Gran Turismo 7 Custom Race thread

[05XR8]

Yeah, I leave full damage and flags on. It’s another good way to see how much avoidance SOPHY calculates. I only turn flags off when I’m spectating.

 

[Revengel]

One note and one question from my race yesterday

View attachment 1463915
Note:
Never use Damage:severe, it's maybe ok for the player, but AI goes to pit as soon as it has any damage to repair it

I actually rather enjoy heavy damage in my custom Sophy races. Among other things, it guarantees (in my opinion) overly aggressive driving will be punished. Sometimes I get caught in it, sometimes it's Sophy on Sophy violence, but it's always a threat. Combined with the fuel multiplier and race length, this adds a level of strategy and (for me at least) enjoyment.

E.G.: Last night I was driving 21 laps at Deep Forest in my 590 cup using my swapped Pantera (at 590.00 PP - all the other cars are between 589.28 and 590.00 PP) and there were four cars that pulled in after the first lap due to damage. This quickly shuffled the field and altered my strategy . . . a bit. The primary car I had my eye on (Elka is normally the quickest around this track at about 1:39.xx) was one of them and I knew she'd have to pit a second time. I planned on doing a one-stop thinking that would give me an advantage. A cat shocked the hamfat out of me and I wound up missing a brake marker and that was it for my race.

When I ran the same race again, Elka wasn't caught in some opening damage (Letty, Ralph Phillips and Shinosuke were) so I decided to try for a non-stop. I just ran out of fuel on the last lap due to a few too many revs and missed shifts (using a clutch/H-Pattern shifter) and Elka still took over first place with 3 laps to go.

Granted, that's just what works for me.

 

[pritchyTim]

AIs we've got in Custom Race mode are "SOPHY + Rubberband" or "PD's + Cafe + Rubberband".
Boost is just another layer of setup for the rubber banding effect.

Brilliant sketch and writeup! Makes a ton of sense, and is consistent with what we all seem to find. Thank you!!

…if i put a runaway redbull (or maybe 3), and possibly also 3 laggards, I can create a 13-and-me grid in the ‘sandwich’ between the dummy racers. Again, consistent with my favorite races where I am in the 2nd tier of 3 classes.

Changing gears a bit, let’s look closer at grid position. Do we find it to be dynamic, constantly reevaluating an AI’s adjusted pace throughout the race? Or is it based on starting grid order? I’ve not tested this closely, but imagine you could set up a grid with the should-be winners in the final grid positions and compare the other cars’ laptimes as they drop from front to back (as the ‘winners’ move forwards).

Would be nice to confirm it’s weighted towards STARTING grid position, as that adds some control for us as we can build grids that slot faster cars towards the back.

Maybe this ends up being another reason to favor short races. Given 2+hrs, that spread will only become more drastic, and regardless of starting position it’ll be hard to override ‘cafe mode’ logic.

Still, would love to find that sweet spot for each of my main grids, promoting close/variable finishing orders and tricking myself into thinking I’ve overcome the artificiality baked into the programming.

 

[Voodoovaj]

Maybe this ends up being another reason to favor short races. Given 2+hrs, that spread will only become more drastic, and regardless of starting position it’ll be hard to override ‘cafe mode’ logic.

Still, would love to find that sweet spot for each of my main grids, promoting close/variable finishing orders and tricking myself into thinking I’ve overcome the artificiality baked into the programming.

I've done hour long races with great results. If I choose a car that is more of challenge, I finish near the back. If I choose one of the better cars, I finish closer to the front.

If there are some runaway cars (for instance, my 787 is a beast at Le Mans), I shuffle them down the grid and it seems to even out.

In every case, I can point to my own mistakes for where I finish rather than something the AI is doing. If you balance your race properly, it should be hardly noticeable.

 

[StuntmanRich]

I’ve been running a DTM championship today using the 155, 190e (engine swapped) and the E36 M3.

All was going great until the round at Spa where one of the two E36s on my grid kept veering off left at the bottom of Eau Rouge and going into the wall every lap. The other M3 was fine. The only difference between the two cars was that the one that made it through safely had 10kg success ballast.

I created a race with 20 of the M3s that couldn’t make Eau Rouge and some of them make the corner while others crash.

Driving the car myself I’d say it’s fine, no issues with wheels hitting arches etc. just a bit tail happy. I used a tune from Youtube.

Does anyone have any tips on what I can do or know what the issue might be?

Edit.: I guess I can try adding ballast like the other car has and see what happens.

Edit 2: i think i solved it. Adding 10kg ballast made them all go through okay. I took away the ballast and stiffened the rear springs and all is fine so I think the rear was compressing too much in the dip and the rear tyres were hitting the arches. Not sure why driving it myself caused no issues but they occurred for the AI.

