Related Experiment Video
Updated: Aug 13, 2026
11:06
A Human-machine-interface Integrating Low-cost Sensors with a Neuromuscular Electrical Stimulation System for Post-stroke Balance Rehabilitation
Published on: April 12, 2016
Real-Time Balancing of Stability and Plasticity in Continual Learning Enables Adaptive Speed Estimation for
Cole B Johnson1, Jairo Maldonado-Contreras2, Kinsey R Herrin3
1School of Computer Science, Georgia Institute of Technology, Atlanta, GA 30332 USA.
IEEE Transactions on Medical Robotics and Bionics
|August 12, 2026
Summary
This study introduces a new framework for continual learning (CL) in prosthetics, improving knowledge retention and adaptation. The novel approach enhances speed estimation for transfemoral prosthesis users, showing significant gains in stability and plasticity.
Area of Science:
- Robotics
- Machine Learning
- Biomedical Engineering
Background:
- Continual learning (CL) in wearable robotics, particularly prosthetics, faces a stability-plasticity dilemma.
- Effective online adaptation requires balancing knowledge retention with new information assimilation.
Purpose of the Study:
- To present a novel online optimizer-based framework for managing the stability-plasticity balance in CL.
- To apply this framework to speed estimation for transfemoral prosthesis (TFA) users.
Main Methods:
- Developed an optimizer-based framework utilizing strategic datapoint replay and learning-rate adjustments.
- Validated the framework through offline tests with 10 TFA users and online tests with 3 TFA and 6 able-bodied (AB) participants.
Main Results:
- Offline validation showed a 39.2% increase in stability and 35.2% boost in plasticity over traditional CL.
- Online trials with AB participants demonstrated significant gains in handling various walking speeds.
- TFA trials improved baseline model plasticity by 67.45% and traditional CL stability by 31.36%, reducing estimation error by 19.47%.
Conclusions:
- The proposed framework effectively manages the stability-plasticity trade-off in continual learning for wearable robotics.
- This approach significantly enhances speed estimation accuracy and adaptability in transfemoral prosthesis users.
