Here is the code for our project on Maximum Entropy-Based Hindsight Goal Generation.
The code was developed by Josephine Koe (5th Year Masters Student at UC Berkeley) and Ritika Shrivastava (5th Year Masters Student at UC Berkeley). For details on Maximum Entropy-Based Hindsight Goal Generation please read our paper.
The code is developed based on OpenAI Baselines (link: https://github.com/openai/baselines).
The code requires python3 (>=3.5) with the development headers. You'll also need system packages CMake, OpenMPI and zlib. Those can be installed using the CS285Project.ipynb To run the code, you need to install OpenAI Gym (link: https://github.com/openai/gym). We use the robotics environment in OpenAI Gym, which needs the MuJoCu physics engine (link: http://www.mujoco.org/).
The experiments were carried out on Google Collab. The model can be trained and evaluated using CS285Project.ipynb. If using Google Collab make sure to change the base repository the code being referenced from.
Josephine_s\ Copy\ of \CS285Project.ipynb contains code for training the model and code to visualize the results.
Thank you for your time. Feel free to reach out if you have any questions or concerns.