ACRONYM Dataset Overview

Dataset Summary

ACRONYM is a dataset containing 17.7 M simulated parallel‑jaw grasps across 8,872 objects, generated using NVIDIA FleX.

Dataset Contents

• Contains samples of grasp data
• Provides tools for visualizing grasps, generating random scenes, and rendering observations

License

• Source code is released under the MIT License
• Dataset is released under the CC BY‑NC 4.0 License

Environment Requirements

• Python 3
• Install dependencies listed in requirements.txt via pip

Usage Examples

Visualize Grasps

acronym_visualize_grasps.py [-h] [--num_grasps NUM_GRASPS] input [input ...]

Generate Random Scenes and Visualize Grasps

generate_scene.py [-h] [--objects OBJECTS [OBJECTS ...]] --support SUPPORT [--support_scale SUPPORT_SCALE] [--show_grasps] [--num_grasps_per_object NUM_GRASPS_PER_OBJECT]

Render and Visualize Observations

render_observations.py [-h] [--objects OBJECTS [OBJECTS ...]] --support SUPPORT [--support_scale SUPPORT_SCALE] [--show_scene]

Obtaining the Full Dataset

1. Download the full dataset (1.6 GB): acronym.tar.gz
2. Download the ShapeNetSem meshes
3. Create watertight versions of the meshes

Dataset Citation

@inproceedings{acronym2020, title = {{ACRONYM}: A Large-Scale Grasp Dataset Based on Simulation}, author = {Eppner, Clemens and Mousavian, Arsalan and Fox, Dieter}, year = {2020}, booktitle = {Under Review at ICRA 2021} }