Using skeletal position to estimate human error rates in telemanipulator operators.

Area of Science:

• Robotics
• Human-Computer Interaction
• Biomechanics

Background:

• Telerobotic operations carry inherent risks due to complex tasks and potential for operator error.
• Existing methods for operator monitoring lack real-time, data-driven behavioral insights.
• Improving safety in teleoperation requires advanced systems for predicting and mitigating human error.

Purpose of the Study:

• To develop and trial a non-invasive system for capturing biomechanical features of teleoperators.
• To create novel human-error rate estimators for industrial teleoperation.
• To lay the groundwork for future systems providing operator feedback to enhance safety.

Main Methods:

• Utilized 3D point-cloud data from depth cameras to estimate operator skeletal pose.
• Conducted in-situ monitoring studies with 14 operators using the MASCOT teleoperation system.
• Employed statistical and machine learning regression techniques (SVR) to estimate collision rates from captured data.

Main Results:

• Successfully captured skeletal pose and collision statistics from 8 hours of operator data.
• Demonstrated the feasibility of using data-driven analysis to estimate collision rates.
• Performed sensitivity analysis on selected input features for the regression models.

Conclusions:

• A non-invasive biomechanical feature capture method is effective for analyzing teleoperator performance.
• Data-driven approaches, particularly SVR, show promise in estimating human error rates in teleoperation.
• The developed system provides a foundation for future research into real-time operator feedback to improve safety.