Human-centric predictive model of task difficulty for human-in-the-loop control tasks.

Area of Science:

• Human-Computer Interaction
• Robotics
• Control Systems Engineering

Background:

• Quantifying manipulation task difficulty in human-in-the-loop systems is challenging.
• Current evaluation methods (performance metrics, user surveys) lack real-time user experience data.

Purpose of the Study:

• To analyze and predict the difficulty of a bivariate pointing task using human-centric measurements.
• To develop and compare data-driven models for real-time task difficulty prediction.

Main Methods:

• Utilized noninvasive sensors to record multimodal human user responses (cognition, physical effort, motion kinematics) for 14 subjects.
• Implemented and compared four models: movement time, fusion (physiological + kinematic), kinematic-only, and physiological-only.
• Employed a data-driven approach using task-independent metrics.

Main Results:

• Significant correlation found between predicted task difficulty and user sensorimotor response.
• The fusion model (physiological + kinematic) achieved the highest accuracy (R2 = 0.927).
• Kinematic-only model (R2 = 0.921) and fusion model outperformed the traditional movement time model (R2 = 0.847).

Conclusions:

• Real-time human-centric data (physiology, kinematics) can effectively predict manipulation task difficulty.
• A fusion model integrating physiological and kinematic metrics offers the most accurate prediction.
• This approach provides a more comprehensive understanding of user experience in human-in-the-loop systems than traditional methods.