Consistent Dynamics

Code for the paper "Learning Powerful Policies by Using Consistent Dynamics Model" presented at Workshop on "Structure & Priors in Reinforcement Learning" at ICLR 2019.

Arxiv link coming soon!

Code Setup

• Copy config/sample.config.yaml to config/config.yaml. For testing, no changes are needed in the newly created config.yaml.
• Install requirements using pip3 install -r requirements.txt.
• Following packages needs to be installed from source:
  • https://github.com/openai/baselines#installation
• From the root dir, run PYTHONPATH=$PWD python3 codes/app/main.py

Notes

• All the logic has a single entry point codes/app/main.py. What task is to be performed is controlled via the config files.

Setup Issues

• If you get an error related to patchelf, try conda install -c anaconda patchelf