AI in RoR

[anarchypark]

military drone AI killed pilot, then antenna, in simulation.
it'll kill WB leader.

[rejndjer]

Clearly, fixed (and dumb as we learned today) if then else commands are AI.... @KPI

Considering how reinforcment learning trains and how slow it is, we would have to run thousands of servers in parallel or wait a few years (like 10 maybe) for it to make its first reasonable decisions... And with that i mean not stutter walk into a wall.

You could try and automate certain things like building groups, but why make the effort if you can simply base party buliding on rules.

hmmm AI was beating dota pros in 1v1 dota after only 2 weeks of "playing" dota back in ~2017 already. next year it came back and defeated best of pros in 5v5. but obviously dota is a game for noobs, while only biggest masterclass profesional player can play ROR

[Sulorie]

Clearly, fixed (and dumb as we learned today) if then else commands are AI.... @KPI

Considering how reinforcment learning trains and how slow it is, we would have to run thousands of servers in parallel or wait a few years (like 10 maybe) for it to make its first reasonable decisions... And with that i mean not stutter walk into a wall.

You could try and automate certain things like building groups, but why make the effort if you can simply base party buliding on rules.

hmmm AI was beating dota pros in 1v1 dota after only 2 weeks of "playing" dota back in ~2017 already. next year it came back and defeated best of pros in 5v5. but obviously dota is a game for noobs, while only biggest masterclass profesional player can play ROR

Tbh, we have nowhere near the amount of computing power they used to train the AI.
Furthermore ROR doesn't have only top-down perspective and more complex controls in addition to a lot more active abilities.

[rejndjer]

Clearly, fixed (and dumb as we learned today) if then else commands are AI.... @KPI

Considering how reinforcment learning trains and how slow it is, we would have to run thousands of servers in parallel or wait a few years (like 10 maybe) for it to make its first reasonable decisions... And with that i mean not stutter walk into a wall.

You could try and automate certain things like building groups, but why make the effort if you can simply base party buliding on rules.

hmmm AI was beating dota pros in 1v1 dota after only 2 weeks of "playing" dota back in ~2017 already. next year it came back and defeated best of pros in 5v5. but obviously dota is a game for noobs, while only biggest masterclass profesional player can play ROR

Tbh, we have nowhere near the amount of computing power they used to train the AI.
Furthermore ROR doesn't have only top-down perspective and more complex controls in addition to a lot more active abilities.

dota is 100000x more complex game than ROR.

[JohnnyWayne]

That doesnt matter, dota as well as sc2 had API changes made for the algorithm to setup upon. Also, as mentioned before, they had a WHOLE LOT of computing power at their disposal. If you have that kind of computing power, still, those algorithms didnt train in real time but in a sped up version of the game and in parallel.

https://www.youtube.com/watch?v=BTLCdge7uSQ

Nothing about this is easy. Considering how many maps ror has and how they are not limited to pvp zones, I'd argue you being wrong about the complexity of the problem. Also there is no real goal defined in ror that an agent could target.

[Sulorie]

Tbh, we have nowhere near the amount of computing power they used to train the AI.
Furthermore ROR doesn't have only top-down perspective and more complex controls in addition to a lot more active abilities.

dota is 100000x more complex game than ROR.

You must be kidding or you didn't read the response completely and stopped at "complex".
DotA is basically a 2D game with 3D models. With significantly less abilities. Letting an AI move and make decisions in a 3D world is an entirely different beast and it is already mind-blowing in DotA.

[Cleo]

Funny thread. But, suggesting that some people who work on an alpha stage of a game in their free time since years to implement a new, highly sophisticated technology that requires a lot of resources while there is an open, years old backlog of minor and major bugs that need to be fixed is not only hilarious, it's borderline stupid.
For the next time, you could try suggesting that they put the server on wheels and see if they can shoot it to mars to drive a couple of rounds there.

[JohnnyWayne]

Tbh, we have nowhere near the amount of computing power they used to train the AI.
Furthermore ROR doesn't have only top-down perspective and more complex controls in addition to a lot more active abilities.

dota is 100000x more complex game than ROR.

You must be kidding or you didn't read the response completely and stopped at "complex".
DotA is basically a 2D game with 3D models. With significantly less abilities. Letting an AI move and make decisions in a 3D world is an entirely different beast and it is already mind-blowing in DotA.

It is about the degrees of freedom you have in the environment of an agent. We have a lot more freedoms in mmorgps: 3d worlds, non pvp zones without borders, allies and enemies classes, their respective skills, number of players, locations that are uncommon and what not.

Funny thread. But, suggesting that some people who work on an alpha stage of a game in their free time since years to implement a new, highly sophisticated technology that requires a lot of resources while there is an open, years old backlog of minor and major bugs that need to be fixed is not only hilarious, it's borderline stupid.
For the next time, you could try suggesting that they put the server on wheels and see if they can shoot it to mars to drive a couple of rounds there.

The skillset you would require to do some sophisticated AI stuff in any game is a different one from game design / development, even if it is overlapping (especially from a performance standpoint - you need to evaluate frames into actions in real time, or at least next to).

[wonshot]

Having bots added to private servers is actually not something completely new.
The DaoC Eden freeshard is the first of the many versions of the older MMO DaoC where they managed to get AIs in, to help with action in case of low population on a global timezone server.

Video of some players fighting against these AI:
https://www.youtube.com/watch?v=_aTdGTB20iI

Now the bots do some basic level of assisting, debuffing, controling and attempting to kite.
I dont think the collision, or the punting/pulling mechanic would make such bots easy to behave on RoR. And honestly, I dont think them being fuctional with pathing would work over so many maps as RoR orvr offers.

The time investment simply would never be worth it I would think. But its a fun idea to think about atleast.

[rejndjer]

Tbh, we have nowhere near the amount of computing power they used to train the AI.
Furthermore ROR doesn't have only top-down perspective and more complex controls in addition to a lot more active abilities.

dota is 100000x more complex game than ROR.

You must be kidding or you didn't read the response completely and stopped at "complex".
DotA is basically a 2D game with 3D models. With significantly less abilities. Letting an AI move and make decisions in a 3D world is an entirely different beast and it is already mind-blowing in DotA.

you cannot be better than AI in mechanical skill. no matter how complex the game is - mechanically - if humans can do it, so can AI. and AI is gonna do it million times better because it literally has million APM. luckily this is a 40apm game, so if you put AI in ROR he wouldn't be able to be better than humans anyway. it's kind of like call of duty games or league of legends. these games are very low-skill-ceiling like ROR, you could add AI there and it would be 50-50 as to who is gonna win. in dota/cs/unreal/quake/starcraft there's no way a human can beat an AI.

P.S. i get it that you need a lot of knowledge and resources to implement this, im just trying to say that in theory if we had these resources it would've been good addition to ROR. also another thing you're forgetting about dota is not that it is just mechanically more intense than ROR, it's also more intense in terms of decision-making. which exists in dota, unlike here. nhf everyone.

https://www.youtube.com/watch?v=pkGa8ICQJS8&t=136s here's a nice little video of AI completely dominating the best dota team of that year. 4 years ago.