Real-Time Continuous Locomotion Mode Recognition and Transition Prediction for Human With Lower Limb Exoskeleton.

Area of Science:

• Robotics
• Biomechanics
• Machine Learning

Background:

• Real-time locomotion mode recognition and transition detection are crucial for safe and effective exoskeleton robot operation.
• Existing methods often struggle with accuracy and timely prediction, limiting exoskeleton adaptability.

Purpose of the Study:

• To develop an innovative framework for precise locomotion mode recognition and transition prediction using only exoskeleton-mounted inertial measurement unit (IMU) signals.
• To enhance the reliability of transition detection and reduce misjudgments.

Main Methods:

• A Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM) model was developed for classification.
• A novel majority filter was implemented to decrease the transition misjudgment rate.
• Locomotion data was collected from six subjects using a rigid exoskeleton with six IMU sensors.

Main Results:

• The framework achieved an average recognition accuracy of 99.58% for five steady locomotion modes (level ground walking, stair/ramp ascent/descent).
• Transitions were recognizable, with an average prediction time of 353 ms, and the majority filter reduced misjudgment rate by 87.04%.
• The model demonstrated real-time performance when tested on a Jetson Nano.

Conclusions:

• The proposed system provides precise locomotion mode recognition and timely transition prediction with high real-time performance.
• The CNN-BiLSTM model combined with a majority filter significantly improves exoskeleton control reliability.
• This framework offers a robust solution for enhancing human-exoskeleton interaction and adaptability.