GraspNeRF: Multiview-based 6-DoF Grasp Detection for Transparent and Specular Objects Using Generalizable NeRF (ICRA 2023)
This is the official repository of GraspNeRF: Multiview-based 6-DoF Grasp Detection for Transparent and Specular Objects Using Generalizable NeRF.
For more information, please visit our project page.
In this work, we propose a multiview RGB-based 6-DoF grasp detection network, GraspNeRF, that leverages the generalizable neural radiance field (NeRF) to achieve material-agnostic object grasping in clutter. Compared to the existing NeRF-based 3-DoF grasp detection methods that rely on densely captured input images and time-consuming per-scene optimization, our system can perform zero-shot NeRF construction with sparse RGB inputs and reliably detect 6-DoF grasps, both in real-time. The proposed framework jointly learns generalizable NeRF and grasp detection in an end-to-end manner, optimizing the scene representation construction for the grasping. For training data, we generate a large-scale photorealistic domain-randomized synthetic dataset of grasping in cluttered tabletop scenes that enables direct transfer to the real world. Experiments in synthetic and real-world environments demonstrate that our method significantly outperforms all the baselines in all the experiments.
This repository provides:
- PyTorch code, and weights of GraspNeRF.
- Grasp Simulator based on blender and pybullet.
- Multiview 6-DoF Grasping Dataset Generator and Examples.
- Please run
pip install -r requirements.txt
to install dependency.
- (optional) Please install blender 2.93.3--Ubuntu if you need simulation.
- Please generate or download and uncompress the example data to
data/
for training, and rendering assets todata/assets
for simulation. Specifically, download imagenet valset todata/assets/imagenet/images/val
which is used as random texture in simulation. - We provide pretrained weights for testing. Please download the checkpoint to
src/nr/ckpt/test
.
Our grasp simulation pipeline is depend on blender and pybullet. Please verify the installation before running simulation.
After the dependency and assets are ready, please run
bash run_simgrasp.sh
After the training data is ready, please run
bash train.sh GPU_ID
e.g. bash train.sh 0
.
bash run_pile_rand.sh
in ./data_generator for pile data generation.
If you find our work useful in your research, please consider citing:
@article{Dai2023GraspNeRF,
title={GraspNeRF: Multiview-based 6-DoF Grasp Detection for Transparent and Specular Objects Using Generalizable NeRF},
author={Qiyu Dai and Yan Zhu and Yiran Geng and Ciyu Ruan and Jiazhao Zhang and He Wang},
booktitle={IEEE International Conference on Robotics and Automation (ICRA)},
year={2023}
This work and the dataset are licensed under CC BY-NC 4.0.
If you have any questions, please open a github issue or contact us:
Qiyu Dai: qiyudai@pku.edu.cn, Yan Zhu: zhuyan_@stu.pku.edu.cn, He Wang: hewang@pku.edu.cn