Adaptive Recalibration of a Prosthetic Leg Neural Control System

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): An estimated 623,000 people were living with major lower limb amputation in the United States in 2005; this number will continue to grow due to population aging and increasing incidence of dysvascular disease. The emerging field of robotic leg prostheses provides exciting possibilities to enhance functional outcomes for these individuals; however, the ability to control of these devices must be improved. Pattern recognition algorithms may be used to decode data from mechanical sensors on the prosthesis to predict ambulation mode, and our preliminary data shows that incorporating neural control information by decoding patterns in electromyographic (EMG) signals improves accuracy. However, EMG signals vary with electrode position, skin/electrode impedance, or muscle fatigue, and it remains unclear how to incorporate these signals within a robust control system that is clinically viable for long-term use. Our long-term goal is to create robust, intuitive, and generalizable control systems for lower-limb prostheses. Our objective in the proposed research is to design and test an adaptive framework-that can compensate for changes in residual limb EMG signals-to control a powered knee and ankle prosthesis. Our central hypothesis, based on preliminary data, is that adaptation of a neural control system may be supervised using mechanical sensor data and a priori gait profile information to more accurately predict ambulation mode. The rationale is that EMG signals provide important neural information to the control system, and that accounting for non-stationary behavior of EMG signals over time will improve system robustness. We will test our hypothesis through the following three specific aims: (1) Develop a gait-pattern estimator to robustly label prior ambulation modes following correct or incorrect control system predictions; (2) Identify an effective method to update the pattern recognition control system for prediction of ambulation modes; and (3) Evaluate a real-time adaptive neural control system in 12 transfemoral amputees. Under Aim 1, we will develop a system that accurately estimates what mode (e.g., walking, stair climbing) the user was operating within during the previous stride. This data will be used as an 'expert' to provide a label to supervise an online adaptive control system. Under Aim 2, the improvement in control accuracy provided by supervised adaptation will be compared to that provided by unsupervised adaptation. Under Aim 3, the adaptive system will be translated to a real-time embedded system and tested by 12 transfemoral amputees. This proposal provides an innovative approach to incorporating neural control information and removes a critical barrier to using EMG signals to improve control of lower limb prostheses. The proposed research is significant because it will result in a robust control system that will allow more intuitive control of powered leg prostheses. This will in turn facilitate use of these devices and improve mobility for tens of thousands of people. This technology may also be translated to improve control of powered exoskeletons-another important emerging field of research.