Sie befinden Sich nicht im Netzwerk der Universität Paderborn. Der Zugriff auf elektronische Ressourcen ist gegebenenfalls nur via VPN oder Shibboleth (DFN-AAI) möglich. mehr Informationen...
The recent development of lower limb exoskeletons has focus on assisting walking on a variety of terrains, such as level ground, stairs, and ramps. However, smooth transitions between different locomotion modes are challenging due to unknown future information. In this paper, we propose a method that adaptively generates a complete gait cycle of trajectory and event prediction for lower limb exoskeletons during the transition stage. First, we design a generative model with the variational autoencoder (VAE) structure to parameterize the gait cycles of trajectory and events into low-dimensional latent representations. Then, the gait trajectory and event prediction are generated by VAE reconstruction, stride time and gait phase alignment. Lastly, reinforcement learning (RL) is adopted to minimize the process error of continuous gait trajectory and event prediction by finding the optimal latent representations of a generative model. The experiments collected gait data from passive and active exoskeletons. The results show that our method significantly outperforms the extend Kalman filter (EKF)-based method and moving horizon estimation (MHE)-based method for both gait trajectory and event prediction tasks during transitions between and within locomotion modes. The capability of our method gives the opportunity to enrich the control mode by providing future information and contributes a valuable tool for natural and smooth transitions among different terrains for the lower limb exoskeleton.