Uses RUDDER on PPO to solve environments with delayed rewards.
Our code is based on the OpenAI Baselines package, in which we included our implementation of RUDDER for PPO for ATARI games with delayed rewards. Additionally, the Tensorflow Layer Library (TeLL) package (v1.0) is required for RUDDER.
RUDDER for PPO2 and more documentation is located in the folder baselines/ppo2_rudder.
Modified/new files:
- baselines/common/atari_wrappers.py
- baselines/common/distributions.py
- baselines/common/vec_env/vec_frame_stack.py
- baselines/ppo2_rudder/
- logger.py
- README.md