takeover decision timing and location. Increased legibility was correlated to supervisor perceptions of path-appropriateness. The most legible paths aligned the supervisor’s egocentric viewpoint to the path goal, while putting targets of avoidance in the supervisor’s periphery. Characterizations of cooperative performance in human-automation interaction systems from this study can inform future ARD system design that mitigates workload in ARD task performance.