Variable speed trajectory prediction method of dynamic prosthesis based on improved SeqGAN

作者：Haowei Han, Honglei An, Hongxu Ma, Qing Wei · 发表于：2021 China Automation Congress (CAC) · 年份：2021 · DOI：10.1109/cac53003.2021.9727962 · 被引用次数：2 · 研究领域：Prosthetics and Rehabilitation Robotics、Muscle activation and electromyography studies、Ergonomics and Human Factors

The speed control of dynamic prosthetic needs virtual constraint to provide control standard, which is mainly provided by fitting regression algorithm at present, but there are still three problems to be solved. The fitting regression method has poor tolerance to the effective error of human motion. At the same time, it needs to not only measure the precise speed, but also collect a large amount of data at different speeds to deal with the problem of speed change. SeqGAN is usually used to deal with discrete text data. The main body of the prediction network is LSTM network, which can deal with more complex continuous sequence prediction. In this paper, SeqGAN is improved to the Feature Generative Adversarial Nets(F-SeqGAN), we collecte the data of constant speed walking joints of the experimental subjects in the discrete speed dimension, to predicte the change of knee and ankle joint angles in the next time stage by the change data of thigh joint angles in the last time stage, avoiding speed judgment and realize speed smooth variable speed trajectory prediction. Compared with Gaussian Process Regression method, it has advantages.