This is a conveyor system. Boxes enter the system at A and move towards C where they exit the system. The same goes for B->D.
Boxes on path A->C have higher priority than boxes B->D. Boxes can stop and wait in each square. If there is a box on A2 and also on B2, the one from A2 should go through X first.
If there are boxes on C2, C1, A2 and B2 then the box from B2 should go through X and box on A2 shuold wait, until C1 is empty. Otherwise it would block the path B->D.
How can I solve this problem with neural networks? So for every turn I would like to input the current state of each block and as a result I would like to get from where to where the next box should move.
For example: System state: BOX present at A1 Result: A1, A2
I don't know if neural networks are a good tool for this problem, but I am just curious. Thanks for input :)
What you are looking for is a policy that maps state to action.
While you can use a neural network to store your policy you need some way to interact with the environment to collect data.
What you are describing is a typical Reinforcement Learning problem. I would suggest you have a look at Q-learning. For the size of the state space, you could easily store your policy in a table, but if you want, neural network are also easy to combine with Q-learning (though when using non-linear approximation schemes convergence is not ensured).