Neural Network Extraction of the Crutch–Leg Synergy to Estimate the Intended Motion for Paraplegic users of Rehabilitation Exoskeletons

An appropriate trajectory generation method is a key factor in the successful performance of rehabilitation exoskeleton robots, which are increasingly used to restore movement for paraplegic people. In order to improve user comfort and provide more natural gaits, recent researches in this area are focused on the online adaptation of the reference joint trajectories, according to the user’s intended motion. Considering the existing synergy between the upper- and lower-body motions, this paper proposes to use neural networks to generate the hip and knee trajectories for the next cycle, according to the crutch trajectory in the current cycle. In this way, the user’s intention will be involved in the online trajectory generation by simply changing the crutch motion. In this work, the training data for the neural network is obtained by capturing the kinematic data of healthy subjects and their crutches. The subjects were asked to wear specially designed shoes to simulate the paraplegic motion better. A computer vision technique is also proposed to combine marker-based and markerless algorithms to extract more reliable data from the videos, recorded by ordinary phone cameras. The results show that the proposed method successfully predicts the joints angles with an MSE of 1.9% for the knee angle is and an MSE of 1.2% for the thigh angle. The results also verify that the proposed method and the trained networks can readily be used for experimental tests with rehabilitation exoskeleton robots, where the crutch angle may be obtained by computer vision, inertial measurement units, or other applicable sensors.