Two-Fingered Grasp Planning for Randomized Bin-Picking

This work aims to increase reliability and reduce cycle time in order to realize a commercially-viable visionguided robotic bin-picking system. We present a novel method for two-fingered grasp generation and target selection for binpicking of randomized parts. We also propose a definition for grasp robustness, and use this to formulate a new grasp quality measure. A densely-sampled set of grasps is generated and evaluated using our proposed quality measure. The highestquality grasps are then used to provide more valid picking options in the context of a randomized pile of parts, and to determine the best part to pick up. Our experimental results show a substantial increase in the average number of valid picking options when compared with a typical industrial approach for target selection.