Reinforcement Learning AI Maze Solver!

This is my Library-Free Javascript Reinforcement Learning AI Maze Solver.

Training Times(Time until maze solved) (Tested on M1 Macbook Air 2020):

Maze 1 (8coreGPU 8core CPU): <2mins

Maze 2 (8coreGPU 8core CPU): <7mins

Maze3 (8Gpu,8Cpu): ~30mins

Maze4 (8GPU,8CPU): >=120mins

By the way, The numbers on the squares indicated the VALUE the bot has given to it (Calculated using TDE Value functions). 

The reward for green squares is 20, red is -5, Yellow/Air is 0.

(Exploitation:Exploration Rate is normally 7:3 however the epsilon-greegy strategy is in use and this value will change throughout.)

Extra Info: In this simulation, the bot has no knowledge of it's surroundings or the positions of the obstacles/rewards and only knows rewards of the squares it has visited. This RL model uses no libraries and is made in vanilla Javascript!