Earlier this semester, Chien-Ming Huang, a principal investigator in the Johns Hopkins Institute for Assured Autonomy, and his group presented a paper at the 2022 Institute of Electrical and Electronics Engineers Robotics and Automation Society International Conference on Intelligent Robots and Systems (IROS). The title of the paper is “Modeling Human Response to Robot Errors for Timely Error Detection.”

Meet the Research Team/Authors:

  • Maia Stiber, CS Department Fellow, Computer Science
  • Russell Taylor, John C. Malone Professor, Computer Science
  • Chien-Ming Huang, John C. Malone Assistant Professor, Computer Science

Research Overview

Errors made by robots are unavoidable when it comes to collaborations between humans and robots, however, these mistakes can have significant impacts. When robots make errors, it often leads to damaging the trust that users have in the robots. Through this research, Stiber, Taylor, and Huang studied the natural social responses that people had to robots when the robots made errors. The responses that humans have are called social signals. The researchers found that these organic social responses effectively signaled timely detection and identification of robot errors. In the process of conducting this research, the paper authors found that their approach is applicable across a wide range of tasks, errors, and responses. The data collected during the study was eventually used to train a machine learning model to detect and pinpoint specific errors made by robots.

Broader Impact of Work

By understanding the natural social reactions that people have to errors made by robots, errors during interactions between humans and robots can be better detected and identified. A unique element of this research is that the methods can be applied across a variety of task contexts, robot errors, and user responses. Using action units (AUs), individual muscular movements of the face defined by the Facial Action Coding System, the research team found that it is possible to detect and localize errors reasonably accurately and efficiently.

Research Highlights

  • Showed that it is possible to detect robot errors using social signals in non-social physical interaction scenarios.
  • Developed real-time detection system to catch robot errors, with human collaboration a key part of the process.
  • Indicates effectiveness of the real-time detection system beyond training settings and generalizes in various contexts
    and with different errors.

Research Image:

Photo courtesy of Maia Stiber.