1.) Mapping Agent(car) coordinates/situation into states

x2-y2 is state 1
x2-y1 is state 2
x1-y2 is state 3
x1-y1 is state 4

2.) Adding random scenarios as inputs

Syntax: state action reward next-state

2 down 0 1 2 left 100 4 4 right 0 2 2 down 0 1 1 left -100 3 1 left -100 3 3 up 100 4 1 up 0 2 3 up 100 4 1 left -100 3 3 right 0 1 1 up 0 2 3 up 100 4
compute policy

Results: (Note: Press compute policy button multiple times until it gets the optimal policy)

Per Action:

Congratulations! found optimal solution from every state

A more complicated example