Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces Seg2Grasp, a modular pipeline that enhances suction grasping in bin picking by addressing the limitations of end-to-end learning in complex environments. The approach consists of three key modules: a Transformer-based Segmentation module for generating class-agnostic object masks, a Grasping module that optimizes suction points using surface normals, and a Classification module utilizing Mask-CLIP for accurate object identification. Real-world experiments show that Seg2Grasp significantly outperforms existing methods in both success rates and adaptability, making it a valuable advancement for automated bin picking in industrial applications.
Seg2Grasp achieves unprecedented adaptability in suction grasping, outperforming traditional methods in complex bin picking scenarios.
Current bin picking methods that rely heavily on end-to-end learning often falter when confronted with unfamiliar or complex objects in unstructured environments. To overcome these limitations, we introduce Seg2Grasp, a modular pipeline designed for robust suction grasping in dynamic and cluttered bin scenarios. Seg2Grasp is built on a three-step process: Segmentation, Grasping, and Classification. The Segmentation module employs a Transformer-based model to generate class-agnostic object masks from RGB-D images, ensuring accurate detection across various conditions. The Grasping module uses surface normals and mask proposals to determine the optimal suction points, enhancing grasp success. Finally, the Classification module leverages fine-tuned open-vocabulary Mask-CLIP for precise object identification, enabling versatile handling of diverse objects. Real-world robotic experiments demonstrate that Seg2Grasp outperforms existing methods in success rates and adaptability, establishing it as a powerful tool for automated bin picking in industrial settings.