Anti-Exploration by Random Network Distillation on PyTorch

This repository contains possible (not ideal one actually) PyTorch implementation of SAC RND with the wandb integration. It is based on official realization written on Jax.

Setup

In order to be able to run code, just install the requirements:

python install -r requirements.txt

Anyway, you would also need to install mujoco stuff by your own, you can follow the steps from the authors

if you want to train the model, setup rnd_config in config.py, initialize SACRNDTrainer in trainer.py and run its train method:

from trainer import SACRNDTrainer

trainer = SACRNDTrainer()
trainer.train()

if you find any bugs and mistakes in the code, please contact me :)