Research On Multi-sensor Data Fusion Of Lower Limb Exoskeleton Robot

The Lower Limb Exoskeleton Robot is a kind of wearable rehabilitation robot to help the paraplegic patients to realize independent living ability,like standing and walking.It has become a hot spot in rehabilitation medicine and Robotics nowadays.The Lower Limb Exoskeleton Robot should be operated in safe and correct mode due to its medical purpose.Thus it deserves robust and efficient sensing ability to first obtain complicated kinetic and environmental information,then ensure patients’ safety by accurate discrimination of kinetic posture.The Multisensor Data Fusion is exactly an effective method to improve sensing ability.This paper analyze the mapping relationship between sensor data and kinetic status,and the stability and accuracy of such relation in view of data fusion.This paper also contributes a novel solution of data fusion in Exoskeleton Robot with better sensing ability,as briefly shown below.First,this paper establish a sensor network in exoskeleton robot for collecting kinetic data,elaborately determine the sensor type and select the CAN network for connection to meet the data acquisition demand.Then,we design a parallel data fusion method for low ranks of multisensor.Each sensor keeps on filtering and separating singularity in data,calculating the tilt angle with the couple of acceleration and palstance.Next,we extract the feature of sample sets,trying to analyze the association between these feature and kinetic status.The procedure includes normalization,feature ranking by combination of distance-based and information gain based method,and PCA based dimension reducing of high rank feature,thus provide offline data set for next fusion steps.Finally,this paper carry out the the high-level data fusion to discriminate different kinetic status.SVM is selected as elementary classifier for its reliability against disturbance.Further more,considering forward correlation among each status in exoskeleton system,we propose a finite-state transfer based SVM classifying algorithm,which evidently enhance the accuracy and predict the abnormal status.The experiment supports the efficiency and robustness of the improved algorithm,and shows its applicability for real time classification.