Predicting Subjective Usability from Kinematic Data in IMU-Based Robotic Teleoperation

Area of Science:

• Robotics and Human-Computer Interaction
• Biomechanics and Motion Analysis

Background:

• Robotic teleoperation is crucial for remote tasks like surgery and inspection.
• Current operator assessment relies on subjective, post-task questionnaires, limiting real-time feedback.
• Wearable Inertial Measurement Unit (IMU) chains offer potential for implicit operator state sensing.

Purpose of the Study:

• To investigate if end-effector kinematics from IMU data can predict operator workload and user experience dimensions.
• To assess the feasibility of inferring subjective states from objective kinematic data in teleoperation.
• To establish an offline proof of concept for real-time adaptive teleoperation systems.

Main Methods:

• Secondary analysis of a public dataset (16 participants, 144 recordings, simulated UR10e arm).
• Three-stage pipeline: bivariate correlation, multivariate regression (10 models, 16 feature sets, LOO validation), and binary classification.
• Statistical validity confirmed via 1000-permutation nested testing; SHAP analysis for feature importance.

Main Results:

• Regression models achieved R2 ≥ 0.50 on 7/10 subjective dimensions (peak R2=0.787 for usability).
• Permutation testing confirmed statistical significance for 8/10 dimensions.
• Binary classification achieved AUC ≥ 0.75 on 9/10 dimensions, with 3 reaching perfect AUC.

Conclusions:

• End-effector kinematics contain significant information about operator experience and workload.
• Kinematics-based inference is feasible, supporting the development of adaptive teleoperation.
• This work provides a foundation for real-time, adaptive human-robot interaction systems.