EMG Control Strategy of a Cable-based Upper Limb Rehabilitation Robot and its Verification

Publication Type:
Issue Date:
Full metadata record
Post-stroke motor recovery highly affected by patients’ active participation in rehabilitation. Surface electromyography (EMG) signals are related to a subject’s intention. It is also one of the most widely used biosignals in the area of motion intention estimation and rehabilitation robot control. However, due to the complicated relationship between multiple muscles and movements, few studies have applied continuous multiple EMG signals to control rehabilitation robots and provide assistant to users’ multi-joint movements in real-time. In this dissertation, new methods for decoding EMG signals during robot-assisted movements are proposed and applied to manipulate a cable-based rehabilitation robot. Different EMG decoders for continuously estimating voluntary motion intention are developed to fulfil different rehabilitation needs. These decoders are used to establish a human-robot cooperative control scheme for promoting users’ active participation in rehabilitation. Firstly, an EMG decoder is build up with a switching mechanism and submodels for decoding EMG signals to motion need forces during a multi-joint complex task in three-dimensional space. The switching mechanism aims to carve up the task into separate simple subtasks. For each simple subtask, a linear six-input three-output time-invariant submodel is established by the state-space modelling method. The inputs are the processed muscle activations of six arm muscles, and the outputs are motion need forces of users when executing the task with visual feedback. The outputs are used to indicate three motors of the robot. The switching logic of the mechanism is to change the parameters of each submodel by times. However, we observed a ‘bump’ behaviour of the estimated forces (i.e., discontinuity) when switching parameters of two submodels. A sudden change in control signals of motors might cause unexpected impacts on patients, so it is unacceptable during rehabilitation. After that, to improve the smoothness of the estimated forces, we attempted to maintain the continuity of the decoder outputs when switching among submodels. A bumpless switching mechanism is proposed by constructing a generic multirealisation for all submodels. The generic multirealisation has a common output matrix, which helps to continuously predict the outputs. For submodels with the same order, the multirealisation is constructed by finding the common denominator matrix of the subsystems’ Matrix Fraction Description (MFD). Furthermore, the best submodel, in terms of goodness of fit, established in each simple subtask, may have a different order. For different-ordered submodels, the generic multirealisation is constructed by finding the common highest-degree-coefficient matrix and expanding the hidden states of submodels […]
Please use this identifier to cite or link to this item: