Anytime Trust Rating Dynamics in a Human-Robot Interaction Task

TL;DR

This study investigates when people rate their trust in a human-robot collaboration by allowing an anytime, single-item trust slider to monitor trust during a remote robot task modeled after an ISS environment. It adopts a dual modeling approach: a decision-making framework for latent trust evolution and a survival-analysis framework to model the timing of trust ratings, with the latter incorporating time-varying covariates such as grasp completion. The results show that while trust evolves with robot performance, the timing of trust ratings is mainly driven by the state and progression of the task, with impending task success increasing the likelihood of rating sooner. The findings support using unobtrusive, one-dimensional trust probes to capture momentary trust in dynamic HRI contexts, and they offer a practical basis for operationalizing anytime trust in future cooperative robotics systems.

Abstract

Objective We model factors contributing to rating timing for a single-dimensional, any-time trust in robotics measure. Background Many studies view trust as a slow-changing value after subjects complete a trial or at regular intervals. Trust is a multifaceted concept that can be measured simultaneously with a human-robot interaction. Method 65 subjects commanded a remote robot arm in a simulated space station. The robot picked and placed stowage commanded by the subject, but the robot's performance varied from trial to trial. Subjects rated their trust on a non-obtrusive trust slider at any time throughout the experiment. Results A Cox Proportional Hazards Model described the time it took subjects to rate their trust in the robot. A retrospective survey indicated that subjects based their trust on the robot's performance or outcome of the task. Strong covariates representing the task's state reflected this in the model. Conclusion Trust and robot task performance contributed little to the timing of the trust rating. The subjects' exit survey responses aligned with the assumption that the robot's task progress was the main reason for the timing of their trust rating. Application Measuring trust in a human-robot interaction task should take as little attention away from the task as possible. This trust rating technique lays the groundwork for single-dimensional trust queries that probe estimated human action.

Figures (8)

• Figure 1: A rendering of the Gazebo environment where the experiment took place. The subject interacted with the robot through a simulated camera and remote user interface. The simulated UR5e robot arm picked stowage blocks from a container and placed them on the rack in a specified configuration.
• Figure 2: A visual of what subjects see during training. The colored squares correspond to stowage spawned in the box below the rack. Each position shown in Instruction and Robot Command, indicated by number, corresponds to a position shown on the rack. Subjects choose stowage from the Intake, then choose a position in Robot Command to indicate where they want the stowage placed on the rack. The algorithm (Gamma or Echo) is indicated above the robot simulation live feed. The trust rating slider is accessible below the live feed. The markup on this image appeared only for instruction.
• Figure 3: Each included subject completes ten trials of 4 grasps each. An algorithm, Echo or Gamma, is assigned randomly for each trial. A grasp is defined as the moment the gripper closes on a piece of stowage to the next time it closes on stowage. This way, trust ratings are associated with the prior grasp. Subjects may adjust their trust rating at any time.
• Figure 4: This comparative box plot shows the change in trust averaged over every grasp in which they changed trust. The capabilities, or algorithms, are shown in different colors where the Gamma algorithm always places stowage correctly, and the Echo algorithm misplaces stowage 50% of the time. This analysis treats every grasp as an independent event.
• Figure 5: Distribution of occurrence of subjects' trust ratings by grasp number. Most ratings occurred during the final grasp.