 

[jesse100190]

when i ran a 10 lap race a dragon trail gardens reverse using 700pp settings the ai pitted like 4-5 times during the race i only pitted once using random ai cars with most of the field pitting on the 1st lap which made no sense when they could of gotten another lap.

 

[CodeRedR51]

Been doing some V8 Supercars inspired races using V8 RWD Gr4 cars with BOP on and, the racing has been good with Sophy, but tonight I decided to try it with BOP OFF and holy cow is the racing more intense. Had to remove 10hp from the Corvettes because they were too quick in a straight line, but all else is pretty close. Just wish you could do standing start with Sophy.

 

[Theufcveteran]

Been doing some V8 Supercars inspired races using V8 RWD Gr4 cars with BOP on and, the racing has been good with Sophy, but tonight I decided to try it with BOP OFF and holy cow is the racing more intense. Had to remove 10hp from the Corvettes because they were too quick in a straight line, but all else is pretty close. Just wish you could do standing start with Sophy.

View attachment 1464166

Try a 15 m rolling start if you haven't already, it seems like in the last update it's been spaced out more so you're not almost bumper to bumper, but it's close enough that you get a good bit of side by side into the first braking zone.

 

[Voodoovaj]

Race Report: Sophy 3 hour endurance at Spa.

I decided to push the endurance time frame a little further. I'm doing 3 hours of Spa with fuel usage set to 4X, basically time compressing a 12 hour race down to 3 hours. I have the tire usage at 2x.

I have my Group C tuned down to 800pp (I've posted the setups in this thread) with the default RM tires. Also, this race is my 2 class Group C/GT set up, just for full disclosure. AS you can see, I took this screenshot while I was pitting near the halfway point.

Race breakdown. I started last (for maximum payout). Obviously, passing the GT class was pretty quick. I took the lead around lap 7 or 8 IIRC.

I'm stopping around every 30 minutes, and I have no idea if I am on pace of a win, or a 3 hour long disappointment. Turn the power down affects the fuel mileage for all cars. When I stopped, there was a group that stopped 1lap after me, and a second group that stopped several laps after me, putting us all out of sync.

I did this last night, and was starting to nod off a little, and at the beginning of the second hour, I ended up driving off track a bit and I lost 4 positions . I managed to get most of the positions back, and I camped behind Derek Bell in the 962C, using his draft to help me save fuel. I was also watching the tail lights to see when he'd indicate for a pit stop. I figured I might be able to pass him in the pits. That's the photo below.

He pulled in almost on empty, while I had 3 laps of fuel. I left the pits 4 seconds up on him, but 20 seconds behind the other group because of being out of sync.

Where am I right now? I paused with 1 hour and 28 minutes to go in 5th place. Which is interesting because, if I had done a 90 minute race, I'd have lost it. I'm going to need to keep pushing to get myself ahead of both packs, so when I finally take that last stop, I'm not left chasing down the lead. I'm going to have to pull in with about 20 minutes to go to ensure I don't run out of fuel before the end. At this point, I'm confident, but another mistake could throw it all away.

Another note, I have damage on light. Pescarolo's 962 was passing a GT and slammed on the brakes at Eau Rouge, and I slammed into his tail. If damage was heavy, I would have been another stop behind. Stay tuned for the final result.

Edit - Race Update #2. 40 minutes to go.

I'm still not sure where it will all play out. Below is a pic from the last stop just before I paused. I plan on running for about 15 minutes, and then stopping for the last time. I think that if I am fueled up with about 20-22 minutes to go, so I can rise it out to the end, I'll be ok.

Note - I think I have figured out the Sophy boost, and I have to agree with the detractors that it's pretty bad. The fastest lap doesn't show in the pit stop screen shots, but it is from another Jag at 2:02.9. I'm pretty confident that I can match that, or come close, IF the car was at 100% power and/or on soft tires (most likely AND). I'll have to try it post race. That would be a clean and simple way to boost the AI, give it 100% power and softer rubber. It would also help explain why there are restriction on the tires (because, then the AI could potentially never boost).

In any event, I might get time tonight to finish up and I will add to the post after the race.

Edit Race Report #3 - End of race.

I forgot to mention that in the middle part of the race, a third group developed with 2 other Jags. One being the same that set the fast lap. I decided that I needed to be in sync with that one. It was going to be the group that would still be running near the front at the end. The other groups would need to stop.

I grabbed a final stop with 22 minutes to go, and it was the right call. I passed the other Jags with a few minutes left to go, and the leading Merc dove into the pits.

It was fun to be on the edge for the entire race, rather than knowing 3 laps in that I would win by an enormous margin. HOWEVER, I did confirm what boost is doing. See the next post for that.

