Deep-Learning to Map a Benchmark Dataset of Non-amputee Ambulation for Controlling an Open Source Bionic Leg

Area of Science:

• Biomedical Engineering
• Robotics
• Rehabilitation Technology

Background:

• Powered lower-limb prosthetics offer a promising solution for amputation patients.
• Current control methods for prosthetic gait trajectories face implementation challenges, impacting safety and intuitiveness.
• A need exists for prosthetic control systems that closely mimic natural locomotion.

Purpose of the Study:

• To propose a novel control strategy for powered lower-limb prosthetics.
• To enable intuitive and safe prosthetic leg control by directly mapping residual limb movement.
• To reduce the need for subject-specific training and post-tuning of prosthetic control systems.

Main Methods:

• A control model was developed to directly map voluntary thigh movement to prosthetic knee and ankle impedance parameters.
• The model was trained using gait trajectory data from intact limb individuals in a public biomechanics dataset.
• The trained model was applied to control a prosthetic leg without requiring post-tuning of the neural network.

Main Results:

• The proposed method achieved intuitive and reliable level-ground walking in able-bodied subjects across various step lengths (self-selected, long, short).
• The prosthetic gait trajectory closely resembled that of the intact limb.
• The model demonstrated the feasibility of using intact limb benchmark data for prosthetic control.

Conclusions:

• Direct mapping of residual limb movement to prosthetic impedance provides an effective control strategy.
• The proposed method eliminates the need for subject-specific training and configuration time.
• This approach facilitates the development of more natural and user-friendly lower-limb prosthetic devices.