By this I mean games that already have online bots like Unreal, Quake, Dark Forces II, Action Half Life, Vampire Slayer, Wizard Wars, The Opera etc. Because such a mod would be very useful for unpathed maps as the bots could just learn the routes themselves. Also incredibly useful for Dark Forces II were most user made level are not mapped for RBots. Why hasn’t anyone made something like this yet?
there was a quite insane ai bot for starcraft2 that did this
That’s good but with the way AI technology would it be possible to improve the bots in the games I mentioned? If so why hasn’t anyone done this yet?
Because
a) it's hard
b) there's not enough demand to make this feature a game changer
There is tech to improve AI since the days of STALKER/FEAR, but it's something really asymmetrical in the payoff/technical implementation ratio, so most studios only do the absolute necessary.
100% and I wish games did this more. Was the Smash Bros amiibo thing a machine learning gimmick?
>b) there's not enough demand to make this feature a game changer
That's the big one. People always ask why muh games don't do X, Y, and Z and it's mostly because of demand and time/cost crunch. I own a Valve Index and can play WWII D-Day maps whenever I want and yet only like 30 people play that game mode. One of the most innovative things but consumers don't care for it. Turns out they don't want cutting edge stuff.
Just wanna add on, PvE is such an unappreciated aspect of games. Look up anything about bots or PvE for any new game, and realize you're the only one who cares. Sucks. PvE is awesome. Favorite game ever is bot survival in mount & blade napoleonic wars, even though the bots are kinda bad.
Absolutely, like i said earlier for DRG it would a be a great thing.
I was talking about modding it into those games like people create character AIs or bots having to be modded into Dark Forces II with RBots.
The big problem would be mapping the model to the game. Most ML models use the python runtime to execute a model. Though it looks like some game engines have some native support for this stuff
RBots released their source code publicly would it be possible to work that to use such machine learning?
Anon, I'm still a git when it comes to ML, but I suspect that the stuff you want to do is quite computationally expensive. It's better not to get fancy and adjust accordingly with a robust and tested method like Utility AI (use in RTS due to efficiency and quite versatile).
People don’t have to pay much or at all to make character AIs but still RTS AI would probably cover what I am after. It would allow the bots to explore and learn the layout of levels without waypoints right? And it could be programmed into something like RBots?
UAI is quite "sandboxy", yes. It's not that suited for scripted events, but if you want a scenario where you drop the bot and it "decides" whatever the hell it wants in order to satisfy its needs is perfect, more on the "emergent behavior" type of stuff.
And would it be possible to program this into something like RBots?
If you want to improve bot behavior, sure, but you would have to adapt the code. I have no idea what architecture they used. It seems pretty crude, though (judging that it has more than 20 years since release).
They have the source code released online, it’s coded using COG the programming language used in DF2JK
Have you tried to use the SDK from RBots to include it in your own maps?
I don’t really have any maps of my own it’s more that I want to explore the vast wealth of DF2JK maps that are out there and do more with them than just walk around them on my own. I also don’t know if I could add waypoints to every map I download. So I wondered why hasn’t anyone improved bots like this using the latest AI technology or at least more advanced AI technology. Surly it is possible and it would at least be interesting to see the results.
because i dont want.
give me some waifus and i will think it.
They’re are plenty of waifus in these games Unreal and Quake have some had battle maidens, Action Half Life has some hot leather matrix style babes, Vampire Slayer has hot female vampire Slayer and best of all Dark Forces II has Mara Jade.
Sure. Probably jusr gather data about user inputs and world conditions, make a model that can predict the inputs and viola. Just generate a bunch of static behaviors, save and label the ones you want to use and hook them up to the game actors.
Would you have to do this for every single level though or could it be done in a way that would allow the bots to learn the structure of a any level without the data being gathered first?
yeah just have AI train against each other until they reach a high skill level. If you train them too long they might be too challenging so you have to find the right balance. It’s totally doable and I expect to see it in Elder Scrolls 6
It might make them more skilled but that wouldn’t give them the ability to transverse levels on their own
>Why hasn’t anyone made something like this yet?
Have you ever heard of GOAP?
>Goal-Oriented Action Planning (aka GOAP, rhymes with soap) refers to a simplfied STRIPS-like planning architecture specifically designed for real-time control of autonomous character behavior in games. I originally implemented GOAP for F.E.A.R. while working at Monolith Productions. This A.I. architecture simultaneously powered Monolith's Condemned: Criminal Origins. (Brian Legge was responsible for the A.I. in Condemned). My GOAP implementation was inspired by conversations within the A.I. Interface Standards Committee's (AIISC) GOAP Working Group, as well as ideas from the Synthetic Characters Group's C4 agent architecture at the MIT Media Lab, and Nils Nilsson's description of STRIPS planning in his AI book.
https://alumni.media.mit.edu/~jorkin/goap.html
It's worth a look.
Also,
Small showcase (2min): https://www.youtube.com/watch?v=0Y1EGevHg1s
Concept Explained (9min): https://www.youtube.com/watch?v=qaIF0iaDWRQ
>Can Modern Machine Learning AI be Used to Make Better Bots for Games?
Generally, no. RL works like shit on everything, but toy tasks.
there was a bot for spring rts, named aai iirc
it was quite stupid for first several matches, then he learned how to play a game and become better and better
wasn't there a starcraft 2 AI bot that literally destroyed the shit out of some proplayers?