 

[Voodoovaj]

OK, this is about the boost (boost weak) that I experienced in the three hour race above. As I mentioned, there were some really fast laps being set by the AI. First by a 962 and then, the fastest lap, by another Jag. I spent time thinking about it, and had an idea that the Jag gained full power AND RS tires.

The Jag set a FL of 2:02.960

First I went and set a TT time with the car, as it was set up in the race. 800pp with RM tires. I set a time a little better than my fastest lap in the race. Sophy is good, but that's TOO good.

Just to eliminate the possibility that it's only RS tires, I did that next. Still way for off Sophy's time. I had gone into the replay post race to watch Sophy from the cockpit. I could see it was very fast, but the line wasn't anything special. I knew I could do better by following a faster line and using some curbs a bit more. There's NO WAY Sophy was only using RS tires to set that time.

Last, I created a new setting sheet, turn the power back to 100% with no ballast and I fitted the RS tires.

(just so you don't have to scroll up) Sophy's best 2:02.960

My best 2:02.013 and I had more to go, but this was good enough to prove my theory.

This would explain why the racing is better with the slower cars, and with everything set to RH tires. Now, I don't know that ALL cars grab RS to boost, but it would make sense with the reports that Sophy is setting unrealistic time. They're only unrealistic on street tires.

It's a little disappointing to find this. Yes, it's still fun, and Sophy does move around the track better and act more realistic. However, to know that it is boosting with this on the fly tire change, and a shift to full power where possible, breaks the immersion.

On the bright side, I am VINDICATED when it comes to the the tire debate from a few weeks back. I could SEEEE the AI go faster. Well, this explains it. I was seeing them boost and switch to stickier tires. It wasn't from my tunes. It was the boost.

Now, I wonder what happens when boost is set to none 🤔

 

[pabloprp69]

OK, this is about the boost (boost weak) that I experienced in the three hour race above. As I mentioned, there were some really fast laps being set by the AI. First by a 962 and then, the fastest lap, by another Jag. I spent time thinking about it, and had an idea that the Jag gained full power AND RS tires.

The Jag set a FL of 2:02.960

First I went and set a TT time with the car, as it was set up in the race. 800pp with RM tires. I set a time a little better than my fastest lap in the race. Sophy is good, but that's TOO good.
View attachment 1464432

Just to eliminate the possibility that it's only RS tires, I did that next. Still way for off Sophy's time. I had gone into the replay post race to watch Sophy from the cockpit. I could see it was very fast, but the line wasn't anything special. I knew I could do better by following a faster line and using some curbs a bit more. There's NO WAY Sophy was only using RS tires to set that time.
View attachment 1464433

Last, I created a new setting sheet, turn the power back to 100% with no ballast and I fitted the RS tires.

(just so you don't have to scroll up) Sophy's best 2:02.960

My best 2:02.013 and I had more to go, but this was good enough to prove my theory.View attachment 1464434

This would explain why the racing is better with the slower cars, and with everything set to RH tires. Now, I don't know that ALL cars grab RS to boost, but it would make sense with the reports that Sophy is setting unrealistic time. They're only unrealistic on street tires.

It's a little disappointing to find this. Yes, it's still fun, and Sophy does move around the track better and act more realistic. However, to know that it is boosting with this on the fly tire change, and a shift to full power where possible, breaks the immersion.

On the bright side, I am VINDICATED when it comes to the the tire debate from a few weeks back. I could SEEEE the AI go faster. Well, this explains it. I was seeing them boost and switch to stickier tires. It wasn't from my tunes. It was the boost.

Now, I wonder what happens when boost is set to none 🤔

I'm trying to follow everything you've said. Is it a theory or you actually saw the AI with RS tires in the replay?

If I understand correctly, the whole race was ran with RM, and towards the end, Sophy was setting unrealistic fast laps?

 

[Theufcveteran]

OK, this is about the boost (boost weak) that I experienced in the three hour race above. As I mentioned, there were some really fast laps being set by the AI. First by a 962 and then, the fastest lap, by another Jag. I spent time thinking about it, and had an idea that the Jag gained full power AND RS tires.

The Jag set a FL of 2:02.960

First I went and set a TT time with the car, as it was set up in the race. 800pp with RM tires. I set a time a little better than my fastest lap in the race. Sophy is good, but that's TOO good.
View attachment 1464432

Just to eliminate the possibility that it's only RS tires, I did that next. Still way for off Sophy's time. I had gone into the replay post race to watch Sophy from the cockpit. I could see it was very fast, but the line wasn't anything special. I knew I could do better by following a faster line and using some curbs a bit more. There's NO WAY Sophy was only using RS tires to set that time.
View attachment 1464433

Last, I created a new setting sheet, turn the power back to 100% with no ballast and I fitted the RS tires.

