Atari Game AI Pong model expansion using Pytorch SB3

https://colab.research.google.com/github/Stable-Baselines-Team/rl-colab-notebooks/blob/master/atari_games.ipynb
https://colab.research.google.com/github/Stable-Baselines-Team/rl-colab-notebooks/blob/sb3/atari_games.ipynb
https://towardsdatascience.com/stable-baselines-a-fork-of-openai-baselines-reinforcement-learning-made-easy-df87c4b2fc82
Added complexities
No fixed walls
Velocity of ball varies
Random ball direction change
Player can jump multiple space units
Cost per move action for player hence minimize actions
* no penalty if missed ball provided no action / move

Reward function
a. based on flags
Flag #1
to commit / play
Flag #2 : type of actions
discrete move
analogous
move up 1 space units
move up 2 space units
move up 4 space units
move up 8 space units
move up 16 space units
move dn 1 space units
move dn 2 space units
move dn 4 space units
move dn 8 space units
move dn 16 space units
Flag #3 : next action step gap
wait time in seconds
1 sec
2 sec
3 sec
4 sec
5 sec
vector distance of ball
1 unit space
2 unit space
3 unit space
4 unit space
5 unit space
no commit (skip) - potential opportunity lost penalty?)
Minus reward
total moves * unit cost