End-to-End Dexterous Grasp Learning from Single-View Point Clouds via a Multi-Object Scene Dataset

Dimensions overall score 8.0

• DGS-Net achieves a grasp success rate of 88.63% in simulation.result 95%
• DGS-Net achieves a grasp success rate of 78.98% on a real robotic platform.result 95%
• DGS-Net exhibits lower penetration with a mean penetration depth of 0.375 mm and penetration volume of 559.45 mm³.result 95%
• The proposed dataset comprises 307 objects, 240 multi-object scenes, and over 350k validated grasps.technical 95%
• Current mainstream grasping datasets predominantly focus on single-object scenarios and predefined grasp configurations, limiting their generalizability in real-world applications.limitation 85%
• The proposed two-stage grasp data generation strategy progresses from dense single-object grasp synthesis to dense scene-level grasp generation.method 90%
• DGS-Net is an end-to-end grasp prediction network capable of learning dense grasp configurations from single-view point clouds in multi-object scenes.technical 90%
• The dataset provides more robust and accurate supervision for dexterous grasp learning by explicitly modeling grasp offsets and pre-grasp configurations.technical 85%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card