(just so you don't have to scroll up) Sophy's best 2:02.960

My best 2:02.013 and I had more to go, but this was good enough to prove my theory.View attachment 1464434

This would explain why the racing is better with the slower cars, and with everything set to RH tires. Now, I don't know that ALL cars grab RS to boost, but it would make sense with the reports that Sophy is setting unrealistic time. They're only unrealistic on street tires.

It's a little disappointing to find this. Yes, it's still fun, and Sophy does move around the track better and act more realistic. However, to know that it is boosting with this on the fly tire change, and a shift to full power where possible, breaks the immersion.

On the bright side, I am VINDICATED when it comes to the the tire debate from a few weeks back. I could SEEEE the AI go faster. Well, this explains it. I was seeing them boost and switch to stickier tires. It wasn't from my tunes. It was the boost.

Now, I wonder what happens when boost is set to none 🤔

The simplest answer is that the noted rubberbanding kicks in when they’re far behind and they start setting insanely fast times that go beyond what is possible.

 

[suomi1]

OK, this is about the boost (boost weak) that I experienced in the three hour race above. As I mentioned, there were some really fast laps being set by the AI. First by a 962 and then, the fastest lap, by another Jag. I spent time thinking about it, and had an idea that the Jag gained full power AND RS tires.

The Jag set a FL of 2:02.960

First I went and set a TT time with the car, as it was set up in the race. 800pp with RM tires. I set a time a little better than my fastest lap in the race. Sophy is good, but that's TOO good.


Just to eliminate the possibility that it's only RS tires, I did that next. Still way for off Sophy's time. I had gone into the replay post race to watch Sophy from the cockpit. I could see it was very fast, but the line wasn't anything special. I knew I could do better by following a faster line and using some curbs a bit more. There's NO WAY Sophy was only using RS tires to set that time.


Last, I created a new setting sheet, turn the power back to 100% with no ballast and I fitted the RS tires.

(just so you don't have to scroll up) Sophy's best 2:02.960

My best 2:02.013 and I had more to go, but this was good enough to prove my theory.

This would explain why the racing is better with the slower cars, and with everything set to RH tires. Now, I don't know that ALL cars grab RS to boost, but it would make sense with the reports that Sophy is setting unrealistic time. They're only unrealistic on street tires.

It's a little disappointing to find this. Yes, it's still fun, and Sophy does move around the track better and act more realistic. However, to know that it is boosting with this on the fly tire change, and a shift to full power where possible, breaks the immersion.

On the bright side, I am VINDICATED when it comes to the the tire debate from a few weeks back. I could SEEEE the AI go faster. Well, this explains it. I was seeing them boost and switch to stickier tires. It wasn't from my tunes. It was the boost.

Now, I wonder what happens when boost is set to none 🤔

I'm gonna quote you :

So you're definition of "rubber banding" is Sophy maintaining itself near your time? I don't believe many would agree with you. We have all seen what Sophy on 100% can do. No human can touch it. So, if Sophy is hovering around your time, that's just you being slow.

Also, I am still EXTEMELY doubtful of your experience with the boost weak setting. Because, I'm also very experienced with this game, and Sophy pulled out an 11 second lead on me over 15 minutes. So, either you are so terrible that the game is "rubber banding" as you say to be as bad as you, OR you believe yourself to be a driving god and think that an unbeatable AI is cheating to keep up with you.

But hey, if you hate it, great, go play something else and let the rest of us enjoy the best feature ever added to this game.

Not near my times, close to my car physically.

Which means Sophy or PD's AI can break the physics to be closer to you if needed.

AI didn't go back to full power, it ignored the physical limits of the car to stay close to you. If you give AI 100% of its power, they will still try to bypass it to catch up if needed. Try to drive a Red Bull against your Group C grid with their basic settings and look at their lap times.

 

[Buckethead75]

The standard IA is so dumb strategy wise. I have recreated a 1995 JGTC field with a mix of road and racing cars, and to make everybody stop once in a 12 laps race at Suzuka I have set fuel consumption on ×3 and tires on x11. That way race cars must stop for tires and road cars for fuel.

But race cars push their tires way too long and instead of stopping around mid race distance, they wait for their tires to be completly dead. They lose almost 10s per lap at the end of their stint, and most of them finish way behind road cars at the end.

It can’t be that complex to check the percentage of wear, compare it to the percentage of race completion and deduce the right strategy to optimize tire performance... OK it’s a bit more complex if you add fuel strategy in the mix but it’s still not rocket science.

 

[Voodoovaj]

I'm trying to follow everything you've said. Is it a theory or you actually saw the AI with RS tires in the replay?

