Study On Unstructured Bin-picking System For Fragile Items

The development of computer vision technology has spawned many solutions to the problem of unstructured bin-picking,which is a big problem on robot manipulation.These solutions usually use rigid grippers or suction cups as end effectors.In order to avoid squeezing the objects to be sorted,complex sensor controlling is needed.On the other hand,the soft grippers have high flexibility due to the adaptability of soft materials,which have many advantages like high adaptability to complex environments and convenience of human-computer interaction.This provides new possibilities for robot to grasp fragile items.In this paper,a gas-driven flexible soft gripper is used for unstructured bin-picking of fragile objects.A vision system using improved instance segmentation network and pose estimation method based on point cloud processing is set up for the soft gripper.A complete unstructured bin-picking system is set up.The main research contents are as follows:(1)Research the design of the soft gripper for fragile objects and do experiments with it.Design a fluid flexible soft gripper for fragile objects.Do finite element analysis based on Yeoh model.Set up experimental platform for single finger performance test.Provide a fitting function relationship between finger deformation and air pressure for vision system according to the simulation and experimental results.Design gas circuit and electronic control system for the soft gripper.(2)Analyze the convolutional neural network and improved instance segmentation algorithm based on Mask r-cnn.Research the algorithm principle of convolutional neural network and Mask r-cnn.Make a training set and adjust the network model according to the environment in this paper so that it can be better used.Use improved method for Mask r-cnn to alleviate the problem of blurred borders.(3)Research the grasping strategy of the soft gripper based on the point cloud.Preprocess the point cloud through RANSAC,filtering,MLS and other algorithms.Combine the mask obtained by instance segmentation algorithms and the segmented point cloud and propose a grasp planning strategy for the soft gripper.(4)Platform construction and experimental verification.Set up an experimental platform of the unstructured bin-picking system using a soft gripper.Design a hand-eye calibration system to complete the conversion from camera coordinate system to robot arm coordinate system and perform calibration accuracy test.Run the entire unstructured bin-picking system for fragile items and analyze the accuracy and reliability through experiments.