Effects of Personalization on Gait-State Tracking Performance Using Extended Kalman Filters

Area of Science:

• Robotics
• Biomechanics
• Human-Computer Interaction

Background:

• Exoskeletons offer mobility assistance but struggle with individual gait variations.
• Current gait-state estimation controllers rely on subject-independent models, leading to kinematic errors.
• Personalization of measurement models in gait-state estimation remains under-explored.

Purpose of the Study:

• To quantify the improvement in gait-state tracking accuracy using a personalized measurement model compared to a subject-independent model within an Extended Kalman Filter (EKF).
• To evaluate the impact of personalization on estimating key gait variables like phase, stride length, and ground inclination.

Main Methods:

• Utilized an Extended Kalman Filter (EKF) for online gait-state estimation.
• Compared a personalized measurement model against a subject-independent model.
• Tuned the measurement model covariance matrix across various parameters to assess performance.

Main Results:

• Personalization improved phase estimation error by an average of 8.5 ± 13.8%.
• Stride length estimation error was reduced by an average of 27.2 ± 8.1% with personalization.
• Ground inclination estimation error decreased by an average of 10.5 ± 13.5%.

Conclusions:

• Personalizing the measurement model significantly enhances the performance of EKF-based gait-state estimation.
• Improved gait-state tracking enables more reliable and adaptive robotic assistance for dynamic human locomotion.
• Findings support the development of personalized control strategies for exoskeletons to accommodate individual gait characteristics.