If I understand correctly, the whole race was ran with RM, and towards the end, Sophy was setting unrealistic fast laps?

Earlier this year, I noted that the AI was driving as if it was on RS tires or SS tires, because I had swapped the tires in the setup. Then I learned that the AI will only use the default tire or RH tires. It appears that they do indeed swap tires during times of boost

The simplest answer is that the noted rubberbanding kicks in when they’re far behind and they start setting insanely fast times that go beyond what is possible.

But it's not insanely fast times. The cars are not going above their capacity. They're going their maximum capacity.

Which means Sophy or PD's AI can break the physics to be closer to you if needed.

AI didn't go back to full power, it ignored the physical limits of the car to stay close to you.

No, it didn't break the laws of physics. It didn't break the physical limits of the car. As I CLEARLY show, the time can be easily replicated if the car is at full power with RS tires.

What you are describing is a more difficult implementation. Simply having the cars step up in tire grade on the fly is a much easier implementation. The same is true of allowing full power. The code won't have to account for anything that "breaks the physics" which would be a testing nightmare. It's easy enough to test a car at max power with soft tires to make sure it doesn't cause a game breaking bug. Having a car go above it's capacity is too risky for development.

If you give AI 100% of its power, they will still try to bypass it to catch up if needed. Try to drive a Red Bull against your Group C grid with their basic settings and look at their lap times.

Swap the tires. Like, I QUITE LITERALLY show you ⬇️⬇️⬇️⬇️⬇️. Also, you forgot about the RSS tires, which are likely still in the game for use by the AI.

(just so you don't have to scroll up) Sophy's best 2:02.960

My best 2:02.013 and I had more to go, but this was good enough to prove my theory.View attachment 1464434

Anyway you slice it, it's disappointing that their solution for an interesting race is merely the opposite of the community's solution. Where we tuned down, they tune up. It does lose it's luster now. It is better in some ways, but obviously, still not as good as a human with the same set up.

 

[Nebuc72]

Earlier this year, I noted that the AI was driving as if it was on RS tires or SS tires, because I had swapped the tires in the setup. Then I learned that the AI will only use the default tire or RH tires. It appears that they do indeed swap tires during times of boost

But it's not insanely fast times. The cars are not going above their capacity. They're going their maximum capacity.

No, it didn't break the laws of physics. It didn't break the physical limits of the car. As I CLEARLY show, the time can be easily replicated if the car is at full power with RS tires.

What you are describing is a more difficult implementation. Simply having the cars step up in tire grade on the fly is a much easier implementation. The same is true of allowing full power. The code won't have to account for anything that "breaks the physics" which would be a testing nightmare. It's easy enough to test a car at max power with soft tires to make sure it doesn't cause a game breaking bug. Having a car go above it's capacity is too risky for development.


Swap the tires. Like, I QUITE LITERALLY show you ⬇️⬇️⬇️⬇️⬇️

I appreciate you have a game design background but this is not how Gran Turismo AI works, or has ever worked.
Their pace is almost always artificially effected by boost tables within each event parameter, including in custom race behaviour.
If you have access to the behind the scenes data of the game you can see the same structure applied through all of the franchise.

It's not 'breaking the physics' moreso than it is just adding additional multipliers onto the cars existing performance. Not literally changing tire grades, but the game would apply say, a 60% grip modifier to the cars behind, a 20% performance increase, etc.
Each AI has a performance rating in individual categories. Acceleration, braking, and so on. There are numerous boost flags that control how each of these ratings are influenced by the 'rubberbanding'.
These can go over 100, which is technically exceeding what the car is capable of, but that doesn't really matter as the AI never work with the same core restraints as the player does in the first place.

This could overall get them to a similar performance level to a RS tyre or full power but that's just coincidence rather than literal design. The design is just "make faster to catch up".

This is sourced from the data tables for GT7. Obviously I am not going to post it public lest I be killed but that is, for better or for worse, how the game works.

 

[suomi1]

I appreciate you have a game design background but this is not how Gran Turismo AI works, or has ever worked.
Their pace is almost always artificially effected by boost tables within each event parameter, including in custom race behaviour.
If you have access to the behind the scenes data of the game you can see the same structure applied through all of the franchise.

It's not 'breaking the physics' moreso than it is just adding additional multipliers onto the cars existing performance. Not literally changing tire grades, but the game would apply say, a 60% grip modifier to the cars behind, a 20% performance increase, etc.
Each AI has a performance rating in individual categories. Acceleration, braking, and so on. There are numerous boost flags that control how each of these ratings are influenced by the 'rubberbanding'.
These can go over 100, which is technically exceeding what the car is capable of, but that doesn't really matter as the AI never work with the same core restraints as the player does in the first place.

