Preference Augmented Q-Learning for Patient Exercise Scheduling in a Robotic Rehabilitation Gym

Area of Science:

• Robotics in Rehabilitation
• Artificial Intelligence in Healthcare
• Human-Computer Interaction

Background:

• Robotic rehabilitation gyms facilitate group therapy sessions, supervised by a single therapist.
• Quantifying patient motivation and fatigue is challenging with current sensor technology.
• Human therapists intuitively assess these factors for optimal patient-robot assignments.

Purpose of the Study:

• To develop an automated patient-robot assignment scheduler for group robotic rehabilitation.
• To incorporate expert assignment preferences and implicitly model non-quantifiable patient states (motivation, fatigue).
• To improve the efficiency and effectiveness of group robotic rehabilitation therapy.

Main Methods:

• Proposed a Preference Augmented Q-Learning assignment scheduler.
• Trained the scheduler using simulated expert assignment preferences.
• Evaluated the scheduler's performance on 1000 simulated patient groups against baseline methods.

Main Results:

• The proposed scheduler significantly outperformed existing baseline assignment methods.
• Performance closely matched a simulated expert with complete patient information and predictive capabilities.
• Demonstrated the effectiveness of integrating expert preferences into reinforcement learning for assignment scheduling.

Conclusions:

• Preference Augmented Q-Learning is a viable approach for optimizing patient-robot assignments in group rehabilitation.
• This method successfully incorporates subjective factors like motivation and fatigue without direct measurement.
• AI-driven scheduling can enhance therapist efficiency and patient outcomes in robotic rehabilitation settings.