Deep-Learning-Based Emergency Stop Prediction for Robotic Lower-Limb Rehabilitation Training Systems

Area of Science:

• Robotics
• Rehabilitation Engineering
• Machine Learning

Background:

• Robotic lower-limb rehabilitation offers advantages like intensive training and quantitative assessment.
• Frequent emergency stops disrupt training, wasting therapist and patient time and resources.
• Early detection of these stops is crucial for effective robotic rehabilitation.

Purpose of the Study:

• To develop a novel deep-learning-based technique for early and accurate detection of emergency stops in robotic rehabilitation training.
• To improve the efficiency and reduce resource wastage in robotic lower-limb rehabilitation.

Main Methods:

• A bidirectional long short-term memory (BiLSTM) prediction model was trained on normal joint data from a robotic training system.
• A real-time threshold-based algorithm utilizing cumulative error was developed for emergency stop detection.
• The model was evaluated for its precision, recall, and F1 score.

Main Results:

• The proposed deep learning technique achieved high performance metrics: precision of 0.94, recall of 0.93, and F1 score of 0.93.
• The prediction model demonstrated robustness against variations in measurement noise.
• Early detection of emergency stops was successfully achieved.

Conclusions:

• The developed deep learning method effectively detects emergency stops in robotic lower-limb rehabilitation training.
• This technique enhances the efficiency and reliability of robotic rehabilitation systems.
• The approach offers a promising solution for minimizing disruptions and optimizing patient recovery.