Robotics · Grasping · VGNBenchmark detail
Grasp benchmark · Parallel-jaw (Franka)

VGN

6-DoFvolumetricreal-timeclutter

The Volumetric Grasping Network (VGN) predicts 6-DoF grasps directly from a Truncated Signed Distance Function (TSDF) of the scene in a single forward pass, making it fast enough for real-time clutter clearing.

It established the volumetric / TSDF input that GIGA later extended with implicit representations.

Primary source
At a glance
Source
Breyer et al., CoRL 2020
Year
2020
Scale
~2M synthetic grasps · 303 training meshes
Gripper
Parallel-jaw (Franka)
Modality
TSDF (from depth)
Best-known
HW: 80% grasp success · 92% clutter clearance · ~10 ms plan
Key results
  • HW: 80% grasp success · 92% clutter clearance
  • Plans a grasp in roughly 10 ms
  • Volumetric TSDF input, single forward pass

SIM = simulation result · HW = physical hardware. Image-wise accuracy is detection quality, not real-robot pick success. Figures cited from Breyer et al., CoRL 2020.

Related benchmarks

← Back to the grasping register
Parallel-jaw (Franka)

ACRONYM

SIM: 59.21% of generated grasps succeed (label generation)