Active-Perceptive Language-Oriented Grasp Policy for Heavily Cluttered Scenes

RA-L 2025

Official code of paper Active-Perceptive Language-Oriented Grasp Policy for Heavily Cluttered Scenes.

Framework

Installation

This code has been tested on Ubuntu20.04 with Python 3.9, Pytorch 2.3.1 and Cuda 11.8.

Create new Conda environment:

conda create -n apeg python=3.9
conda activate apeg

Install torch 2.3.1 wit cuda 11.8:

pip install torch==2.3.1 torchvision==0.18.1 torchaudio==2.3.1 --index-url https://download.pytorch.org/whl/cu118

Install pytorch3d (please refer to this guide):

pip install pytorch3d

Install maniskill (manually modified) and CustomTasks:

pip install -e .
cd CustomTasks
pip install -e .

Install LocalGrasp requirements:

cd ../localgrasp
SKLEARN_ALLOW_DEPRECATED_SKLEARN_PACKAGE_INSTALL=True pip install -r requirements.txt

Install other requirements:

cd ../
pip install -r requirements.txt

Usage

Checkpoint

Checkpoints can be downloaded from Here.

Demo

Run demo code. To train the model, please set the parameter 'evaluate' to False and change the epi_ids.

cd ViLG/scripts
python main.py

Citation

Please cite our paper in your publications if it helps your research.

@article{dai2025active,
  title={Active-Perceptive Language-Oriented Grasp Policy for Heavily Cluttered Scenes},
  author={Dai, Yixiang and Chen, Siang and Yang, Kaiqin and Hu, Dingchang and Xie, Pengwei and Li, Guosheng and Shen, Yuan and Wang, Guijin},
  journal={IEEE Robotics and Automation Letters},
  year={2025},
  publisher={IEEE}
}