Auto Snakes

A small project I made testing out Q learning in games. Each snake has a 3 x 3 vision box around its head that detect if that cell is a wall. Every game tick, chooses a the direction that has the highest probability of giving it points. 

A snake get 1 point when it approaches its goal, and is penalized 1 point when it does not. It gets 10 points if it reaches its goal, and is penalized 20 points if it dies.