Adaptive Locomotion Transition Recognition With Wearable Sensors for Lower Limb Robotic Prosthesis

Locomotion transition recognition with external disturbances is a key issue in lower limb robotic prostheses. Redonning the prosthetic socket and individual differences are encountered frequently in daily life. They change the feature distribution and further fail the previously trained recognition model. To bridge the technical gap between laboratory validation and practical demands, researchers need to construct an adaptive recognition method with high accuracy, quick response, time-efficient calibration, and minimal interference to the human body. Our study developed an adaptive recognition algorithm based on two miniaturized inertial sensors (anterior thigh and forefoot) and a foot pressure sensor, which can easily be integrated into the prosthesis. The algorithm fused probability-based fuzzy classifiers, which took heuristic-based features as inputs with a dynamic-time-warping-based automatic template generation block. It enables the recognition model to quickly fit the parameters for an untrained mode after external disturbances. We validated the proposed adaptive recognition method on 13 healthy subjects, performing 18 motion transitions with random interday intervals and across user evaluation, both offline and online. Even with the data of one subject as the prior knowledge, the other subjects produced an average accuracy of 98.04% and an average transition time latency of −180.99 ms without manual training in interday and intersubject uses. We also tested the method on two subjects using robotic prostheses, one person with a transfemoral amputation and the other person with a transtibial amputation. The average accuracy was still as high as 98.06% (transfemoral) and 100% (transtibial) without manual calibration after seven days of intervals. The results proved that the adaptive recognition method is robust to sensor redonning and user changes. Compared with the state-of-the-art methods, our study makes a further step toward the practical use of robotic prostheses. Future studies will be conducted to implement the algorithm on the robotic prosthesis for more extensive tests.