Design of a Bio-Inspired Gait Phase Decoder Based on Temporal Convolution Network Architecture With Contralateral...

Area of Science:

• Biomedical Engineering
• Robotics
• Neuroscience

Background:

• Inter-leg coordination is crucial for prosthetic wearer safety, particularly for those with high-level amputations.
• Current lower-limb prosthesis controllers rely on past signals, leading to delayed responses and increased fall risks.

Purpose of the Study:

• To propose a bio-inspired gait pattern generation architecture for improved bilateral coordination in lower-limb prostheses.
• To enhance prosthetic leg control by integrating motion intention and kinematic data.

Main Methods:

• Developed an artificial movement pattern generator (MPG) utilizing a temporal convolution network.
• Fused surface electromyography (sEMG) from the impaired leg with prosthetic leg kinematic data.
• The MPG predicts four sub-gait phases, decoding motion intention and status.

Main Results:

• The gait phase decoder demonstrated high intra-subject consistency.
• Decoding accuracy ranged from 89.27% to 91.16% across various walking speeds.
• Achieved 90.30% accuracy in estimating prosthetic leg gait phase for a hip disarticulation amputee.

Conclusions:

• The proposed architecture offers a viable solution for bilateral coordination issues in lower-limb prostheses.
• This bio-inspired approach enhances walking coordination for amputees, especially those with hip-level amputations.