Abstract:

The robot technologies have a lot of progress recently. In the near future, robots will appear much more frequently in daily human life, such as in schools, hospitals, offices, museums, and the home. In order to assist and interact with humans, the robots need to understand the humans’ intentions and respond accordingly.

Therefore, the aim of this dissertation is to develop human intention estimation methods using electromyography (EMG), electroencephalography (EEG), and dynamic information; and to design an elastic and variable stiffness mechanism with intrinsic safety, and advanced control for assistive and rehabilitation exercises. We use a muscle model to find the relationship between EMG and torque. Moreover, an exoskeleton graphical model is proposed to merge the EMG signal and dynamic information, and enhance the stabilization of the robot-human system. For extremely weak patients, independent component analysis-multiple kernel learning (ICA-MKL) is also proposed to increase the classification accuracy of the EEG human intention estimation. In order to satisfy the demand of the assistive and rehabilitation exercises, two different kinds of mechanism approaches are proposed. Backdrivable torsion spring actuators (BTSA) are recommended because they are light weight, compact, and mainly used in assistive control. Considering their performance and safety, variable stiffness and coupled elastic actuation are designed to adaptively change for different tasks. Finally, intelligent and safety controls are also developed to help humans in assistive exercises, rehabilitation, and walking, and guarantee safety under variable environments.