Dual-Branch Fusion Network: Precise Decoding of Lower Limb Multi-Joint Torque.

Area of Science:

• Biomedical Engineering
• Robotics
• Machine Learning

Background:

• Accurate real-time lower-limb joint torque estimation is crucial for adaptive human-exoskeleton interaction.
• Existing methods struggle with diverse locomotion and dynamic environments.

Purpose of the Study:

• To develop a novel framework for accurate, real-time lower-limb joint torque estimation across diverse locomotion conditions.
• To improve adaptive human-exoskeleton interaction through precise torque control.

Main Methods:

• A dual-branch architecture combining Temporal Convolutional Networks (TCN) and Transformers was developed.
• TCN processed local temporal dynamics, while Transformers captured global dependencies.
• A joint-specific, task-aware residual fusion mechanism with residual enhancement was employed for feature synthesis.

Main Results:

• The framework achieved high accuracy across twelve locomotion patterns with low root mean square errors (e.g., 0.1405 Nm/kg for knee) and high Pearson correlation coefficients (e.g., 0.9904 for ankle).
• Maintained a low latency of 4.2912 ms, demonstrating computational efficiency.
• Showcased strong adaptability on public datasets.

Conclusions:

• The proposed method effectively balances high estimation accuracy with the computational efficiency required for real-time applications.
• It successfully addresses limitations in adapting to dynamic environments for human-exoskeleton systems.
• This advancement provides a fast, reliable solution for adaptive exoskeleton torque control, enhancing natural human-robot interaction.