Influencing Behavioral Attributions to Robot Motion During Task Execution

N. Walker, C. Mavrogiannis, S. Srinivasa, and M. Cakmak, “Influencing Behavioral Attributions to Robot Motion During Task Execution,” Jun. 2021.

Abstract

Recent literature has proposed algorithms for autonomous generation of robot motion that communicates functional attributes of a robot’s state such as intent or incapability. However, less is known about how to automate the generation of motion for communicating higher-level behavioral attributes such as curiosity or competence. In this work, we consider a coverage task resembling robot vacuum cleaning in a household. Through a virtual interface, we collect a dataset of human attributions to robot trajectories during task execution and extract a probabilistic model that maps robot trajectories to human attributions. We then incorporate this model into an trajectory generation mechanism that balances between task completion and communication of a desired behavioral attribution. Through an online user study on a different household layout, we find that our prediction model accurately captures human attribution for coverage tasks. Further, our generation mechanism produces trajectories that are thematically consistent, but more research is required to understand how to balance attribution and task performance.

BibTeX Entry

@inproceedings{walker2021influencing,
  title = {Influencing Behavioral Attributions to Robot Motion During Task Execution},
  author = {Walker, Nick and Mavrogiannis, Christoforos and Srinivasa, Siddhartha and Cakmak, Maya},
  booktitle = {Proceedings of the 2021 ICRA Workshop on Modern Approaches for Intrinsically-Motivated Intelligent Behavior},
  location = {Xi'an, China},
  month = jun,
  year = {2021},
  wwwtype = {workshop},
  wwwpdf = {https://hcrlab.cs.washington.edu/assets/pdfs/2021/walker2021influencing.pdf}
}