Abstract:We propose a fast and accurate method of 6D object pose estimation for bin-picking of mechanical parts by a robot manipulator. We extend the single-shot approach to stereo vision by application of attention architecture. Our convolutional neural network model regresses to object locations and rotations from either a left image or a right image without depth information. Then, a stereo feature matching module, designated as Stereo Grid Attention, generates stereo grid matching maps. The important point of our method is only to calculate disparity of the objects found by the attention from stereo images, instead of calculating a point cloud over the entire image. The disparity value is then used to calculate the depth to the objects by the principle of triangulation. Our method also achieves a rapid processing speed of pose estimation by the single-shot architecture and it is possible to process a 1024 x 1024 pixels image in 75 milliseconds on the Jetson AGX Xavier implemented with half-float model. Weakly textured mechanical parts are used to exemplify the method. First, we create original synthetic datasets for training and evaluating of the proposed model. This dataset is created by capturing and rendering numerous 3D models of several types of mechanical parts in virtual space. Finally, we use a robotic manipulator with an electromagnetic gripper to pick up the mechanical parts in a cluttered state to verify the validity of our method in an actual scene. When a raw stereo image is used by the proposed method from our stereo camera to detect black steel screws, stainless screws, and DC motor parts, i.e., cases, rotor cores and commutator caps, the bin-picking tasks are successful with 76.3%, 64.0%, 50.5%, 89.1% and 64.2% probability, respectively.

A Pose Estimation Approach Based on Keypoints Detection for Robotic Bin-picking Application

Fast and Robust Object Pose Estimation Based on Point Pair Feature for Bin Picking

Fast and Accurate Pose Estimation for Industrial Workpieces Robotic Picking

Sparse Convolution Based 6D Pose Estimation for Robotic Bin-Picking with Point Clouds

A 6D Pose Estimation for Robotic Bin-Picking Using Point-Pair Features with Curvature (Cur-Ppf).

Realtime 3D Reconstruction at Scale and Object Pose Estimation for Bin Picking System.

A Robotic Semantic Grasping Method for Pick-and-place Tasks

A Robust Pixel-Wise Prediction Network with Applications to Industrial Robotic Grasping

AttentionVote: A Coarse-to-fine Voting Network of Anchor-Free 6D Pose Estimation on Point Cloud for Robotic Bin-Picking Application

Semantic Part Segmentation Method Based 3D Object Pose Estimation with RGB-D Images for Bin-Picking

Efficient 3D Object Recognition for Unadjusted Bin Picking Automation

Robotic Grasping With Multi-View Image Acquisition and Model-Based Pose Estimation

DBPF: A Framework for Efficient and Robust Dynamic Bin-Picking

A Novel Robotic Grasp Detection Framework Using Low-Cost RGB-D Camera for Industrial Bin Picking

PPR-Net: Point-wise Pose Regression Network for Instance Segmentation and 6D Pose Estimation in Bin-picking Scenarios.

Deep Learning Based 6D Pose Estimation for Bin-Picking Using 3D Point Clouds

3D object recognition and pose estimation for random bin-picking using Partition Viewpoint Feature Histograms

An Industrial Bin Picking Framework for Assembly Tasks.

Object Pose Estimation for Robotic Grasping based on Multi-view Keypoint Detection

6D Pose Estimation Based on 3D Edge Binocular Reprojection Optimization for Robotic Assembly

Stereo Vision Based Single-Shot 6D Object Pose Estimation for Bin-Picking by a Robot Manipulator