This could overall get them to a similar performance level to a RS tyre or full power but that's just coincidence rather than literal design. The design is just "make faster to catch up".

This is sourced from the data tables for GT7. Obviously I am not going to post it public lest I be killed but that is, for better or for worse, how the game works.

Well, thank you for this clear explanation.

No, it didn't break the laws of physics. It didn't break the physical limits of the car. As I CLEARLY show, the time can be easily replicated if the car is at full power with RS tires.

What you are describing is a more difficult implementation. Simply having the cars step up in tire grade on the fly is a much easier implementation. The same is true of allowing full power. The code won't have to account for anything that "breaks the physics" which would be a testing nightmare. It's easy enough to test a car at max power with soft tires to make sure it doesn't cause a game breaking bug. Having a car go above it's capacity is too risky for development.

That's the easiest thing on Earth to implement, just some flags and percentages as @Nebuc72 said. This thing exists into video games since at least 1992 and Super Mario Kart.

Swap the tires. Like, I QUITE LITERALLY show you ⬇️⬇️⬇️⬇️⬇️. Also, you forgot about the RSS tires, which are likely still in the game for use by the AI.

No you didn't. You compared your lap time with a 100% power Jag, no ballast and RS tires to AI's with 800PP and RM tires. Assuming this :

I spent time thinking about it, and had an idea that the Jag gained full power AND RS tires.

No, they just had the percentage given by rubberbanding.

Again, go try to fight an AI Group C with a Red Bull, giving them 100% power / no ballast / RM tires (not RS because AI can't wear them, obviously) / a better setup if you want. Crush them with your alien car, then look at their lap times in the replay.
Reality is sometimes stubborn, you have to try it by yourself.

 

[Voodoovaj]

I appreciate you have a game design background but this is not how Gran Turismo AI works, or has ever worked.
Their pace is almost always artificially effected by boost tables within each event parameter, including in custom race behaviour.
If you have access to the behind the scenes data of the game you can see the same structure applied through all of the franchise.

I'd love to see these, but it's neither here nor there, what you've described is how game physics works. It's just a bunch of parameters.

Grip for instance would be a base value and you can add or subtract from it depending on the "tire" and the material is on. So, RS tires would be Grip+++ on asphalt. Grip --- in heavy rain.

It's not 'breaking the physics' moreso than it is just adding additional multipliers onto the cars existing performance. Not literally changing tire grades, but the game would apply say, a 60% grip modifier to the cars behind,

That's literally what video game tires are. You can add complexity to the model and it's behaviour, but at the end of the day, it's a grip modifier.

Each AI has a performance rating in individual categories. Acceleration, braking, and so on. There are numerous boost flags that control how each of these ratings are influenced by the 'rubberbanding'.
These can go over 100, which is technically exceeding what the car is capable of, but that doesn't really matter as the AI never work with the same core restraints as the player does in the first place.

Again, I have never seen any cars exceed their maximum potential. "Over 100" is a tough thing to discuss because what is 100? Is 100 the maximum a car can do based on it's settings file? Is 100 what car can do with all the power and grip multipliers available to it?

I've seen people in the same car as me go 2-3 seconds faster than I can even HOPE to muster. We see really fast times every weekly time trials and no one is cheating. So, with that in mind, and again, that I have never ever seen a time that I cannot replicate (either in game or posted on this forum), I'm having a hard time accepting that they are bestowing something upon the AI that it outside the boundaries of possibility.

It COULD be, but I have yet to see it.

This could overall get them to a similar performance level to a RS tyre or full power but that's just coincidence rather than literal design. The design is just "make faster to catch up".

This is sourced from the data tables for GT7. Obviously I am not going to post it public lest I be killed but that is, for better or for worse, how the game works.

I'd be surprised that they were sloppy like that (modifiers on top of modifiers). I'm not saying that you're wrong, I'm just surprised that this method would be deemed acceptable in this day and age. I have seen some CRAZY stuff (Like NES legacy code in PS3 games). If you told me there was legacy code from GT1, I'd believe you.

(side story, I was at EA Redwood Shores back in the day and was told a rumor that during the Madden 96 issues, they tried to get a European outsourcer to help with the core game. They delivered futbol instead of football and THAT was the nail in the coffin for the game and the relationship between EA and Visual Concepts. True story? Crazy rumor? I don't know. It was an exec who should know, but he could have just been messing with me )

At the end of the day, my point still holds, I have yet to see anything beyond the possibility of each individual car's capacity. Sophy did a time faster than RM tires and 800pp allowed, but it didn't go faster than the car's maximum. I went faster than Sophy in my test, and I didn't invoke any special procedure to do it. I just turned everything to the max.

