Predicting reactive stepping in response to perturbations by using a classification approach.

BACKGROUND: People use various strategies to maintain balance, such as taking a reactive step or rotating the upper body. To gain insight in human balance control, it is useful to know what makes people switch from one strategy to another. In previous studies the transition from a non-stepping balance response to reactive stepping was often described by an (extended) inverted pendulum model using a limited number of features. The goal of this study is to predict whether people will take a reactive step to recover from a push and to investigate what features are most relevant for that prediction by using a data-driven approach. METHODS: Ten subjects participated in an experiment in which they received forward pushes to which they had to respond naturally with or without stepping. The collected kinematic and center of pressure data were used to train several classification algorithms to predict reactive stepping. The classification algorithms that performed best were used to determine the most important features through recursive feature elimination. RESULTS: The neural networks performed better than the other classification algorithms. The prediction accuracy depended on the length of the observation time window: the longer the allowed time between the push and the prediction, the higher the accuracy. Using a neural network with one hidden layer and eight neurons, and a feature set consisting of various kinematic and center of pressure related features, an accuracy of 0.91 was obtained for predictions made up until the moment of step leg unloading, in combination with a sensitivity of 0.79 and a specificity 0.97. The most important features were the acceleration and velocity of the center of mass, and the position of the cervical joint center. CONCLUSION: Using our classification-based method the occurrence of reactive stepping could be predicted with a high accuracy, higher than previous methods for predicting natural reactive stepping. The feature set used for that prediction was different from the ones reported in other step prediction studies. Given the high step prediction performance, our method has the potential to be used for triggering reactive stepping in balance controllers of bipedal robots (e.g. exoskeletons).

Automatic versus manual tuning of robot-assisted gait training in people with neurological disorders.

BACKGROUND: In clinical practice, therapists choose the amount of assistance for robot-assisted training. This can result in outcomes that are influenced by subjective decisions and tuning of training parameters can be time-consuming. Therefore, various algorithms to automatically tune the assistance have been developed. However, the assistance applied by these algorithms has not been directly compared to manually-tuned assistance yet. In this study, we focused on subtask-based assistance and compared automatically-tuned (AT) robotic assistance with manually-tuned (MT) robotic assistance. METHODS: Ten people with neurological disorders (six stroke, four spinal cord injury) walked in the LOPES II gait trainer with AT and MT assistance. In both cases, assistance was adjusted separately for various subtasks of walking (in this study defined as control of: weight shift, lateral foot placement, trailing and leading limb angle, prepositioning, stability during stance, foot clearance). For the MT approach, robotic assistance was tuned by an experienced therapist and for the AT approach an algorithm that adjusted the assistance based on performances for the different subtasks was used. Time needed to tune the assistance, assistance levels and deviations from reference trajectories were compared between both approaches. In addition, participants evaluated safety, comfort, effect and amount of assistance for the AT and MT approach. RESULTS: For the AT algorithm, stable assistance levels were reached quicker than for the MT approach. Considerable differences in the assistance per subtask provided by the two approaches were found. The amount of assistance was more often higher for the MT approach than for the AT approach. Despite this, the largest deviations from the reference trajectories were found for the MT algorithm. Participants did not clearly prefer one approach over the other regarding safety, comfort, effect and amount of assistance. CONCLUSION: Automatic tuning had the following advantages compared to manual tuning: quicker tuning of the assistance, lower assistance levels, separate tuning of each subtask and good performance for all subtasks. Future clinical trials need to show whether these apparent advantages result in better clinical outcomes.