PPO

Proximal Policy Optimization implementation with Tensorflow.

https://arxiv.org/pdf/1707.06347.pdf

requirements

• Python3

dependencies

• tensorflow
• gym[atari]
• opencv-python
• git+https://github.com/imai-laboratory/lightsaber

usage

training

$ python train.py --gpu {0 or -1} --render --final-steps 10000000

playing

$ python play.py --gpu {0 or -1} --render --load {path of models}

implementation

The code base is build up on these projects

• DQN
• OpenAI Baselines

Ongoing Work

Augmenting the observation space to be aware of damage classes of the agent.