Now, if someone can show me a Honda FIT setting F3500 lap times, or something along those lines, then I'll believe it. But, to me, it seems the cars are constrained to their theoretical maximums.


AT THE END OF THE DAY, I think what's disappointing is that this is a bit of a broken promise. Sophy in short races looks like what it was billed to be. It was promised to be better than the best human, but of course it would be if it cheats.

It's simply disappointing that this is the method they have chosen. I think we can all agree on that. It breaks the immersion. I don't do the longer single player races because there is no feeling that the AI has a desire or capacity to win. Sophy was showing a lot of promise, but once you push it, it falls apart and relies on an inelegant method to remain competitive.

It would be better if they simply gave it special AI tires and slight PP boost from the start, and let the chips fall where they may. Oh well, I guess I don't need to test longer races.

 

[XSquareStickIt]

I've never seen under the bonnet to know how GT7 code works, but I do believe that, in rubber banding to catch up to the player, Sophy AI does bend physics to perform beyond what should theoretically be possible for the car.

To show this, I'm going to quote myself from earlier in the thread.

​

In fact, SophyKin has always upshifted from 4th gear to 5th gear at 202km/h. After I pass her, she gets up to around 216km/h before upshifting at the exact same rpm. It's baffling!

Also, in the braking zones, SophyKin closes massively on my lighter 1992 NSX, and she does this using roughly 60% of the brake pedal. Rather ridiculous.

If anyone would like to study the replay in–game, I've shared the replay publicly:

NSX Trophy (Sophy Custom Race)​

#gtplanet #nsxtrophy #rubberband​

They way I look at it is that, when rubber banding, Sophy seems to drive on a conveyor belt going forward, and the car's speedo reads the air speed of the car rather than the wheel speed, hence why there was such a disparity in shift km/h despite ostensibly shifting at the exact same rpm (I assume Sophy drives with the game's AT shifting). Under braking, that conveyor belt Sophy drives on reverses direction to let it stop better. I think that's also why they're so keen on dive bombing from far behind the player.

So yeah, I'm of the firm belief that Sophy will bend physics and cheat to perform what should otherwise be physically impossible when rubber banding towards the player.

 

[MGR]

Further to this post where I tested the different boost modes in Sophy custom races, I ran another set to illustrate how the AI boost/handicap is being applied.

The previous tests would suggest that with Weak Boost (and professional level difficulty) Sophy AI is running at 100% power and grip provided the player's car is somewhere behind the leader. Once the player leads the way, only then does Sophy being 'boosting'. The tests also showed that 'Boost Weak' used the most aggressive form of AI boosting, while 'Boost Off' demonstrated the highest AI handicap when the player was lagging behind. So I used those three scenarios in the tests below.

We already know that AI power boosting was playing a role, but I wanted to test the minimum mid corner speeds around turn 6 and 7 of the Red Bull Ring to work out the influence of any grip multiplier. I selected those corners because of their long, constant radius which should help with data consistency.

• The middle row in each table (Boost Weak + AI leading) can be used as a reference point to show the performance of the current Sophy AI agent when her vehicle is unmanipulated (ie 100% power and grip)
• The top rows demonstrate the maximum performance penalty applied to Sophy of any of the boost modes (Book Off + AI leading).
• The bottom rows show the maximum performance boost of any of the boost mode scenarios (Boost Weak + AI trailing behind)

Refer to this post for the previous tests.

 

[Nebuc72]

I'd love to see these, but it's neither here nor there, what you've described is how game physics works. It's just a bunch of parameters.

Grip for instance would be a base value and you can add or subtract from it depending on the "tire" and the material is on. So, RS tires would be Grip+++ on asphalt. Grip --- in heavy rain.

That's literally what video game tires are. You can add complexity to the model and it's behaviour, but at the end of the day, it's a grip modifier.

Yeah, I know. But the AI pace is dictated by applying modifiers on top of whatever 'state' the car has entered the track with.
It's not that literally the AI will 'step up' from RH to RS tires on the fly, but they will just get a flat modifier that might resemble that performance level.

I'd be surprised that they were sloppy like that (modifiers on top of modifiers). I'm not saying that you're wrong, I'm just surprised that this method would be deemed acceptable in this day and age.

Tell me about it man!
I mean, there's a reason AI has been such a hot topic in GT for so long, isn't there? Because it's all antiquated design, poorly done, stuck about 20 years in the past. All events have custom boost tables attached to them that dictate the pace of the AI witihn the event. There's a global boost setting per event and then each car also has its own boost parameters on top of it for various facets of their handling.

It very much is modifiers upon modifiers, probably a few more layers of modifiers on top of that.

It's the whole reason Sophy has these problems, they cannot seem to let go of this boost table system.



