Support NeoGAF

[Digity.exe]

Possible deep dive into an upgraded AI within GT7?

 

[AJUMP23]

Sony Drivatars?

 

[Bernd Lauert]

Drivatars incoming?

 

[Calverz]

Hasan?

 

[SLB1904]

Oh boy

 

[alucard0712_rus]

What it has to do with in-game AI in GT7?
You can do a lot of things with AI and PD are nerds. I will not be surprised if they created some sort of a track generator using AI.

 

[M1chl]

Drivatar then, probably. Cool technology and it made perfect sense.

 

[Punished Miku]

That's a beautiful X in the title there. Impressive work.

 

[Anos Voldigoad]

Improved Ai? That dude who said bad shit about GT7 ai in the SoP thread never gave me samples of good racing ai in video game and now we get this hahaha. Say what you want, I'm all in!

 

[Digity.exe]

What it has to do with in-game AI in GT7?
You can do a lot of things with AI and PD are nerds. I will not be surprised if they created some sort of a track generator using AI.

Never suggested that's what it is exactly, only a possibility. It could be many things.

 

[onQ123]

Probably have more to do with the weather & stars

 

[Dream-Knife]

That's a beautiful X in the title there. Impressive work.

It's a trendy thing these days for collaborations. They're trying to make it sound high end. Leffot does this.

 

[sncvsrtoip]

Sony Drivatars?

please no, idea that just doesn't work, machine learning can be used not only to mimic wrong behaviour of players but also to have very fast ai with good race craft

 

[a multicultural team of v]

My guess would be that they ran all the data from their GT eSports league through an AI to create something. There is probably a shitload of data there, enough to cover nearly any situation or scenario that could occur during a race.

 

[Elios83]

It would be really interesting if advanced AIs based on actual human drivers found the way into GT7.
But it's better to wait and see what the announcement is about.

 

[AJUMP23]

please no, idea that just doesn't work, machine learning can be used not only to mimic wrong behaviour of players but also to have very fast ai with good race craft

Often with poliphony games the AI just runs the perfect line. I actually want an AI that will give me an opening due to being drafted or driver mistakes. Allow me to take an opening like Senna.

 

[alucard0712_rus]

I wish for A-B procedural 'tracks' or more open-road locations or ability to drive in procedural city. One can dream.

 

[Connificent]

Hasan?

 

[Andodalf]

NFTs

 

[alabtrosMyster]

I think that they will use the muscle deformation technology from spiderman to deform tires in real time!

 

[UnNamed]

I hope they fix the shitty penalties in GT Sport using the AI.

 

[SpokkX]

Hope this means the cars dont run on rails anymore

 

[Oddvintagechap]

I was thinking the other day, Sony collects alot of driving data. Why not create a game using google maps, collect even more driving data. Use it to power the AI of their cars.

 

[SLB1904]

I hope they fix the shitty penalties in GT Sport using the AI.

Don't ram and don't get rammed youll be ok

 

[Roxkis_ii]

My guess, not knowing anything, is Sony has made a self driving AI using the data from GT sport. They currently hold all record lap times in Forza Horizon 5.

 

[Loxus]

I'm curious to see how good this will turn out.

When looking at the Sony AI & PlayStation collaboration.

Sony's AI subsidiary is developing smarter opponents and teammates for PlayStation games
"Sony AI [...] has begun a collaboration with PlayStation that will make game experiences even richer and more enjoyable," say notes from a recent strategy presentation given by Sony CEO Kenichiro Yoshida. "By leveraging reinforcement learning, we are developing Game AI Agents that can be a player's in-game opponent or collaboration partner."


Reinforcement learning (RL) is an area of machine learning concerned with how intelligent agents ought to take actions in an environment in order to maximize the notion of cumulative reward. Reinforcement learning is one of three basic machine learning paradigms, alongside supervised learning and unsupervised learning.

In all, this sounds like some serious stuff.

 

[kingpotato]

I was thinking the other day, Sony collects alot of driving data. Why not create a game using google maps, collect even more driving data. Use it to power the AI of their cars.

