Enhancing Human-Robot Collaborative Predictability through Rational Action Modeling of Robot Trajectories

Abstract

The integration of collaborative robots into manufacturing environments has opened new opportunities for flexible production processes. This flexibility is often demonstrated through adaptive robot behavior, where collaborative robots, with the help of AI methods, dynamically change their trajectories in response to changes in the workspace, task demands, or actions of the human operator. While the ability of collaborative robots to adapt promises significant benefits for more efficient operations and higher process stability, it may increase the difficulty of predicting changes in robot behavior and actions from a human perspective. The seamless human-robot collaboration requires that the robot make its intentions clear and predictable to the human partner. With mathematical definitions of predictability, this paper proposes a modeling approach consistent with the principle of rational action to enable robots to generate more predictable trajectories for human partners. The proposed modeling leverages cost optimization formulations grounded in the theory of rational action to model human predictability inferences about robot trajectory. The basic assumption is that humans will view the shortest, most direct trajectory to a goal as the most predictable trajectory a robot can take. To evaluate the proposed approach, a study was conducted with 39 participants. The study compared humans’ ability to predict robot trajectory generated using the new approach with conventional trajectories not considering predictability. The results reveal an 8-18.5% enhancement for the proposed methods in enabling accurate trajectory prediction by human observers. According to the Likert scale results, the trajectories generated by optimizing predictability cost functions were perceived as more comprehensible and better aligned with participant’s anticipations compared to conventional trajectories. Consequently, the study demonstrates a marked improvement in augmenting the predictability of robot trajectory for human collaborators through the developed approach.