This repo contains code for the paper HyperDynamics: Meta-Learning Object and Agent Dynamics with Hypernetworks (ICLR 2021). This repo borrows code structure from learning_to_adapt.
Project site: zhouxian.github.io/hyperdynamics
If you use this code, please cite this paper:
@inproceedings{xian2021hyperdynamics,
author = {Zhou Xian and
Shamit Lal and
Hsiao{-}Yu Tung and
Emmanouil Antonios Platanios and
Katerina Fragkiadaki},
title = {HyperDynamics: Meta-Learning Object and Agent Dynamics with Hypernetworks},
booktitle = {9th International Conference on Learning Representations, {ICLR} 2021,
Virtual Event, Austria, May 3-7, 2021},
year = {2021},
}
Download and install mujoco (version 131).
Set up conda env and install dependencies:
conda env create -f environment.yml
conda activate hyperdyn
. source_this.sh
Training on the task Cheetah-Slop with tensorboard logging:
# HyperDynamics
python scripts/train_hyperdyn.py --en basic
# MAML baseline
python scripts/train_maml.py --en basic
Tensorboard log files are saved under log/{model}/{en}.
