Artificial Intelligence meets Kreedz 1.6 - Brainstorming
Hello everyone.
I am actually brainstorming about project ideas which could unite any given aspect of the 1.6 KZ microcosmos with some type of AI. My favorite idea so far is to start a YT-Series, where I pick a map and create an AI-Bot which plays Kreedz. I would take a potential viewer on a journey how to program such a thing. But..
I would like to make use of the swarm intelligence of the XJ community and because of that I would love to as you, if there are any project ideas which you think would be interesting.
Kind regards, hefA.
I dont really see other area other than map completion tbh, somethong like this where you cam see the progress each ai makes wpuld be interesting to watch i guess: https://youtu.be/WSW-5m8lRMs
Im not too familiar with how ai work or what you will use tho. I guess it will be something like the TAS in speedrunning community, but im not aware as to how you implement such thing as i didnt get in that area of programming so far
Definetely would be super interesting to see how far map times can go tho
10/03/2021 21:42Very interesting idea. I'd like it if you elaborate a bit more on what specific things we could suggest to you because as the guy above me said, what other area then completing maps? Maybe specific things such as a optimally perfect bhop run on fastrun, or a completion of a very technical extreme map?
I'm quite unaware of just how strong you can make this A.I to be as I have a limited knowledge of this area. I know of Google's deep mind A.I project which was able to dominate the top chess engine after learning to play the game within 8 hours, however my understanding of this is that within that time period it had played millions of games against itself in order to achieve that high level of ability, and I don't know if you have the hardware capable of achieving such a thing.
I am extremely curious about your project!
A.I. learns how to LJ
Learn him to perform from 250 lj block to 258 (or 259?) with jumpoff 0.0x and landing 0.0x units every block seperately.. :D
I feel like climb and slide would be the most interesting ones tbh, basic lj and bhop would be just perfect stats all the time, like it would take little time to learn to constantly do 299 bhop pre and 100% sync strafes imo as there are scripts for that sort of stuff already, but taking the optimal route on some maps would be cool to see, maybe even discovering new routes
12/03/2021 13:53generative RNN based scroll emulator and strafe helper 
auto-generated kz maps. or making on existing map easier or harder, by shortening or lenghtening the distance between blocks.
14/03/2021 02:39This is a script generated map in csgo, by a guy from discord called iTs. It isn't really an AI but just a randomised block generator with some rules. So yea a basic kz map generator seems more than realistic
An AI that changes the difficulties in maps. Basically adds/removes blocks or makes the blocks bigger/smaller depending upon the difficulty chosen by the player (won't work on public servers for obvious reasons). This can in turn add so many new LAN-only routes to the records list and let people enjoy death maps on easy difficulties and vice-versa
15/03/2021 11:10^ I personally don't like these ideas. Sure they could be useful, but the beauty in A.I is showing the magnificence of its capabilities. That's what we saw in chess, and what we saw in DotA 2 with OpenAI where they were able to beat a lot of top players and clans (but ultimately lost against the best teams in the world). Now, 1.6 is a lot less complex then DotA 2. I would love to see the brilliance of an A.I demolishing a top player in a 1on1, or completing extreme maps effortlessly, like that map vly_nandu or whatever it is called. Showing the strength of the A.I as a player is what excites me, not creating maps.
16/03/2021 11:46Originally posted by pepi-back^ I personally don't like these ideas. Sure they could be useful, but the beauty in A.I is showing the magnificence of its capabilities. That's what we saw in chess, and what we saw in DotA 2 with OpenAI where they were able to beat a lot of top players and clans (but ultimately lost against the best teams in the world). Now, 1.6 is a lot less complex then DotA 2. I would love to see the brilliance of an A.I demolishing a top player in a 1on1, or completing extreme maps effortlessly, like that map vly_nandu or whatever it is called. Showing the strength of the A.I as a player is what excites me, not creating maps.
CS 1.6 has significantly more freedom and complexity of movement than Dota 2 does. It would be pretty hard for an AI to find the optimal route in most maps. An AI beating vly_nantu_d is a distant dream at this time. I'd love to see it though.
17/03/2021 07:39Originally posted by ChichinThis is a script generated map in csgo, by a guy from discord called iTs. It isn't really an AI but just a randomised block generator with some rules. So yea a basic kz map generator seems more than realistic
That map is garbage.
Guess we just got jbaited -,-
Hello folks,
thank you very much for sharing your thoughts! All ideas are very interesting, but as Flibo pointed out, a 3D environment can get pretty tough very fast.
Regarding to an AI which would target the limits: Please consider, that Bots like AlphaGo or AlphaStar cost milions of $ (hardware, energy, personell etc.), so creating an AI that would play better than the top players or even bad players is out of scope of ANY private project. But I really like the attitude and optimism of yours!
The big Problem that I see is the way of understanding the environment - I dont think it is easily possible to get to the engine of 1.6, so the bot wold have to interact with the environment through vision, which limits it's possibilities.
And the rest of ideas, like the auto generating of maps, Anti-Cheat, Auto-Map-Changer, LJ-Jumping Bot etc. partly suffer of similar limitations, but seem more realistic to me.
BTW : Are thery people who are also active in the AI/ML field and would like to participate in the project?
@hefakz any updates, after 4 years? xd
Please log in to post replies.