Robot Dexterous Grasping in Cluttered Scenes Based on Single-View Point Cloud

Grasping is a basic but challenging task in intelligent robotic manipulation, and grasping pose detection is the key in this task. Most current work is shifting from 2D planar grasping to more flexible six-degree-of-freedom (6-DoF) grasping, and some significant progress has been made. However, there are still limitations such as low success rate and poor robustness in cluttered scenes. In this paper, we investigate 6-DoF grasping in cluttered scenes, and propose a cascaded multi-target learning network based on self-attention mechanism and multiscale sampling. The self-attention mechanism effectively improves the network’s attention to the correct grasping pose, while multiscale sampling improves the network’s adaptability to grasping objects of different sizes. In addition, we propose a dual-objective evaluation metric based on the force-closure metric and the center-of-mass distance metric, which can make a more reasonable and reliable evaluation of the grasping poses in the dataset. Our model is evaluated on large-scale benchmarks as well as the real robot system. The proposed method achieves state-of-the-art results on GraspNet-1Billion (8.8+AP), and shows 95.24% success rates in real cluttered scenes. Note to Practitioners—This paper presents a framework for 6-DoF grasping in cluttered scenes using single-view point clouds, improving success rates without requiring complete point clouds. By introducing a self-attention mechanism and multiscale sampling, the model adapts to varying object sizes and enhances focus on the correct grasping pose. A dual-objective evaluation metric based on force-closure and center-of-mass distance also refines dataset quality, balancing force and geometric criteria for more reliable grasping poses. This approach supports direct deployment in real-world robotic applications like automated sorting and assembly, with high robustness and efficiency in cluttered environments.