The GOGO engine (Good Grasps Only Data Engine)

Image from the paper showing the online data collection while executing the task.

Good Grasps Only: A data engine for self-supervised fine-tuning of pose estimation using grasp poses for verification

Paper | Model

This site is a collection of links related to the project GOGO engine.

The GOGO engine uses keymatchnet for the pose estimation. You can find the repostisory for keymatchnet at https://github.com/fhagelskjaer/keymatchnet.

With keymatchnet installed, you can run keymatchnet with the trained weights from GOGO engine the following way:

keymatchnet --model_root "frhag/keymatchnet_gogo_engine" --obj testdata/09_BGPSL6-9-L30-F7.stl --scene testdata/point_cloud.pcd --visu True

The code for running the in bin pose estimation and grasp estimation is available at https://github.com/fhagelskjaer/sdui40binpick.

To cite:
@article{hagelskjaer2024good,
  title={Good Grasps Only: A data engine for self-supervised fine-tuning of pose estimation using grasp poses for verification},
  author={Hagelskj{\ae}r, Frederik},
  journal={arXiv preprint arXiv:2409.11512},
  year={2024}
}