Improving Workers' Musculoskeletal Health During Human-Robot Collaboration Through Reinforcement Learning.

Area of Science:

• Robotics
• Occupational Safety
• Artificial Intelligence

Background:

• Human-robot collaboration is increasing in industrial settings.
• Awkward worker postures during collaboration can lead to musculoskeletal disorders.
• Existing methods may not adequately address dynamic posture optimization.

Purpose of the Study:

• To develop a novel model-free reinforcement learning method to improve worker postures.
• To reduce the risk of musculoskeletal disorders in human-robot collaboration.
• To create adaptive and personalized ergonomic solutions.

Main Methods:

• Utilized 3D human skeleton reconstruction to calculate continuous awkward posture (CAP) scores.
• Developed an online gradient-based reinforcement learning algorithm.
• Dynamically adjusted robot end-effector positions and orientations to improve CAP scores.

Main Results:

• Significantly improved participants' CAP scores in a collaborative task.
• Demonstrated superior performance compared to fixed or individual height robot positions.
• Participants preferred the working postures achieved with the proposed method.

Conclusions:

• The model-free reinforcement learning approach optimizes worker postures without biomechanical models.
• The data-driven method offers personalized ergonomic improvements.
• This technology enhances occupational safety in robot-implemented factories.