Research On Autonomous Grasping Technology Of Wheelchair Mounted Robotic Arm Based On Multi View Templates

Based on a survey of the needs of persons with disabilities and the elderly,the group developed an easy-to-use,responsive,and operationally personalized WMRA(wheelchair mounted robotic arm),which can provide various life services such as assisting meals,drinking water,opening the door,etc.In particular,autonomous grasping is the core step for accomplishing these auxiliary tasks.In order to achieve autonomous grasping,the WMRA needs to perform teaching exercises for different household items in advance,record the identification template of the captured items and the corresponding grasping posture.Then,the vision system uses offline training data to identify the target object and estimate the space pose of the object relative to the template.Finally,the grasping posture and grasping trajectory are determined,so that the WMRA completes the grasping process.Aimed at the case where the target template is a multi-viewpoint cloud,the project establishes the visual system of the WMRA using PCL(Point Cloud Library)and ROS(Robot Operation System).The specific research content is as follows:(1)Using the global feature VFH as a recognition feature,an object recognition system oriented to a multi-view point cloud template was built to determine the target of capture.The algorithm first reduces the computational complexity and noise interference through the filtering process.Then the scene object is segmented by using plane segmentation and Euclidean distance segmentation.Finally,the target object is obtained through VFH feature matching.(2)A composite target pose estimation algorithm is proposed.The algorithm first uses template-selecting method based on VFH to obtain the template which is closest to the target pose,and then uses the improved keypoint-registrating method to perform more accurate pose estimation.(3)The project conducted a study of the coordinate system calibration method and the establishment of an autonomous grabbing system based on ROS.In order to determine final gripping position and grabbing posture,the system establishes four coordinate systems,and carries out the determination of the internal reference matrix and the calibration study between the coordinate systems.Using the communication mode of nodes,topics,and services in ROS,information transfer between the robot arm,camera,microphone,and wheelchair was realized,and completing the construction of WMRA offline training and online grasping system.(4)Three sets of experiments were set up to evaluate the visual algorithm and autonomous grasping system.The first set of experiments performed 400 object identifications with different template numbers and background settings for 10 common home objects.The results showed that the success rate of the objec t recognition algorithm was above 90%,and the calculation time was less than 0.5 seconds.The second group of experiments evaluated the pose and pose estimation algorithm proposed in this study.The results show that the pose estimation algorithm has a significant improvement in accuracy over the existing template selection method.In addition,the improvement in the point cloud registration phase reduced the computation time of the entire pose estimation algorithm by more than 20%.The third group of experiments conducted 40 different poses for 5 different objects respectively,which proved that the system can meet the real-time and accuracy requirements of WMRA for autonomous grasping.