Can I change tickrate and remove GUI on my own server?

[asdfgzxcvb]

I am wondering about making an AI that can play 2009scape on my own server. I understand official support is not-existent for my own server.

Like Blizzard's Starcraft and Google Deepmind's AlphaStar, they trained their AI model on a custom version of the game Starcraft where AI agents can play a version of the game that runs faster (has a quicker tickrate).

For example, say I want an agent (a bot, lol) that learns how to do quests. Instead of having it do trial-and-error learning on a 0.6 second game tick-cycle, is there a way to change the game tickrate to 0.06 seconds? This way, the AI agent can perform actions and learn faster because it can do more actions in less time.

Likewise, as far as the GUI goes, since cloud GPUs have a lot of latency and will not work for performance for my AI, I need to run a Tesla A100 GPU locally on my linux box. I am wondering if there is a 'archlinux' version of 2009scape that could be ran and actions could be performed without needing to run much graphics at all.

How bare-bones can the game 2009scape become?


Hausknecht, Matthew, Joel Lehman, Risto Miikkulainen, and Peter Stone. (2014). “A Neuroevolution Approach to General Atari Game Playing.” IEEE Transactions on Computational Intelligence and AI in Games 6 (4): 355–66. https://doi.org/10.1109/tciaig.2013.2294713. https://www.cs.utexas.edu/users/pstone/ ... hauskn.pdf

‌UCF. (2015). The Hybercube-based NeuroEvolution of Augmenting Topologies (HyperNEAT) Users Page. Evolutionary Complexity (EPlex) Research Group at the University of Central Florida. http://eplex.cs.ucf.edu/hyperNEATpage/HyperNEAT.html. http://eplex.cs.ucf.edu/

Vallade, B., David, A., & Nakashima, T. (2015). Three Layers Framework Concept for Adjustable Artificial Intelligence. Journal of Advanced Computational Intelligence and Intelligent Informatics, 19(6), 867–879. https://doi.org/10.20965/jaciii.2015.p0867

[gregf36665]

You could go and modify the server tick duration to any arbitrary duration but weird things could start happening: https://gitlab.com/2009scape/2009scape/ ... ckUtils.kt

As far as the client goes it depends if you want to train your AI on the client or just train it on what packets it needs to send. You might be able to modify a client to train your AI but again all bets are off on what will work. Technically you might want to just look at packets to figure out how to train your AI.
https://gitlab.com/2009scape/rt4-client

Of course it goes without saying that this is so far beyond what was expected of this project that all bets are off when it comes to support.

Good luck and be careful, if your AI starts running on the live server it’ll be banned for botting.

[asdfgzxcvb]

It seems choosing an objective function or setting an objective (like complete quests) is the wrong approach. It seems to be better to use HyperNEAT algorithms choose sequences of actions and just see what interesting ways an AI plays the game contra human gameplay.

Kenneth O. Stanley and Joel Lehman. 2013. Novelty Search and the Myth of the Objective (at EPIA 2013). https://youtu.be/JlSLJ8ciHDE?t=3880

You could go and modify the server tick duration to any arbitrary duration but weird things could start happening: https://gitlab.com/2009scape/2009scape/ ... ckUtils.kt

As far as the client goes it depends if you want to train your AI on the client or just train it on what packets it needs to send. You might be able to modify a client to train your AI but again all bets are off on what will work. Technically you might want to just look at packets to figure out how to train your AI.
https://gitlab.com/2009scape/rt4-client

Of course it goes without saying that this is so far beyond what was expected of this project that all bets are off when it comes to support.

Good luck and be careful, if your AI starts running on the live server it’ll be banned for botting.

[Red Bracket]

I've seen someone try something similar with 2006rebotted. The problem was at a high tickrate, the game became nonfunctional. At least if you'll be able to run thousands of bots at once.

[asdfgzxcvb]

It would be interesting to have a group of bots that can only trade amongst themselves like group ironman. Thus, emergent gameplay and surprising strategies might be found.

I've seen someone try something similar with 2006rebotted. The problem was at a high tickrate, the game became nonfunctional. At least if you'll be able to run thousands of bots at once.

[daymansalvort]

Yes, you can change the GUI.