Behavioral Cloning of Physiotherapists in Adapting Robot Control Parameter

Area of Science:

• Robotics
• Artificial Intelligence
• Rehabilitation Medicine

Background:

• Robot-aided rehabilitation enhances upper-limb disorder treatment.
• Artificial intelligence (AI) can learn physiotherapist expertise for autonomous robot assistance modulation.
• Real-time multimodal patient data is crucial for adaptive robotic therapy.

Purpose of the Study:

• To propose and validate the Physiotherapist-Supervised Parameter Adaptation (PSPA) strategy for online tuning robot assistance.
• To replicate physiotherapists' decision-making in robot-assisted rehabilitation using behavioral cloning.
• To compare PSPA's performance in tailoring assistance with established performance-based (PB) methods.

Main Methods:

• Experimental validation in a clinical setting with ten post-surgical orthopedic patients.
• Utilized KUKA LWR 4+ robot for robot-aided rehabilitation sessions.
• Collected multimodal patient data (biomechanical and physiological) labeled by supervising physiotherapists.

Main Results:

• Combining biomechanical and physiological features significantly improved classification performance.
• The Multi-layer Perceptron classifier demonstrated the highest performance.
• PSPA showed greater alignment with physiotherapists' decisions compared to the PB approach (ΔF1-score = 15.40 ± 30.33%).

Conclusions:

• The PSPA methodology effectively replicates physiotherapist decision-making for personalized robot-assisted rehabilitation.
• Integrating multimodal features enhances the accuracy of AI-driven rehabilitation strategies.
• PSPA offers a promising approach for optimizing robot assistance in physical therapy.