Robotic grasping from single-view observations remains a critical challenge in manipulation. However, existing methods still struggle to generate reliable grasp candidates and stably evaluate grasp feasibility under incomplete geometric information. To address these limitations, we present SuperGrasp, a new two-stage framework for single-view parallel-jaw grasping. In the first stage, we introduce a Similarity Matching Module that efficiently retrieves valid and diverse grasp candidates by matching the input single-view point cloud with a precomputed primitive dataset based on superquadric coefficients. In the second stage, we propose E-RNet, an end-to-end network that expands the grasp-aware region and takes the initial grasp closure region as a local anchor region, capturing the contextual relationship between the local region and its surrounding spatial neighborhood, thereby enabling more accurate and reliable grasp evaluation and introducing small-range local refinement to improve grasp adaptability. To enhance generalization, we construct a primitive dataset containing 1.2k standard geometric primitives for similarity matching and collect a point cloud dataset of 100k samples from 124 objects, annotated with stable grasp labels for network training. Extensive experiments in both simulation and real-world environments demonstrate that our method achieves stable grasping performance and good generalization across novel objects and clutter scenes.
Figures & tables
Fig. 1: Overview of the task pipeline. Under dense clutter and single-view observation constraints, the tabletop decluttering task is accomplished through a two-stage pipeline consisting of grasp candidate generation, grasp evaluation, and refinement.
Fig. 2: Convex superquadric shapes under different parameter settings.
Fig. 3: The primitive object database constructed in this work, which contains 1.2k standard geometric objects of various shapes and scales, including cylinders, frustums, elliptical cylinders, and cuboids.
Fig. 4: Overview: Our framework consists of two stages. In the first stage, we first estimate the superquadric coefficients of the target object and compute their similarity to those of the objects in the database. We then select the top- N most similar objects and transfer their grasp poses to the target object through a transformation matrix T , followed by a coarse filtering step to remove noisy candidates. In the second stage, we enlarge the grasp-aware region of each initial grasp and feed the cropped regional point cloud into a PointNet++ network for feature extraction. We then further crop the features corresponding to the points within the initial gripper closing region and aggregate them via max pooling to obtain the grasp anchor feature. Finally, the anchor feature is fed into an evaluation network and a refinement network to predict grasp feasibility and refinement feasibility, respectively.
Fig. 5: Gripper configuration and refinement strategy. In (a), the x -, y -, and z -axes define the gripper coordinate system, w represents the gripper width, L denotes the fingertip length, and Fd is the applied disturbance force directed vertically downward. In (b), the initial gripper is first extended downward by 0.008m along the z -axis, and is then rotated by ±15∘ about the z -axis.
Method
Seen(obj=10, scene=30)
Seen(obj=20, scene=30)
Unseen(obj=10, scene=30)
GSR
TSR
GA
GSR
TSR
GA
GSR
TSR
GA
PointNetGPD [ 7 ]
78.76%
29.67%
113
59.43%
34.67%
350
33.90%
6.67%
59
Contact-GraspNet [ 16 ]
78.25%
80.33%
308
75.42%
82.83%
659
63.91%
69.67%
327
Chen et al. [ 5 ]
–
–
–
–
–
–
58.14%
83.33%
430
Ours (S1) + PointNetGPD [ 7 ]
63.34%
72.00%
341
38.42%
47.83%
747
31.62%
37.00%
351
PointNetGPD [ 7 ] + Ours (S2)
93.60%
53.67%
172
79.02%
48.33%
367
87.85%
31.33%
107
TABLE I: Comparison of different methods in simulation
Fig. 6: Simulation experiment settings. (a) Scene with 10 seen objects. (b) Scene with 20 seen objects. (c) Scene with 10 novel objects.
Fig. 7: Real-world experiment settings. We design two real-world test scenarios: (1) the objects within the green box are similar to those in the training set; and (2) the objects within the red box differ significantly from the training set in both shape and size.
Method
Similar (obj=10, scene=10)
Novel (obj=10, scene=10)
GSR
TSR
GSR
TSR
PointNetGPD [ 7 ]
58.82%
20.00%
52.38%
11.00%
Contact-GraspNet [ 16 ]
78.95%
60.00%
74.65%
53.00%
Chen et al. [ 5 ]
–
–
69.23%
90.00%
Ours
98.00%
98.00%
94.12%
96.00%
TABLE II: Comparison of different methods in the real world
Dexterous grasping in multi-object scene constitutes a fundamental challenge in robotic manipulation. Current mainstream grasping datasets predominantly focus on single-object scenarios and predefined grasp configurations, often neglecting environmental interference and the modeling of dexterous pre-grasp gesture, thereby limiting their generalizability in real-world applications. To address this, we propose DGS-Net, an end-to-end grasp prediction network capable of learning dense grasp configurations from single-view point clouds in multi-object scene. Furthermore, we propose a two-stage grasp data generation strategy that progresses from dense single-object grasp synthesis to dense scene-level grasp generation. Our dataset comprises 307 objects, 240 multi-object scenes, and over 350k validated grasps. By explicitly modeling grasp offsets and pre-grasp configurations, the dataset provides more robust and accurate supervision for dexterous grasp learning. Experimental results show that DGS-Net achieves grasp success rates of 88.63% in simulation and 78.98% on a real robotic platform, while exhibiting lower penetration with a mean penetration depth of 0.375 mm and penetration volume of 559.45 mm^3, outperforming existing methods and demonstrating strong effectiveness and generalization capability. Our dataset is available at https://github.com/4taotao8/DGS-Net.
Single-view RGB-D grasp detection remains a common choice in 6-DoF robotic grasping systems, which typically requires a depth sensor. While RGB-only 6-DoF grasp methods has been studied recently, their inaccurate geometric representation is not directly suitable for physically reliable robotic manipulation, thereby hindering reliable grasp generation. To address these limitations, we propose MG-Grasp, a novel depth-free 6-DoF grasping framework that achieves high-quality object grasping. Leveraging two-view 3D foundation model with camera intrinsic/extrinsic, our method reconstructs metric-scale and multi-view consistent dense point clouds from sparse RGB images and generates stable 6-DoF grasp. Experiments on GraspNet-1Billion dataset and real world demonstrate that MG-Grasp achieves state-of-the-art (SOTA) grasp performance among RGB-based 6-DoF grasping methods.
Kangxu Wang, Siang Chen, Chenxing Jiang +3
The Department of Electronic Engineering, Tsinghua University, Beijing 100084, China. · The Department of Electronic and Computer Engineering, The Hong Kong University of Science and Technology, Hong Kong, China.
Dual-arm robotic grasping is essential for manipulating large, heavy, and geometrically complex objects that cannot be reliably handled using a single manipulator. These large objects often contain only sparse graspable regions determined by local geometric properties such as thickness, edge structure, and gripper clearance. Prior bimanual grasping methods assume access to a full point cloud of the object which inherently contains this geometric information, but may not be accessible in real scenarios. This work proposes PartialBiGrasp, a dual-arm grasp generation framework that operates directly on partial point cloud observations. Our model learns geometric features implicitly through convolutional occupancy networks, enabling local reasoning about graspability, collision-free contact regions, and object thickness. We leverage this understanding to generate force-closure compliant grasp pairs, which are further refined using a sampling-based optimization to correct for ambiguity caused by incomplete geometry. We evaluate our approach using analytical force-closure metrics, large-scale simulation experiments, and real-world robot evaluations on noisy partial point clouds of novel objects, demonstrating robust and physically stable dual-arm grasp generation.