A hallmark of human locomotion is that it continuously adapts to changes in the environment and predictively adjusts to changes in the terrain, both of which are major challenges to lower limb amputees due to the limitations in prostheses and control algorithms. Here, the ability of a single-network nonlinear autoregressive model to continuously predict future ankle kinematics and kinetics simultaneously across ambulation conditions using lower limb surface electromyography (EMG) signals was examined. Ankle plantarflexor and dorsiflexor EMG from ten healthy young adults were mapped to normal ranges of ankle angle and ankle moment during level overground walking, stair ascent, and stair descent, including transitions between terrains (i.e., transitions to/from staircase). Prediction performance was characterized as a function of the time between current EMG/angle/moment inputs and future angle/moment model predictions (prediction interval), the number of past EMG/angle/moment input values over time (sampling window), and the number of units in the network hidden layer that minimized error between experimentally measured values (targets) and model predictions of ankle angle and moment. Ankle angle and moment predictions were robust across ambulation conditions with root mean squared errors less than 1° and 0.04 Nm/kg, respectively, and cross-correlations (R2) greater than 0.99 for prediction intervals of 58 ms. Model predictions at critical points of trip-related fall risk fell within the variability of the ankle angle and moment targets (Benjamini-Hochberg adjusted p > 0.065). EMG contribution to ankle angle and moment predictions occurred consistently across ambulation conditions and model outputs. EMG signals had the greatest impact on noncyclic regions of gait such as double limb support, transitions between terrains, and around plantarflexion and moment peaks. The use of natural muscle activation patterns to continuously predict variations in normal gait and the model’s predictive capabilities to counteract electromechanical inherent delays suggest that this approach could provide robust and intuitive user-driven real-time control of a wide variety of lower limb robotic devices, including active powered ankle-foot prostheses.

No abstract

Background Despite performance improvements in active lower limb prostheses, there remains a need for control techniques that incorporate direct user intent (e.g., myoelectric control) to limit the physical and cognitive demands and provide continuous, natural gait across terrains. Methods The ability of a nonlinear autoregressive neural network with exogenous inputs (NARX) to continuously predict future (up to 142 ms ahead of time) ankle angle and moment of three transtibial amputees was examined across ambulation conditions (level overground walking, stair ascent, and stair descent) and terrain transitions. Within-socket residual EMG of the prosthetic side, in conjunction with sound-limb shank velocity, were used as inputs to the single-network NARX model to predict sound-limb ankle dynamics. By overlaying the ankle dynamics of the sound limb onto the prosthesis, the approach is a step forward to establish a more normal gait by creating symmetric gait patterns. The NARX model was trained and tested as a closed-loop network (model predictions fed back as recurrent inputs, rather than error-free targets) to ensure accuracy and stability when implemented in a feedback control system. Results Ankle angle and moment predictions of amputee models were accurate across ambulation conditions and terrain transitions with root-mean-square errors (RMSE) less than 3.7 degrees and 0.22 Nm/kg, respectively, and cross-correlations (R2) greater than 0.89 and 0.93, respectively, for predictions 58 ms ahead of time. The closed-loop NARX model had similar performance when characterizing normal ranges of ankle dynamics across able-bodied participants (n = 6; RMSEθ < 2.7°, R2θ > 0.95, RMSEM < 0.11 Nm/kg, R2M > 0.98 for predictions 58 ms ahead of time). Model performance was stable across a range of different EMG profiles, leveraging both EMG and shank velocity inputs for the prediction of ankle dynamics across ambulation conditions. Conclusions The use of natural, yet altered in amputees, muscle activity with information about limb state, coupled with the closed-loop predictive design, could provide intuitive user-driven and robust control by counteracting delays and proactively modifying gait in response to observed changes in terrain. The model takes an important step toward continuous real-time feedback control of active ankle-foot prostheses and robotic devices.