Research On The Perception Of Body Motion Trend Based On The Solutionand Prediction Of Lower Extremity Joint Torques

As we all knows, wearer’s move intent information is necessary for controlling the exoskeleton robot so as to provide the fore assistance to human. Analysis on human biological signals(e.g. EEG, EMG, and so on) is a method to exactly perceive human movement trend intention, but it must be attached to the skin surface and is easy to fall and shift. Therefore, this method is not stable and can’t be used reliably in the complex environment. However, through displacement sensors, acceleration sensors and gyroscopes set up in exoskeleton robot to detect the motion intention of wearer, this is simple, stable and highly reliable approach. If these move signals are used to detect the human move intent with characteristic algorithm, there is no doubt that this will provide a favorable condition for practical applications. In this paper, an algorithm on human dynamic model is used to obtain the joint torque. Then, move periodic properties of lower limb are also considered. We proposed an algorithm which can predict change trend of wearer’s joint torque. Finally, the prediction of move intent is realized.Firstly, move signals are measured from exoskeleton robot to obtain the wearer’s joint torque. A simple lower limb model is established, which is used to analyze the kinematic and dynamic model, and obtain the current joint torque information. This is the base of our prediction algorithm. In research, we find that joint torque change is ahead of joint angle change. This implies joint torque change of upper limb can be used to predict the joint move intent.Secondly, to predict the wearer move intent exactly, a joint move information data set contained typical gait and joint torque information under recent gait is constructed. Then, joint torque of current gait is compare with gait in data set. According to the comparison result, the current gait is divided into stable gait and variability gait. In stable gait move, the similar gait torque in data set is used to be the prediction of current gait. In variability gait, an evaluation function of variability is established to evaluate the variability of gait. For small variation move, reflective algorithm adjusts the data set. This used to meet the needs of similarity and predict the move intent on the way of stable gait prediction. For the large variation move, short-term time-series forecasting model is used to predict the needing joint torque of wearer.Finally, to confirm the efficiency of our algorithm, the algorithm is applied to our exoskeleton robot experimental system. In experiment, the lower limb joint torque of stable gait move and variability gait move is predicted. The prediction values reflect the move intent of wearer. After adjusting prediction result by the power assist factor, it can be used to control the exoskeleton robot directly. For assessing the performance of power assist, performance evaluation index is proposed to evaluate the efficiency of power assist. According to the evaluation result, the algorithm is confirmed to be efficiency for providing the power assist for wearer.