Grasp benchmark · Parallel-jaw
GIGA
6-DoFimplicit geometryclutterparallel-jaw
GIGA (Grasp detection via Implicit Geometry and Affordance) jointly learns grasp affordance and 3D scene geometry through implicit neural representations, improving cluttered-scene grasping by sharing structure between the two tasks.
It cleanly quantifies the clutter gap: in simulation its grasp success rate falls from 87.9% on packed scenes to 69.8% on heaped piles.
Primary source →At a glance
- Source
- Jiang et al., RSS 2021
- Year
- 2021
- Scale
- Built on VGN synthetic setup · affordance + implicit geometry
- Gripper
- Parallel-jaw
- Modality
- TSDF + implicit
- Best-known
- HW: 83.3% packed · 86.9% pile · SIM: 87.9% / 69.8%
Key results
- HW: 83.3% packed · 86.9% pile grasp success
- SIM: 87.9% packed · 69.8% pile — the clutter gap, quantified
- Joint affordance + implicit geometry learning
SIM = simulation result · HW = physical hardware. Image-wise accuracy is detection quality, not real-robot pick success. Figures cited from Jiang et al., RSS 2021.
Related benchmarks
← Back to the grasping registerParallel-jaw
GraspNet-1Billion →
De-facto clutter benchmark · AnyGrasp current SOTA (AP)
Parallel-jaw
Dex-Net 2.0 →
HW: 93% on adversarial · 99% precision on 40 novel objects (YuMi)
Parallel-jaw
Grasp-Anything →
Language-driven grasp synthesis · open-vocabulary scenes
Parallel-jaw
Jacquard →
~95% image-wise (GR-ConvNet-class)