What would Google maps data be used for? I wouldn't think it would be useful for racing and I hope you're not suggesting it's used for generating roads. If you end up with something like flight sim, no thank you.

Spoiler: Trigger warning for flight sim fanatics

 

[ethomaz]

It is PD helping with data for AI Deep Learning Study.
They are working with them for some time already.

BTW it is not Drivatar like because there is no human data inputed.
The first study is based in the physical rules of the car, track, and environment... the AI Become way too faster like a "super-human" beating humans by over a lap lol
The second study (probably that one they will show now) takes more things in mind... like the track, head-up display, force feedback, etc... to try to make the AI drive more like humans.

Polyphony Digital Helps New AI Deep Learning Study Using Gran Turismo

A new paper posted to the arXiv research repository has demonstrated the possibility of teaching AI how to race in Gran Turismo using only the same visual cues as human drivers. The research, conducted by Sony AI -- a recently created subsidiary of Sony to study applications for AI -- follows up

www.gtplanet.net

"In the previous study, the authors used a reinforcement learning method to teach an AI to go as fast as it possibly could based purely on its physical interactions with the environment. It learned what paths resulted in collisions and slower lap times, and eventually achieved "super-human" performance, beating even the best Gran Turismo Sport drivers over a lap.

The new paper focuses on visual cues instead. In essence the paper's lead author, Ryuji Imamura, proposes that the information conveyed while driving — the track itself, the head-up display, and force feedback — are sufficient for a human driver to learn how to drive quickly, and thus should be enough for an AI to learn to drive at human speeds."

The results of the second study.

 

[Oddvintagechap]

What would Google maps data be used for? I wouldn't think it would be useful for racing and I hope you're not suggesting it's used for generating roads. If you end up with something like flight sim, no thank you.

Spoiler: Trigger warning for flight sim fanatics

They can use the GT engine and use the mapping of Japan to basically use drivers inputteach the AI how to read the roads, since they don't own Uber. That game looks like that becase youre not suppose to be there, thats effectively the game's grass.

How about gameplay? i dunno. GT driving school? A game that teaches people how to drive along with your local driving laws? in VR?

Anyways just thoughts.

 

[ethomaz]

Sony Drivatars?

Not exactly.

Drivatars are based in player inputed data to generate the AI and that is why anything dumb players do the AI will learn lol

The Sony AI is based in the actual game data (track, car, physics, force feedback, etc) and not player.
The issue of the first study was that the AI become a "super-human" beating the best players in GTS by over a lap (that means even real life professional will be beat by over a lap).

Now there is the second study to try to make the AI become more human.

 

[sn0man]

I was thinking the other day, Sony collects alot of driving data. Why not create a game using google maps, collect even more driving data. Use it to power the AI of their cars.

Okay. This is smart. Maybe GTS is training data for Sony to make a car.

 

[ethomaz]

Okay. This is smart. Maybe GTS is training data for Sony to make a car.

That I believe is the end goal of the studies being done.

 

[yurinka]

GT7 secretly is Knack Kart game

 

[Anos Voldigoad]

 

[mckmas8808]

Super interested.

 

[Inviusx]

I'm curious. I have zero expectations.

My guess would be they are using player data to enhance the capabilities of their AI. Oh and then Kaz is going to offer some pie in the sky comment like "we may even be able to use this data to enhance the AI of our drivers in future GT titles" or something like that.

I'm getting Folding at Home vibes from this. Who here remembers Folding at Home?

 

[ReBurn]

The real reason online is required probably

 

[SilentUser]

I'm curious. I have zero expectations.

My guess would be they are using player data to enhance the capabilities of their AI. Oh and then Kaz is going to offer some pie in the sky comment like "we may even be able to use this data to enhance the AI of our drivers in future GT titles" or something like that.

I'm getting Folding at Home vibes from this. Who here remembers Folding at Home?

I do I'm also curious as it could be something cool, but I'm not going to get hyped, though.

 

[GHG]

Needs to be some kind of improvement because from what I've seen so far the AI looks the same as always - pretty shit.

I'm buying it primarily for Sport mode so don't really care either way but a surprise would be nice.

 

[Tommi84]

Don't know what it its, but sonsofbitches, I'm IN!!

