Research On The Motion Intention Prediction Technology For The Control Of Lower Extremity Exoskeleton Robot

The lower extremity exoskeleton robot is a kind of wearable power assisted mechanical device,which can be installed on the human body through the form of bandages.As a typical human-computer interaction system,the wearer can control exoskeleton through his own movement.However,the structure of human-computer integration puts forward high requirements for the control system.The exoskeleton robot needs to provide good assistance effect,at the same time,the safety of the wearer needs to be paid attention to all the time during the whole movement.Therefore,the exoskeleton robot needs to detect the wearer's motion attitude and motion intention in real time.In this case,the motion intention prediction technology becomes a key part of exoskeleton robot system and a high precision algorithm of motion intention prediction means better effect of assistance.With the development of exoskeleton robot in the direction of more flexible and intelligent,the traditional algorithm of motion intention prediction no longer meets the requirements of prediction accuracy.Aiming at this problem,combining with the related technologies in the field of artificial intelligence,this paper uses long-shortterm memory neural network to carry out research on high-precision prediction of motion intention.The specific research contents include:Based on gait analysis,this paper compares the current researches on using sensors to obtain human motion intentions,and chooses to use multiple inertial sensors and a simple exoskeleton platform to build an exoskeleton robot's motion data collection system.The system is used to collect motion data in five scenarios: walking on flat ground,going up stairs,going down stairs,going uphill and downhill,and construct a gait motion data set of exoskeleton robot for lower limbs.Afterwards,this paper uses BP neural network and time series prediction model to verify the key algorithms in the extremity exoskeleton robot on the proposed data set,which provides a data basis for the subsequent research of motion intention prediction algorithm.In the research of motion intention prediction algorithm,this paper adopts a motion intention prediction model based on the long short term memory network which is widely used and performs well in motion intention prediction tasks.At the same time,this paper studies and analyzes the existing optimization algorithm based on the long short term memory network.Seq2 seq model and AE-LSTM model is compared in our experiment.The experimental results show that the prediction accuracy of the basic long short-term memory network in the comparative experiment is low,which can't meet the high accuracy requirements of the motion intention prediction task in the lower extremity exoskeleton robot system.In the optimization of the long short term memory network,the prediction structure of long-short term memory networks is analyzed from the perspective of spatiotemporal data optimization.The design experiment proves that the basic longshort term memory network is difficult to model the spatial position relationship of exoskeleton robot when it moves.Therefore,on the basis of long-short term memory network,this paper optimizes the structure of long-short term memory network in combination with the characteristics of motion symmetry and short-term prediction in the mobile task of lower extremity exoskeleton robot,and proposes the optimization based on gait symmetry and the optimization based on short-term prediction task.The performance of the proposed algorithm is evaluated on the data set.The experimental results show that the proposed optimization structure can effectively improve the prediction accuracy of long-short term memory networks in motion intention prediction task.