Across-user adaptation for a powered lower limb prosthesis.

Area of Science:

• Biomedical Engineering
• Robotics
• Artificial Intelligence

Background:

• Pattern recognition algorithms enhance control of powered lower limb prostheses by detecting user intent for seamless mode transitions.
• Current algorithms require extensive, time-consuming subject-specific data, limiting their generalization across users.

Purpose of the Study:

• To develop and evaluate an adaptive pattern recognition algorithm capable of automatically learning subject-specific data from new users during ambulation.
• To improve the generalization and reduce the error rates of pattern recognition algorithms for powered lower limb prostheses.

Main Methods:

• An adaptive pattern recognition algorithm was developed to learn new subject-specific data during ambulation.
• The algorithm was tested on a transfemoral amputee using a powered knee-ankle prosthesis.
• Initial training data came from two other transfemoral amputee subjects.

Main Results:

• Before adaptation, the algorithm, trained on data from other users, exhibited high error rates (4.21%) hindering continuous ambulation.
• After adaptation, error rates significantly decreased to 1.25%.
• The adaptive algorithm enabled the novel user to successfully complete all locomotion mode transitions.

Conclusions:

• Adaptive pattern recognition algorithms can effectively learn subject-specific data from novel users during ambulation.
• Adaptation significantly reduces error rates, improving the performance and usability of powered lower limb prostheses.
• This approach enhances the generalization of pattern recognition algorithms, making them more practical for widespread clinical use.