Bridging the Performance Gap: The Impact of Grasping Behavior and Riemannian Geometry on Prosthetic Intent...

Area of Science:

• Biomedical Engineering
• Rehabilitation Engineering
• Signal Processing

Background:

• Myoelectric prosthesis control faces challenges due to the discrepancy between offline intent recognition accuracy and online performance.
• This offline-online gap is a significant barrier to effective prosthetic limb control.

Purpose of the Study:

• To investigate the mechanisms behind the offline-online gap in myoelectric prosthesis control from a grasping behavior perspective.
• To propose a novel Riemannian geometry-based signal processing scheme to address this performance degradation.

Main Methods:

• Hypothesized feature distribution mismatch between steady-state training data and dynamic online control requirements.
• Quantified performance disparities of various training models using datasets from amputees and able-bodied subjects.
• Developed a Riemannian geometry-based scheme to enhance feature separability for improved prosthetic control.

Main Results:

• Models trained on steady-state data showed significant performance degradation (5.85%–18.16%) in dynamic phases.
• Models trained exclusively on dynamic data exhibited superior intent recognition.
• The proposed Riemannian geometry-based scheme enhanced dynamic phase performance by up to 13.82%.

Conclusions:

• The mismatch between training data characteristics and dynamic task requirements is a critical cause of the performance gap.
• The proposed scheme effectively improves decoding of dynamic signals by optimizing geometric features in Riemannian space.
• This research provides insights into the offline-online gap and offers an engineering solution for robust prosthetic systems.