Gran Turismo Sophy: Sony AI x Polyphony Digital

  • Thread starter Magog
  • 1,719 comments
  • 194,217 views

This interview was published in January and the developers mention endurance races with dynamic weather as one of their goals for Sophy.

I hope we learn some new things as they sure showcase it on the World Tour events.
Thank you for sharing this, that sounds hopeful!
 
It looks like Sony AI is simultaneously working on another AI model (not Sophy) potentially for use in Gran Turismo.

Essentially it appears to be trained using "vision" such as what we see when playing the game, rather than being fed input data (like slip angle and ratio, wheel load, etc) the way Sophy is. It has currently only been trained in time trial settings, but it sounds like they plan to train it in more of a race setting in the future.

Full details in this Traxion.gg article.
 
It looks like Sony AI is simultaneously working on another AI model (not Sophy) potentially for use in Gran Turismo.
Basically this is news from 2021:


The research predates Sophy and Sony AI, but became part of Sony AI when that formed. I saw the most recent paper was published in August, and it had expanded the track selection to three rather than just the original one, and while still better than the best human players it wasn't quite as quick as Sophy.
 
Basically this is news from 2021:


The research predates Sophy and Sony AI, but became part of Sony AI when that formed. I saw the most recent paper was published in August, and it had expanded the track selection to three rather than just the original one, and while still better than the best human players it wasn't quite as quick as Sophy.
Interesting, thanks. The article made it seem like it was a new interview. Cool to see in the paper summary that it's already being trained with weather on those three tracks.
 
Interesting, thanks. The article made it seem like it was a new interview. Cool to see in the paper summary that it's already being trained with weather on those three tracks.
Well, the paper itself is definitely new (well, August), and I think they were at some AI awards event so Dr. Subramanian might have said something at it, it just looks like the same vision-based RL agent from 2020 with more development rather than a new addition alongside Sophy.
 
That article seems quite a bit different than anything I'd seen. At least through the screen shots it was provided so it's got to be somewhat new.

The discussion about feeding it visible information vs precise track data is a warm welcome for what else is in store for the new AI in GT though
 
That article seems quite a bit different than anything I'd seen. At least through the screen shots it was provided so it's got to be somewhat new.

The discussion about feeding it visible information vs precise track data is a warm welcome for what else is in store for the new AI in GT though
Here's the pre-Sony AI paper, with the AI trained on visual cues:

 
This is the information I was talking about. Outlined in this paper here.

Yes, that's the paper presented at the conference in August; it even cites the 2021 Fuchs paper in the Related Work section (naturally :lol: )

As far as I can tell, the difference between the 2021 version of the vision-based reinforcement learning agent and the 2024 one is that it dispenses with what the paper calls "forward-looking" or "global features" - knowledge of the course layout is given as an example - and only uses local information (what it can see right now), so that the agent is mimicking human proprioceptive behaviours.
 
Fascinated by the technology, but must admit that it is a slippery slope when this matures enough that anybody at home can use it – camera pointed at screen, device hooked up to a controller… Equals basically auto grinding credits at one end, and the end of competitive online time trials on the other? Pros and cons, for sure.
 
Fascinated by the technology, but must admit that it is a slippery slope when this matures enough that anybody at home can use it – camera pointed at screen, device hooked up to a controller… Equals basically auto grinding credits at one end, and the end of competitive online time trials on the other? Pros and cons, for sure.
You could already do that.
Sophy - with other cars on the track at the same time - is a step further than time trialing.
Also Sophy doesnt suffer from reaction and input delay as any user-solution would, because user can not run his AI-Agent direclty on the console but requires external devices to achieve the same.
 
Fascinated by the technology, but must admit that it is a slippery slope when this matures enough that anybody at home can use it – camera pointed at screen, device hooked up to a controller… Equals basically auto grinding credits at one end, and the end of competitive online time trials on the other? Pros and cons, for sure.
Someone has already done this with Lego Technic… I’ll see if I can find the YouTube link

Here it is:
 
Last edited:
I wasn’t saying it was impossible for the auto-grinding, that’s kind of irrelevant in my opinion. I was more going after the machine learning AI of a home user setting time trial lap times that a human driver have difficulty beating. That would be a conundrum to solve. How do you know if someone is as good as their lap times claim to be? Just taking the edge cases into consideration is all 😄👍🏻
 
I wasn’t saying it was impossible for the auto-grinding, that’s kind of irrelevant in my opinion. I was more going after the machine learning AI of a home user setting time trial lap times that a human driver have difficulty beating. That would be a conundrum to solve. How do you know if someone is as good as their lap times claim to be? Just taking the edge cases into consideration is all 😄👍🏻
You dont know unless they show evidence of them driving (as long as it doesnt look toooooo perfect).
TAS (tool assisted speedrunning) is a thing in Trackmania and theoretically could be applied to all racing games that are deterministic (which is the case for GT7 time trialing, as weather and wind dont change, nor does fuel or tyrewear).
The only thing (in my opinion) preventing this is GT7 being a console game, which causes reaction and input delay to every external device (as said before), as you cant run your ML AI on the console ;)
 
I was having a great battle with the newer generation of M3 and it got ruined after the first lap.
I haven't seen anything like that but the speed of all the Sophy cars got reset at the same exact time.
And of course after that point they all got much slower.
The Sophy races need good mixing of cars and not these stupid systems like boost.
VID_20241009_230701.gif

(The GIF may not work but I guess you get it)
 
Last edited:
Sophy AI in the first lap is just a rocket. I don't know why after the second lap decrease the speed and the aggressivity.
Discussion about it here.

