R2D2 (Recurrent Experience Replay in Distributed Reinforcement Learning)

introduction

An Implementation of Recurrent Experience Replay in Distributed Reinforcement Learning (Kapturowski et al. 2019) in PyTorch.

Training

First adjust parameter settings in config.py (number of actors, environment name, etc.).

Then run:

python3 train.py

Testing

python3 test.py

Result

The code was trained and tested in Atari game 'MsPacmac'