A2G: Leveraging Intuitive Physics for Force-Efficient Robotic Grasping

This is the code of "A2G: Intuitive Physics Helps Robot Grasp Better to Improve the Manipulation Performance"

We will update the way to use it as soon as possible.

paper, video

Citation

If you find our work useful, please cite it.

@ARTICLE{cheng2024a2g,
  author={Cheng, Yaofeng and Liu, Shengkai and Zha, Fusheng and Guo, Wei and Du, Haoyu and Wang, Pengfei and Bing, Zhenshan},
  journal={IEEE Robotics and Automation Letters}, 
  title={A2G: Leveraging Intuitive Physics for Force-Efficient Robotic Grasping}, 
  year={2024},
  volume={9},
  number={7},
  pages={6376-6383},
  keywords={Task analysis;Force;Physics;Point cloud compression;Market research;Grasping;Grippers;Deep learning;grasp position logic;task-oriented grasp},
  doi={10.1109/LRA.2024.3401675}}

Train and inference

cd A2G

# train
bash ./scripts/run_handle.sh

# inference
bash ./scripts/inference_handle.sh

new dataset can be generated with run_handle.sh, if you don't want to generate new data, just add the dataset_path in the sh file.