Human-assisting intelligent systems demand certain methods to precisely predict motorized limb joint angles. This paper presents the application of deep-recurrent neural networks (RNNs), which is a type of neural network for processing sequential data, for predicting the knee joint angle in real-time. This model is created based on a combination of electromyographic (EMG) signals, (with electrodes being placed on three leg muscles), and inertial measurements of the upper and lower legs. The data collected from different subjects when they performed different gaits were used to construct the model, which was evaluated in a real-time setting. The proposed RNN model based on fusion information contains a balance between computational complexity and prediction accuracy. Results on a microcontroller show that, within a predicted horizon of 50 ms, the model has a low prediction error of ±12.93 degrees.

Research on the lower limb exoskeleton for rehabilitation have developed rapidly to meet the need of the aging population. The rehabilitation exoskeleton system is a wearable man–machine integrated mechanical device. In recent years, the vigorous development of exoskeletal technology has brought new ideas to the rehabilitation and medical treatment of patients with motion dysfunction, which is expected to help such people complete their daily physiological activities or even reshape their motion function. The rehabilitation exoskeletons conduct assistance based on detecting intention, control algorithm, and high-performance actuators. In this paper, we review rehabilitation exoskeletons from the aspects of the overall design, driving unit, intention perception, compliant control, and efficiency validation. We discussed the complexity and coupling of the man–machine integration system, and we hope to provide a guideline when designing a rehabilitation exoskeleton system for the lower limbs of elderly and disabled patients.

This paper analyzes human motion, more specifically the human gait in the sagittal plane. A video camera is used to acquire images of a walking person, fitted with a set of white light-emitting diodes (LEDs). The acquired trajectories of the light points are then used to specify joint trajectories in a biped robot. To analyze the stability of the human gait, a system was also developed to acquire the center of pressure (CoP). This system uses eight force sensors, four under each foot. The influence of the human torso angle on the CoP position during walking was confirmed. Some experiments were carried out on a biped robot, and the results show that the acquired human gait can be used in a biped robot, after scale conversion.

The real-time balance control of an eight-link biped robot using a zero moment point (ZMP) dynamic model is difficult due to the processing time of the corresponding equations. To overcome this limitation, two alternative intelligent computing control techniques were compared: one based on support vector regression (SVR) and another based on a first-order Takagi-Sugeno-Kang (TSK)-type neural-fuzzy (NF) network. Both methods use the ZMP error and its variation as inputs and the output is the correction of the robot's torso necessary for its sagittal balance. The SVR and the NF were trained based on simulation data and their performance was verified with a real biped robot. Two performance indexes are proposed to evaluate and compare the online performance of the two control methods. The ZMP is calculated by reading four force sensors placed under each robot's foot. The gait implemented in this biped is similar to a human gait that was acquired and adapted to the robot's size. Some experiments are presented and the results show that the implemented gait combined either with the SVR controller or with the TSK NF network controller can be used to control this biped robot. The SVR and the NF controllers exhibit similar stability, but the SVR controller runs about 50 times faster.

As the aging of the population aggravates, the problem of fall has aroused more attention. We have proposed a novel pre-impact fall detection algorithm and protection system. The real-time algorithm used the thresholds of vertical velocity and displacement profile to detect fall events in switching periods of fast motion. It can effectively solve the problem of false alarms of jump and jogging for daily-life applications. A second-order lag filtering algorithm method was utilized to reduce the drift of the vertical velocity with good performance. In addition, the vertical displacement profile was calculated by using the vertical velocity estimation. The algorithm got 93.6% sensitivity and 95.6% specificity, which was higher than the algorithm without displacement with 75.6% specificity. The results demonstrated the effectiveness of the proposed detection method, which were evaluated by subjects wearing fall detection and airbag system.