Bots for Auidosurf

[DevilDimon]

I'm planning to write a bot for this game for my machine learning course at university.
I have two questions for you guys:
1) Has anyone ever written a bot for Audiosurf (1 & 2)?
2) Do I get banned if I inject a bot into the game? I'm not planning to ruin leaderboards, I'm only interested in testing the bot, so I'll do it offline. If I do, is it going to be a VAC-ban or Audiosurf-only-ban?
Let me know what you think. Any help in programming the actual bot behaviour would also be appreciated.

[skzk]

2) There is no bot detection system like VAC as far as I'm aware but users can be banned for cheating.

[Kyanite]

This is possible to do, because I have tried to create something similar and it is already ingame, but is called a minion. It is seen in gamemodes such as "Turbo" and "Nitro".

And you are allowed to do create this bot-thingy as long as i don't get unto the leaderboards. So I would recommend you to download a few MP3s from the internet and use those, since they are mostly encoded with almost no details. Or you could do it offline.

[skzk]

Originally posted by Rusty!:

a minion.

I don't think that's what he meant.

[Kyanite]

Originally posted by skzk:

Originally posted by Rusty!:

a minion.

I don't think that's what he meant.

@skzk - The minion in game is an automated bot that is activated through the left mouse button. With a litte tweaking it could be the player ship that moves on its own without the player doing anything. But yeah this is the easiest way to create what I think he means... xD

[Poonker]

But you know, writing a script that will just catch all the blocks won't give you high scores, right?

[Kyanite]

Originally posted by Poonker:

But you know, writing a script that will just catch all the blocks won't give you high scores, right? :Original_Assassin:

Doing the script right you can program the bot to catch colored blocks only and avoid greys.

[''Y'']

Originally posted by Rusty!:

Originally posted by Poonker:

But you know, writing a script that will just catch all the blocks won't give you high scores, right? :Original_Assassin:

Doing the script right you can program the bot to catch colored blocks only and avoid greys.

Which is useless in normal Mono, as it is very often necessary to hit a grey in order to reset the match timer so you can maximize your cluster size, or to keep a big cluster on the grid for an approaching powerblock. And it's rather useless in Ninja, as it is often best to skip certain colors to avoid force clearing so you can get a bigger cluster, especially for approaching powerblocks.

So unless the bot can determine when to do those things, it is indeed generally useless for high scores (unless of course the scoreboard is filled with noobs :p).

[DevilDimon]

That is exactly where neural networks come in.

[DevilDimon]

Originally posted by Luzzifus:

Actually I think this is not a very good use case for neural networks.

Thank you for the reply, great food for thought. I'm still in the process of choosing a game to work with, and Audiosurf was not very high on the list for a completely different reason (the algorithm for real-time track detection).

[Luzzifus]

Originally posted by DevilDimon:

That is exactly where neural networks come in.

Very interesting approach. But if you're going for a NN, a bot would slow the learning process down (bot as in an algorithm that actually plays the game). Alternatively you could write a score simulator, since AS2 allows you to get the exact timing of all the blocks and other objects on a track for each generated song/run (theoretically, by writing a mod or altering an existing one). Also the game mechanics are simple, well documented (mod script code) and/or well known. This would help tremendously speeding up the training process.

One difficulty in training for AS would be overfitting. You have to make sure that your NN is not learning single tracks, but actually general technique. That was my main problem when I thought about something like this some time ago: What would be the best input definition for the NN, which would make sure that enough of the track is observed to actually be able to optimize on the one hand, and not specializing on trained songs only (=overfitting) on the other hand?

Originally posted by DevilDimon:

Thank you for the reply, great food for thought. I'm still in the process of choosing a game to work with, and Audiosurf was not very high on the list for a completely different reason (the algorithm for real-time track detection).

Oh you read that faster than I could delete it. ^^
Well I think it might be possible but still think it is rather hard to get a NN working for AS2. If you like the challenge, I'd really like to know the outcome!

[†нэ$†øиэd$pîdэя]

You know if you are writing this just for a grade in class why don't you work with the developer so he can, if he's willing, write a bot detection program and that way you both can test off of each other. You get your grade he gets his cheater recognition.
On the other had if this isn't for a class or grade or whatever and you just want to try to get high scores or feel useful by being on the top scores then you seriously need a better life lol. but im sure its just for a class and grade.