Graph networks for learnable physical simulation
This repository is a partial implementation of Graph networks as learnable physics engines for inference and control.
- DeepMind control suite
- pytorch 0.4.1 (other versions untested)
Generate data with
gen_data.py script, you should get control signals and resulting 6-link swimmers states.
Learn data distribution first with
python test_normalizer.py. It will generate
normalize.pth. Then run
python train_gn.py to train the model. The learning rate schedule corresponds to "fast training" in original paper.
python evaluate_gn.py <model path>