Subject-specific gait parameters prediction for robotic gait rehabilitation via generalized regression neural network

Gait pattern planning is important in robotic gait rehabilitation, whereby patients learned the pattern provided to them Gait pattern is related to gait parameters, such as cadence, stride length, and walking speed. Therefore, the planning of gait parameters for natural walking should be addressed in order to generate gait pattern for specific subjects. The present work utilizes generalized regression neural networks (GRNNs) to predict natural gait parameters for a given subject. The inputs of GRNNs are age, gender, body height, and body weight of the targeted subject. First of all, speed mode (normal/slow) must be chosen by the therapist. When speed mode is specified, the trained “Walking Speed” GRNN (WS-GRNN) outputs a selectable range of walking speed for a given subject. Subsequently, the therapist can select and recommend a walking speed, which will be used as an input to “Stride Length” GRNN (SL-GRNN) for the generation of stride length in the next step. Finally, cadence is calculated from walking speed and stride length. This model is easy to use to obtain gait parameters, since the therapist only needs to predefine the speed mode and select a walking speed from the range that is recommended by WS-GRNN. Results and t-test shows that outputs predicted by the GRNNs are closed to the experimental data. The efficiency and accuracy of the GRNNs are discussed in the conclusion.