MX Simulator

Have anyone worked/ are working or thinking of working on some machine learning, presumably a neural network for mx simulator? I got the thought a while back that it would be funny to write a neural network to learn to play mx simulator, and eventually, as you would expect, get very good at it, and that way you maybe could draw some knowledge from it.

Edit mx.exe to mx.exe --learn to teach the AI robots.

Then do --practice to make them practice the lines.

How does the teaching and learning actually work? Like, do I have to load into a game with bots and run a race, and then they copy me?

This subject interests me a lot. A little explanation would be good.

Jrife548 wrote:How does the teaching and learning actually work? Like, do I have to load into a game with bots and run a race, and then they copy me?

The bots just test the different lines the real player has completed. There's no neural network or anything like that.

jlv wrote:

Jrife548 wrote:How does the teaching and learning actually work? Like, do I have to load into a game with bots and run a race, and then they copy me?

The bots just test the different lines the real player has completed. There's no neural network or anything like that.

Would it be easy to improve them as: the line that seem to work the best between timing gate a-b if the preferred line? To make "the best lap"?

Wahlamt wrote:Would it be easy to improve them as: the line that seem to work the best between timing gate a-b if the preferred line? To make "the best lap"?

That's what it does now. Although it's not necessarily per gate. If you look at the stock tracks or my sx tracks you'll notice some gates have 100000 added to them. That tells the bots not to switch lines at that point. Seems best to let them switch on the entry for each corner.