Manufacturing and production industries are increasingly turning to robots to carry out repetitive picking operations in an efficient manner. This paper focuses on tackling the novel challenge of automating the bin picking process for entangled objects, for which there is very little research. The chosen case study are sets of freely curved tubes, which are prone to occlusions and entanglement. The proposed algorithm builds a representation of the tubes as an ordered list of cylinders and joints using a point cloud acquired by a 3D scanner. This representation enables the detection of occlusions in the tubes. The solution also performs grasp planning and motion planning, by evaluating post-grasp trajectories via simulation using Gazebo and the ODE physics engine. A force/torque sensor is used to determine how many items were picked by a robot gripper and in which direction it should rotate to solve cases of entanglement. Real-life experiments with sets of PVC tubes and rubber radiator hoses showed that the robot was able to pick a single tube on the first try with success rates of 99% and 93%, respectively. This study indicates that using simulation for motion planning is a promising solution to deal with entangled objects.

Bin picking is a challenging task involving many research domains within the perception and grasping fields, for which there are no perfect and reliable solutions available that are applicable to a wide range of unstructured and cluttered environments present in industrial factories and logistics centers. This paper contributes with research on the topic of object segmentation in cluttered scenarios, independent of previous object shape knowledge, for textured and textureless objects. In addition, it addresses the demand for extended datasets in deep learning tasks with realistic data. We propose a solution using a Mask R-CNN for 2D object segmentation, trained with real data acquired from a RGB-D sensor and synthetic data generated in Blender, combined with 3D point-cloud segmentation to extract a segmented point cloud belonging to a single object from the bin. Next, it is employed a re-configurable pipeline for 6-DoF object pose estimation, followed by a grasp planner to select a feasible grasp pose. The experimental results show that the object segmentation approach is efficient and accurate in cluttered scenarios with several occlusions. The neural network model was trained with both real and simulated data, enhancing the success rate from the previous classical segmentation, displaying an overall grasping success rate of 87.5%.