 

[SLB1904]

I do I'm also curious as it could be something cool, but I'm not going to get hyped, though.

it must be something cool right? or why would they make a special announcement for it?

 

[CamHostage]

Not exactly.

Drivatars are based in player inputed data to generate the AI and that is why anything dumb players do the AI will learn lol

The Sony AI is based in the actual game data (track, car, physics, force feedback, etc) and not player.
The issue of the first study was that the AI become a "super-human" beating the best players in GTS by over a lap (that means even real life professional will be beat by over a lap).

Now there is the second study to try to make the AI become more human.

It's interesting what people consider "bad AI", because, pretty obviously, game developers could make perfectly great AI which would whip your ass day and night whenever you picked up a controller. People want fair AI (but AI that they can still beat more often than not) and they want AI that seems like it's doing what a person would do against them, even if they're doing something a crazy person would do if this were anything but a video game. They want "Realistic AI," by a gamer's definition of realistic (which usually involves a lot of cheats of reaction/expression animations & routines to sell the idea of AI being active, or in this case a number of added factors that temper the AI's skills); "Good AI" has, oddly enough, rarely had a place in gaming.

So in GT, the game designers don't want to design around crashes and bad driving lines because that's "bad AI" for a racing experience on paper, but then they have to because game players just go pinballing through the field (which is partly because they're bad drivers and partly because most games don't have a permanent damage punishment stat, but also partly because a video game will inherently only allow so much presence of awareness of the vehicle when played off of a flat TV several feet away; VR could help but it's still an abstraction) and rarely maintain a perfect line even when they're skilled enough to drive at that level. Good opponent AI in a realistic racing game would be something that constantly is going, "WTF is up with this maniac?"

 

[Inviusx]

It's interesting what people consider "bad AI", because, pretty obviously, game developers could make perfectly great AI which would whip your ass day and night whenever you picked up a controller. People want fair AI (but AI that they can still beat more often than not) and they want AI that seems like it's doing what a person would do against them, even if they're doing something a crazy person would do if this were anything but a video game. They want "Realistic AI," by a gamer's definition of realistic (which usually involves a lot of cheats of reaction/expression animations & routines to sell the idea of AI being active, or in this case a number of added factors that temper the AI's skills); "Good AI" has, oddly enough, rarely had a place in gaming.

So in GT, the game designers don't want to design around crashes and bad driving lines because that's "bad AI" for a racing experience on paper, but then they have to because game players just go pinballing through the field (which is partly because they're bad drivers and partly because most games don't have a permanent damage punishment stat, but also partly because a video game will inherently only allow so much presence of awareness of the vehicle when played off of a flat TV several feet away; VR could help but it's still an abstraction) and rarely maintain a perfect line even when they're skilled enough to drive at that level. Good opponent AI in a realistic racing game would be something that constantly is going, "WTF is up with this maniac?"

You're right, when people say bad AI they usually mean boring AI, which perfectly sums up the AI racers in GT.

The F1 games have it right. Firstly there are multiple tiers of difficulty so you can tune how difficult you want it but most importantly it attacks/defends and also makes mistakes. All things lacking with GT AI.

 

[Tiago Rodrigues]

Gran turismo 7s hype is honestly hitting me out of nowhere since that state of play.
Let me see what this is about

 

[01011001]

let's hope it's not just another Drivatar moment... while those are kinda fun, they are ultimately very underwhelming to say the least

 

[captainraincoat]

Monza t-bone….bought to you by the cloud

 

[mclaren777]

Improved Ai? That dude who said bad shit about GT7 ai in the SoP thread never gave me samples of good racing ai in video game and now we get this hahaha. Say what you want, I'm all in!

I wasn't the guy you're talking about so please forgive me for jumping in, but I'd say that rFactor 2 probably has the best AI in the business.

Edit: I also think it has the best physics/FFB because it accurately models the slip angle of tires, but that's a different conversation entirely.

 

[Anos Voldigoad]

Hyped for the reveal!

 

[Anos Voldigoad]

 

[Thirty7ven]

Challenging A.I is easy, what's hard is for them to behave realistically.