Robotic Grasping And Compliant Placement Method Based On Vision And Force Sense

The rapid development of robotics has changed the form of manufacturing,and robots are widely used in military,medical and life fields.The main operations performed by the robot are grasping and placing.It is of great significance to enhance the robot’s grasping ability by giving the robot the ability to understand the scene through the vision sensor.However,when the target object has an irregular shape and is placed in a cluttered manner,only obtaining the position information and rotation angle of the object through the two-dimensional recognition algorithm does not meet the complex grasping requirements.Traditional 3D recognition algorithms run slowly,do not meet real-time requirements,and perform poorly in scenes with complex backgrounds and occlusions.This paper studies the pose estimation method,and proposes a real-time 6D pose estimation algorithm based on deep learning to improve the accuracy of pose estimation.When the robot performs the placement operation,it is easy to make rigid contact with the placement platform only by using position control,and it is easy to damage the grasped object.Fragile objects are well protected by the guided placement of force sensors.In this paper,the compliance control method based on the admittance principle is adopted for the placement operation,the contact force is sensed by the force sensor,and then the position of the end of the robot is adjusted to avoid damage to the object.The main research contents and achievements of this paper are as follows:A robot grasping and compliant placement system based on vision and force sense guidance is built.An experimental platform was built based on the UR3 e robot,and the internal parameters of the camera were calibrated,and the projection weight error met the experimental requirements.The "Eye-to-Hand" hand-eye system is constructed and calibrated,the rotation matrix and translation matrix are obtained,and the relative positional relationship between the robot end and the camera is obtained.This paper improves the YOLO-6D algorithm.The original algorithm has a faster running speed,but the accuracy is low in the scene with complex background and occlusion.In this paper,the original YOLO V2 detection network is replaced by the YOLO V3 network,which can extract more local features from the image and improve the algorithm’s recognition ability in occlusion scenes.And introduce SENet to form an attention mechanism,which can tilt more resources to useful features and reduce the interference of complex backgrounds on target object recognition.In addition,the pose estimation method is adjusted,and the cell group is selected for RANSAC-based EPn P pose estimation,and the accuracy of pose estimation is improved through multiple iterations.Experiments show that the algorithm in this paper can effectively improve the detection ability of occluded objects and improve the accuracy of pose estimation.Compared with other common pose estimation algorithms,it has better performance.For the compliant placement task,a compliance control method based on the admittance principle is adopted.The accuracy of the compliant control method mainly depends on the position control accuracy of the robot,which requires less computation and stronger robustness.The UR3 e collaborative robot has high position control accuracy,and the control method based on the admittance principle can achieve better control accuracy and meet real-time requirements.Afterwards,the multi-dimensional force control effect of the robot is tested through the force sense guided teaching experiment.The robot can track the expected trajectory well,which proves that the compliance control method in this paper has a good compliance control effect.Relying on the built experimental platform,the target object pose estimation experiment and the compliant placement experiment were carried out.The experimental results of pose estimation and grasping prove that the pose estimation algorithm in this paper has high performance and can accurately estimate the pose of occluded objects.In the compliant placement experiment,the compliance control method based on the admittance principle makes the contact force change slowly and the fluctuation is small when the grasped object contacts the experimental platform,which can well protect the fragile objects.