No abstract

Deep reinforcement learning (DRL) provides a new solution for rehabilitation robot trajectory planning in the unstructured working environment, which can bring great convenience to patients. Previous researches mainly focused on optimization strategies but ignored the construction of reward functions, which leads to low efficiency. Different from traditional sparse reward function, this paper proposes two dense reward functions. First, azimuth reward function mainly provides a global guidance and reasonable constraints in the exploration. To further improve the efficiency, a process-oriented aspiration reward function is proposed, it is capable of accelerating the exploration process and avoid locally optimal solution. Experiments show that the proposed reward functions are able to accelerate the convergence rate by 38.4% on average with the mainstream DRL methods. The mean of convergence also increases by 9.5%, and the percentage of standard deviation decreases by 21.2%–23.3%. Results show that the proposed reward functions can significantly improve learning efficiency of DRL methods, and then provide practical possibility for automatic trajectory planning of rehabilitation robot.

The portable and inexpensive hand rehabilitation robot has become a practical rehabilitation device for patients with hand dysfunction. A pneumatic rehabilitation glove with an active trigger control system is proposed, which is based on surface electromyography (sEMG) signals. It can trigger the hand movement based on the patient’s hand movement trend, which may improve the enthusiasm and efficiency of patient training. Firstly, analysis of sEMG sensor installation position on human’s arm and signal acquisition process were carried out. Then, according to the statistical law, three optimal eigenvalues of sEMG signals were selected as the follow-up neural network classification input. Using the back propagation (BP) neural network, the classifier of hand movement is established. Moreover, the mapping relationship between hand sEMG signals and hand actions is built by training and testing. Different patients choose the same optimal eigenvalues, and the calculation formula of eigenvalues’ amplitude is unique. Due to the differences among individuals, the weights and thresholds of each node in the BP neural network model corresponding to different patients are not the same. Therefore, the BP neural network model library is established, and the corresponding network is called for operation when different patients are trained. Finally, based on sEMG signal trigger, the pneumatic glove training control algorithm was proposed. The combination of the trigger signal waveform and the motion signal waveform indicates that the pneumatic rehabilitation glove is triggered to drive the patient’s hand movement. Preliminary tests have confirmed that the accuracy rate of trend recognition for hand movement is about 90%. In the future, clinical trials of patients will be conducted to prove the effectiveness of this system.