Human-robot teaming for real-time data foraging decisions

Robot appendage-based sensing can bring directly into the field a new generation of instrumentation that affords in situ measurements of a quality and accuracy that has heretofore only been available back at the lab. Human interfaces with appropriate real-time decision support for handling accumulating in situ measurements are needed to advance the operational efficacy - and hence the ultimate value - of the geoscience mission. Here we present a new human-robot teaming workflow for field data collection decisions, applied to the study of soil rheology in a simulated sand dune environment and along a forested hillslope in Southeastern Pennsylvania (USA).

Expert geoscientist participants repeatedly chose between various data collection objectives: to improve spatial coverage, to improve coverage of the dynamic range of a variable of interest, or to examine potential discrepancies with a hypothesis. A corresponding quantitative decision model was run that allowed a robot to suggest potential data collection locations to the geoscientist based on their reported objective. Geoscientists either accepted the suggestion and used the robot to collect data from the suggested location, or rejected the suggestion and selected a new location that better fit their objective. Overall geoscientists reported the robot suggestions were high quality and well aligned with their data collection strategies. Results also suggested that geoscientists' objectives could be reasonably well-predicted by the degree of information coverage and the degree of fit between the collected data and the hypothesis. This work will enable the development of cognitively-compatible legged robotic assistants that can infer humans' dynamic scientific objectives and aid with the real-time identification of potentially valuable data collection locations, helping to reduce human cognitive load and thereby enhance geoscience outcomes.