Assistive Force Cycle Prediction Method for Lower Limb Exoskeleton Based on Independent Attitude Processing Sensor

In order to make the exoskeleton suitable for the wearers, it is necessary to provide fitting assistive force, wherein assistive force cycle is an important parameter for generating the assistive force curve. The assistive force cycle always wants to match the gait cycle and needs to be determined at the beginning of each gait. Due to the unknown current gait cycle, the assistive force cycle is often defined as the previous gait cycle or the average of multiple previous cycles. However, this strategy is not accurate. For this reason, an assistive force cycle prediction method based on long short-term memory (LSTM) network with independent attitude processing (IAP) sensors is proposed in this article. First, the IAP sensors are utilized to collect and process the gait information to generate the input features. Meanwhile, a variable window method is proposed for detecting assistive force timing and calculating assistive force cycle. Then, the input features and corresponding assistive force cycles are transferred to the LSTM to generate the predictive model. Finally, the assistive force cycles are predicted on four gait modes (uniform walking, variable walking, upstairs, and uphill). In the validation experiment, mean square error (MSE) is selected as the evaluation index of the model prediction effect. Compared with the other traditional cycle calculation method, MSEs based on LSTM are all lower than 0.1 on the four kinds of gaits. The average accuracy of the assistive force cycle prediction by the proposed method can reach 97.23%.