Yes. AlphaStar by DeepMind (Google).
https://www.deepmind.com/blog/alphastar-mastering-the-real-time-strategy-game-starcraft-ii
You're probably referring to reinforcement learning.
>Why hasn’t anyone made something like this yet?
Because it's incredibly hard to implement, they managed to train a bot to play old Atari games using reinforcement learing:
because the results would be derp. you get better results by coding behavior by raycasting shit on the map but i still wanna try this someday
>Because it's incredibly hard to implement, they managed to train a bot to play old Atari games using reinforcement learing:
lol no it isnt. i made a bot that could play breakout too plus this shit is old as fuck
I would be happy with derp if the bots could navigate the map on their own and in fact a certain amount of derp is what gives bots a certain charm. The bots that came with skulltag are down right retarded but that just makes them hilarious and I wouldn’t have them any other way.
How big is the DF II mp base? I found a RBot link that was old as fuck (2002) and no info on its architecture, maybe I could try to implement something easy but modern and better if the code is open source.
The code is open source and the DF2JK map base is in the thousands but nobody really plays multiplayer anymore hence the need to rely on bots. Which is ok I like playing with bots but the thing is that most maps are not designed for bots. So I was wondering if there is an AI solution to this.
If I had something to release and I wanted visibility, where should I publish?
Never mind, just found the repo for OpenJKDF2.
There’s also The Massassi Temple and DF2JK.net for general mods, maps, files and projects
people have done it, but the ROI is too low. pretty sure the bots that defeated dota 2 champs used machine learning
Too much work, you’re asking ok much from devs who today don’t want to compress their files or make actual levels because it’s “too hard for them” and instead opt for large file sizes and empty worlds respectively
Not talking about it a feature in modern games but modding it into the games mentioned above which either have mods that change bot behaviour already or have mods that ad bots in. Thus conceivably providing a base to be programmed to have some kind of ML
Test
Test? A test of what?
Yes it just takes time, money and hardware to train an AI.
already exists.
Does it? Where?
>Can Modern Machine Learning AI be Used to Make Better Bots for Games?
Yes
>Why hasn’t anyone made something like this yet?
Because nobody fucking cares plus you don't even need machine learning to make good game bots
Not to make good bots but perhaps you might to improve existing bots allowing them to explore maps without them having to be pathed
>allowing them to explore maps without them having to be pathed
what does that mean? why is it advantagous to build a pathing map with AI instead of traditional methods?
Because there are so many maps that are unpathed, this especially a problem with Dark Forces II where most of user made levels are unpathed.
what does "pathing" mean?
Usually, bots are retarded and need help navigating the map through predefined 'paths'. Pic related. So it is necessary to run waypoints (colored marks) all over the place for the bot to be able to "see" where he can go.
Just to complement it: Like
said, this could be done automatically, but the result can be a bit... rudimentary.
before they became fags openai beat the top dota Team handily with their bots
You don't need AI to do that
It's trivial to generate pathing maps automatically
Is it? I thought you need to set them using a level editor or an sdk of some kind.
No, most games generate pathing data automatically
Yes, but this is not the case with Dark Forces II, Jedi Academy, Unreal Tournament, Action Half Life, Wizard Wars etc.
So why would anyone bother putting the time and effort into seeting up an AI system to generate pathing maps for old games nobody cares about
Because those old games are actually good unlike modern games which are dumbed down and shit. Plus those games have a wealth of brilliant levels that people have crafted and if you want to play them you need bots.
That's your personal opinion
Most people don't give a shit about them, so they aren't going to bother
And if they did bother there's easier ways to do it than using machine learning
Most people as in zoom zooms?
What easier ways are those?
I played those games when they came out and they definitely aren't worth playing now
Easier way would be to create a tool to edit the compiled maps and place them manually
Did you really favourite Wizard Wars, Jedi Academy or Dark Forces 2 map?
There are thousands of DF2JK levels the idea of having improved AI is so as to have to edit each one manually
You could make a procedural tool to generate the pathing maps
If you cared about what most people think you would not be here.
There isn't much prestige involved in "helping" those games, anon. So if isn't going to be a labor of love from a player that genuinely likes the game (and probably plays it), it's simply not happening. I mean that's how it usually does, but there's definitely some autists out there that would could crack a new "AI" in a week. I'm going to try with Dark Forces II since there's an open source version and some work already done, but I'm no wizard and at any point if it presents itself as some daunting task, I'm out.
Its good your giving it a shot, I wish you luck. If you have any difficulties it might not be a bad idea to ask for advice on Massassi Temple or JKDF2.net. If you make any progress those would probably be the best places to post as well, it will be interesting to see the results.
>Massassi Temple or JKDF2.net
Thanks, I'll be posting there.
it would be nice if there was a way to add ai team mates to any given game that has multiplayer built in for solo
(referring to DRG)
But maybe it'll get there eventually, eh?
anyone have that old ass quake copypasta of the guy who left he server running and the bots just stopped fighting eventually?
It's not something that's easy to do, first you need to decompile the map format which is a pain if it's BSP, then you need to create a tool to do some geometry analysis and build a pathing map, which is hard to do without gamedev experience
It's not that hard, really