python3 -m pip install -r requirements.txt
Below command can collect dataset.
./collect_data.py
Below command can train the NewtonianVAE model
./train.py
./check_correlation.py
./check_operability.py
pixyz>=0.3.3
torch>=1.12.0
numpy>=1.23.1
PyYaml>=5.3.1
matplotlib>=3.5.2
tqdm>=4.64.0
dm_control>=1.0.8
tensorboardX>=2.5.1
- Jaques, Miguel, Michael Burke, and Timothy M. Hospedales. "Newtonianvae: Proportional control and goal identification from pixels via physical latent spaces." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021.
- Watters, Nicholas, et al. "Spatial broadcast decoder: A simple architecture for learning disentangled representations in vaes." arXiv preprint arXiv:1901.07017 (2019).
- Okumura, Ryo, Nobuki Nishio, and Tadahiro Taniguchi. "Tactile-Sensitive NewtonianVAE for High-Accuracy Industrial Connector-Socket Insertion." arXiv preprint arXiv:2203.05955 (2022).