Bin picking of deformable linear objects using object-oriented grasp planning

Abstract

Picking deformable linear objects (DLOs) from an unorganized supply is required for many industrial handling and assembly tasks. Automated picking with robots can address labour shortage, but remains an unresolved challenge for DLOs. This paper proposes an approach for bin picking of DLOs using object-oriented grasp planning. Therefore, DLOs are localized in 2D images through general purpose instance segmentation, and the DLO topology is determined. Subsequently, grasp pose candidates are sampled and evaluated, and suitable instances for picking are determined. Thus, this approach derives a collision-free grasp pose for picking a suitable object. Real-world experiments demonstrate a success rate of up to 97 % for picking DLOs from overlapping and disordered arrangements in the first attempt.