/MapGo

TensorFlow implementation of the paper MapGo: Model-Assisted Policy Optimization for Goal-Oriented Tasks

Primary LanguagePythonMIT LicenseMIT

MapGo: Model-Assisted Policy Optimization for Goal-Oriented Tasks

This is the TensorFlow implementation of the paper MapGo: Model-Assisted Policy Optimization for Goal-Oriented Tasks accepted by IJCAI2021.

Requirements:

pip install -r requirements.txt

To run:

example:

python mapgo.py --tag=test --env=AntLocomotion-v0 --buffer_size=100000 --pi_lr=1e-3 --q_lr=1e-3 --timesteps=100 --foresight_length=30 --model_based_training=True --fgi=True

Hyperparameters are in config.py.

Acknowledgements

Our implementation is based on HGG and mbpo codebase. Some of the environments are based on multiworld.