Object recognition: Bin-picking for industrial use

Abstract

This paper shows a method for object pose detection and gripping point determination that is successfully applied to industrial applications running a three-shift system. The industrial applications are fully automated feeding systems, commonly known as bin-picking. The proposed method for object detection is a generic approach to detect 6 degrees of freedom of any solid objects with arbitrary geometry. The proposed method is using 3D range data and is based on a heuristic tree search using a 6D pose voting scheme. For gripping point determination the object removal is simulated using the range data and a CAD model of the gripper in order to avoid any kind of collisions. During long-term operations in different use-cases, the method showed its usability regarding the crucial requirements, such as robustness, accuracy, portability and speed

Automatic Grasp Pose Generation for Parallel Jaw Grippers

“…While automatic grasp pose synthesis approaches increasingly gain popularity, many established bin-picking applications still utilize a manual grasp pose definition [24,25]. This usually requires expert knowledge of the object and application, leading to a work-intensive reconfiguration process [26, 27,28,29,30]. Some of these approaches enable the user to add degrees of freedom around defined grasps to further increase the total number of grasp poses and simplify the teach-in process [26,27].…”

“…Mahler et al [31] complex algorithmic top-down Spenrath et al [28,24,25] complex manual 6D Gatrell [32] simple algorithmic 6D Smith et al [33] simple algorithmic top-down Buchholz et al [27] complex manual 6D Mousavian et al [21] complex random 6D Brook et al [23] complex random 6D Bohg et al [22] complex random 6D Depierre et al [34] complex random top-down Johns et al [19] complex random top-down Saxena et al [35] complex manual 6D ours complex algorithmic 6D…”

“…To evaluate our approach in a real-world task, an analytical bin-picking solution 3 [28] was provided with the option to import the generated grasps. The approach records a 3D point cloud of the scene, estimates 6D object poses using template matching [44,45], and selects a collision-free and kinematically feasible grasp pose on the localized objects using heuristic search [24,25,26].…”

Deep Workpiece Region Segmentation for Bin Picking

“…The other class of approaches tries to find suitable grasp locations by first estimating the pose of object and then extracting grasp points. Such a method is being used by the Bin Picking 3D software, described in [4], [5], [6], which is used by our approach as post-processing step to determine object pose in a point cloud. It searches for local maxima in the depth map to start a pose estimation of the known object by fitting the CAD model and determining its pose using a look-up table.…”

“…In these experiments, an object pose estimation (OPE) technique using generalized Hough transformation [5], [6] is applied on the real point clouds. We estimated the ground truth via manually adjusting the CAD model in the scene and then applying the ICP algorithm.…”