GraspNet-1Billion: Robotic Grasping Benchmark with 1 Billion Poses

GraspNet-1Billion provides 1 billion grasp poses for robotic manipulation research, maintained by the GraspNet team. The dataset serves as a large-scale benchmark for grasping tasks and was last updated in June 2025.

Use Cases

• Training robotic models for grasp detection using the 1 billion grasp labels
• Evaluating grasping algorithms using the grasp pose data
• Benchmarking manipulation performance in robotic grasping tasks

Strengths

• 1 billion grasp poses
• MIT license
• Dedicated API for data access

Limitations

• Large scale requires significant storage and computational resources
• Limited to the specific domain of robotic grasping

Provenance

Source

GraspNet

Freshness

Last updated June 2025