taxposed
Science Score: 54.0%
This score indicates how likely this project is to be science-related based on various indicators:
-
✓CITATION.cff file
Found CITATION.cff file -
✓codemeta.json file
Found codemeta.json file -
✓.zenodo.json file
Found .zenodo.json file -
○DOI references
-
✓Academic publication links
Links to: arxiv.org -
○Academic email domains
-
○Institutional organization owner
-
○JOSS paper metadata
-
○Scientific vocabulary similarity
Low similarity (13.1%) to scientific vocabulary
Repository
Basic Info
- Host: GitHub
- Owner: jwang078
- Language: Python
- Default Branch: multimodal
- Size: 4.18 MB
Statistics
- Stars: 2
- Watchers: 1
- Forks: 1
- Open Issues: 0
- Releases: 0
Metadata Files
README.md
TAXPoseD
Learning Distributional Demonstration Spaces for Task-Specific Cross-Pose Estimation
TAX-PoseD, a method for learning relative placement prediction tasks, learns a spatially-grounded latent distribution over demonstrations without human annotations, using an architecture with geometric inductive biases.
Repository structure
multimodal- Stable latest branch (you are here)multimodal_dev- Latest branchmultimodal_icra2024- ICRA 2024 paper's model configurations
Installation
To install dependencies like pytorch, ndf_robot and other libraries, please follow the instructions in the TAX-Pose Github repo.
Then, install this repository with:
pip install -e .
Datasets
In our paper, we use the same 1-rack NDF training dataset as TAX-Pose, as described here.
We have also experimented with environments from the RPDiff paper.
Training a TAX-PoseD model
The current best model config has a long name: joint_train_pzX-dgcnn-transformer_pzY-pn2_hybridlatentz-global_gradclip1e-4_se3-action_upright-anchor_flow-fix-post-encoder-one-flow-head_joint2global-pzY-sample_anchor2action2global-opt2-pzX-sample_mod_easy_rack
python train_residual_flow_multimodal.py --config-path=../configs/rpdiff_data --config-name=joint_train_pzX-dgcnn-transformer_pzY-pn2_hybridlatentz-global_gradclip1e-4_se3-action_upright-anchor_flow-fix-post-encoder-one-flow-head_joint2global-pzY-sample_anchor2action2global-opt2-pzX-sample_mod_easy_rack dataset_root=TODO test_dataset_root=TODO log_dir=TODO rpdiff_descriptions_path=TODO
The ICRA submission model configurations can be run on the multimodal_icra2024 branch.
Cite
@article{wang2024taxposed,
title={Learning Distributional Demonstration Spaces for Task-Specific Cross-Pose Estimation},
author={Wang, Jenny and Donca, Octavian and Held, David},
journal={IEEE International Conference on Robotics and Automation (ICRA), 2024},
year={2024}
}
Owner
- Name: jwang078
- Login: jwang078
- Kind: user
- Repositories: 10
- Profile: https://github.com/jwang078
Citation (CITATION.cff)
@article{wang2024taxposed,
title={Learning Distributional Demonstration Spaces for Task-Specific Cross-Pose Estimation},
author={Wang, Jenny and Donca, Octavian and Held, David},
journal={IEEE International Conference on Robotics and Automation (ICRA), 2024},
year={2024}
}