Playing Atari Tennis with Deep Reinforcement Learning

Introduction:

An AI agent which is able to play Atari-2600 Tennis and surpass human player
🔴Red Side: Agent
🔵Blue Side: Opponent

Build Requirement:

Python 3.5
Pytorch 0.4.1
CUDA version is depend on GPU
Simply Run it with python main.py

Result:

Testing average return
Test the model for 30 Games per 200 episodes during training

Training return
Red line is the average return of every 10 episodes in the training sequence