Learning Skill Training Schedules From Domain Experts for a Multi-Patient Multi-Robot Rehabilitation Gym

Area of Science:

• Robotics
• Artificial Intelligence
• Rehabilitation Medicine

Background:

• Robotic gyms enable simultaneous patient rehabilitation under therapist supervision.
• Optimizing patient-robot assignments can enhance multi-patient training outcomes.
• Dynamic assignment strategies are needed to adapt to real-time patient data.

Purpose of the Study:

• To develop and evaluate a supervised learning approach for dynamic patient-robot assignment.
• To leverage domain expert knowledge for automated assignment policies.
• To improve patient skill gain in simulated robotic rehabilitation environments.

Main Methods:

• A neural network model was developed to predict patient-robot assignment priorities.
• The model was trained on a synthetic dataset simulating expert assignment behavior.
• The approach was evaluated in simulated rehabilitation gym scenarios with varying complexities.

Main Results:

• The assignment algorithm achieved high accuracy in imitating expert behavior (75.4%–84.5%).
• The proposed method significantly outperformed baseline assignment strategies in terms of mean skill gain.
• The system demonstrated effective patient training scheduling without complete knowledge of skill acquisition dynamics.

Conclusions:

• Learned dynamic assignment policies can effectively optimize patient-robot allocation in rehabilitation settings.
• Supervised learning from expert behavior provides a viable method for creating automated assignment systems.
• This approach enhances the efficiency and effectiveness of robotic-assisted physical therapy.