Other than not understanding why PD don’t offer AI sliders, the blame for “Slow Down SOPHY” could be due to players. Even during the debut World Tour demonstrations players were crying how they couldn’t possibly have a chance to win races versus the agent. We got SOPHY and it was challenging. Players still complained and what happened is what we have now.

Sure, PD’s 27 years of programming computer controlled cars should have seen GT Sport and GT7 with competitive AI from the start. However, Kaz’ plan-dream-vision to implement something new to the GT franchise looks like it’s is going to be continually developed.
We had a good agent at the start and now it’s as if PD pressed a button for SOPHY to just be a normal bot. Looks this way to me on the surface.
 
We had a good agent at the start and now it’s as if PD pressed a button for SOPHY to just be a normal bot. Looks this way to me on the surface.
In fact I reckon the normal AI is much-faster in, for example, a Grp 3 custom race with BOP on and Boost set to Weak.

Sophy is a sleepy win first time and that's over the 3 laps we have now, whereas the aforementioned custom race is a challenge to win in the 10 -> 12min timeframe (5 -> 8 laps depending on track) I usually go for (works well when taking turns with friends).

Chalk and cheese. I'm sooo keen for PD to give us Sophy 3.0 in the (hopefully) big update coming soon... with a decently-granular difficulty option that maxes out at "insane".
 
Last edited:
In fact I reckon the normal AI is much-faster in, for example, a Grp 3 custom race with BOP on and Boost set to Weak.

Sophy is a sleepy win first time and that's over the 3 laps we have now, whereas the aforementioned custom race is a challenge to win in the 10 -> 12min timeframe (5 -> 8 laps depending on track) I usually go for (works well when taking turns with friends).

Chalk and cheese. I'm sooo keen for PD to give us Sophy 3.0 in the (hopefully) big update coming soon... with a decently-granular difficulty option that maxes out at "insane".
Not just Custom Races, but as we all have mentioned many times, the Chilli race programming. I found the Fuji race is the toughest for me.
Anyway, it’s good that SOPHY can drive most cars and now at a few more layouts, but the bad seems like it’ll be around for a while.
 
I wasn’t saying it was impossible for the auto-grinding, that’s kind of irrelevant in my opinion. I was more going after the machine learning AI of a home user setting time trial lap times that a human driver have difficulty beating. That would be a conundrum to solve. How do you know if someone is as good as their lap times claim to be? Just taking the edge cases into consideration is all 😄👍🏻
For your information, this is the case on rfacor 2
In trials, the AI improves its time as it completes each lap.

Note that when the session ends, AI forgets its learning, but you can unlock this option so that it remembers its learning.
Sorry for going off topic butwe could imagine this for the future of the gran turismo license.

Even if Sophy gives me a hard time...
 
Last edited:
In fact I reckon the normal AI is much-faster in, for example, a Grp 3 custom race with BOP on and Boost set to Weak.

Sophy is a sleepy win first time and that's over the 3 laps we have now, whereas the aforementioned custom race is a challenge to win in the 10 -> 12min timeframe (5 -> 8 laps depending on track) I usually go for (works well when taking turns with friends).

Chalk and cheese. I'm sooo keen for PD to give us Sophy 3.0 in the (hopefully) big update coming soon... with a decently-granular difficulty option that maxes out at "insane".


That because Sophy is right now blocked in Arcade Quick Race format. It should be compared to normal AI in the same mode. Why? Because this Quick Race has fixed „rubber banding” settings based on selected difficulty, and yes it has rubber banding on hard too. Sophy is not dumbed down it is just limited to one mode where there is rubber banding always on

I think that we will see real Sophy when PD will add it to custom race and we could define „boost” for it or in career mode in paprika events.
 
I posted a video in the video thread about this exact thing. Extremely frustrating.

I don't understand the point of nerfing an AI in a specific game mode that allows you to confront the difficulty of an AI that the studio has hyped for us.
A Sophy AI known to gran turismo players, since the studio has been teasing us with Sophy for months and months, confronting this AI with exceptional players, so devellopers shouldn't be afraid of players being humiliated by an AI. .

Or maybe is it specific to the vehicle she's driving ?(I can't test it, my ps5 is dead at the moment).
 
Last edited:
Sophy AI in the first lap is just a rocket. I don't know why after the second lap decrease the speed and the aggressivity.
Same. I get passed by AI cars on the first lap and then catch and pass them on the second lap. Boost mode is active in Sophy races. You just couldn't make it up.
 
That because Sophy is right now blocked in Arcade Quick Race format. It should be compared to normal AI in the same mode. Why? Because this Quick Race has fixed „rubber banding” settings based on selected difficulty, and yes it has rubber banding on hard too. Sophy is not dumbed down it is just limited to one mode where there is rubber banding always on

I think that we will see real Sophy when PD will add it to custom race and we could define „boost” for it or in career mode in paprika events.
Cynic in me thinks they're probably gonna hold off on a global Sophy integration across the game for GT8, would be a really good selling point. I wouldn't be surprised if that's also the reason why we haven't gotten any new weather options.
 
That because Sophy is right now blocked in Arcade Quick Race format. It should be compared to normal AI in the same mode. Why? Because this Quick Race has fixed „rubber banding” settings based on selected difficulty, and yes it has rubber banding on hard too. Sophy is not dumbed down it is just limited to one mode where there is rubber banding always on

I think that we will see real Sophy when PD will add it to custom race and we could define „boost” for it or in career mode in paprika events.
The same boost is still there in Custom Races even with boost off. It's just completely baked into the programming.
 
Back