Locomotion prediction based upon data-driven classification of intrinsically driven transitions

Accurately predicting the locomotion state of a person’s next step is critical for control of multifunctional prosthetic devices. The output from the Beta Process Auto Regressive Hidden Markov Model (BP-AR-HMM) provides consistent and concise state output when classifying patterns throughout a gait cycle in non-steady state locomotion. In this study the state outputs from the BP-AR-HMM were used in a Long Short Term Memory network (LSTM) to estimate whether the next foot contact would be a walking or a running step. Results indicate that the LSTM can accurately predict the next step using minimal sensor data in non-steady state locomotion.