Nenkai (retired GT modder) did a short Twitter thread on some of the issues with it back near launch.


It's not insight into Sophy specifically but just shows how many layers of boost there are that go into the events. It's incredibly convoluted and as you say I have no idea why they still think it's an acceptable application.

 

[suomi1]

I'm having a hard time accepting that they are bestowing something upon the AI that it outside the boundaries of possibility.

It COULD be, but I have yet to see it.

At the end of the day, my point still holds, I have yet to see anything beyond the possibility of each individual car's capacity.

We can't help you seing if you refuse to see what can be seen doing a few simple tests.

I went faster than Sophy in my test, and I didn't invoke any special procedure to do it. I just turned everything to the max.

Not any special procedure other than tune everything to the max to beat an AI that was racing an heavier and less powered car. Your answer is here, just accept it.

Now, if someone can show me a Honda FIT setting F3500 lap times, or something along those lines, then I'll believe it. But, to me, it seems the cars are constrained to their theoretical maximums.

Modifiers don't apply at levels as extreme as these, as you can well imagine. Decredibilizing an analysis by caricaturing things in this way cannot lead to a productive discussion.

I mean, there's a reason AI has been such a hot topic in GT for so long, isn't there? Because it's all antiquated design, poorly done, stuck about 20 years in the past.

Tell me we still have traces of "Motor Toon Gran Prix" code lines into GT6 and I wouldn't be that surprised. lol

 

[Voodoovaj]

Not any special procedure other than tune everything to the max to beat an AI that was racing an heavier and less powered car. Your answer is here, just accept it.

You're making the assumption that the AI was using the same set up that I was. That's a fundamental misunderstand of the virtual world.

The AI is using the same car file, that's all. It can do whatever it likes to that file within the game system.

Modifiers don't apply at levels as extreme as these, as you can well imagine.

Modifiers can apply to everything. There are no "extreme cases". There's merely your (specific) inability to replicate a Sophy lap, which I have shown is not extreme in the least.

Decredibilizing an analysis by caricaturing things in this way cannot lead to a productive discussion.

Correct, so you should stop doing it.

Tell me about it man!
I mean, there's a reason AI has been such a hot topic in GT for so long, isn't there? Because it's all antiquated design, poorly done, stuck about 20 years in the past. All events have custom boost tables attached to them that dictate the pace of the AI witihn the event. There's a global boost setting per event and then each car also has its own boost parameters on top of it for various facets of their handling.

It very much is modifiers upon modifiers, probably a few more layers of modifiers on top of that.

It's the whole reason Sophy has these problems, they cannot seem to let go of this boost table system.

I was chatting with a buddy about it last night. "Why didn't they...yada, yada, yada"

I want to be clear, these Sophy races are still super fun when they are short. Still, why, when you have a blank slate, would you opt to keep doing the same old thing? I gave them the benefit of the doubt that Sophy takes more power, and that's why it's not on PS4, but is that legit? I don't think so now.

I think the biggest fault is that they chose to make everything revolve around the player, when (what I think we'd all like to see) is for the player to blend into the race. Have the race unfold naturally based on car to car interactions, rather than trying to force an orderly progression of cars to pass.

 

[suomi1]

You're making the assumption that the AI was using the same set up that I was. That's a fundamental misunderstand of the virtual world.

The AI is using the same car file, that's all. It can do whatever it likes to that file within the game system.

Modifiers can apply to everything. There are no "extreme cases". There's merely your (specific) inability to replicate a Sophy lap, which I have shown is not extreme in the least.

Correct, so you should stop doing it.

Dear Lord! Let's stop this, indeed.

Whaaaaatever...... I just had my best race so far on GT7 applying what we know about Sophy and rubberbanding. A six hour one at that. I finished the race standing in front of the TV. I will eventually make a race repport tomorrow if I have some time to do so.

I thank you again @bliprunner , sometimes the answer does not come from ourselves.

 

[Aedum]

Dear Lord! Let's stop this, indeed.

Whaaaaatever...... I just had my best race so far on GT7 applying what we know about Sophy and rubberbanding. A six hour one at that. I finished the race standing in front of the TV. I will eventually make a race repport tomorrow if I have some time to do so.

I thank you again @bliprunner , sometimes the answer does not come from ourselves.

Glad to hear this.

 

[05XR8]

Battle packs, Switchbacks, divebombs, missing apexes, love taps, bump-n-runs, blocking, side-o-by-side-o, paint scraping. If this ain’t simulating real racing, I’m being duped by the real racing I watch that does everything in my virtual races.




If boosting and rubberbanding and impossible physics continue to produce the racing I’m enjoying, I’m all for it!

 

[CodeRedR51]

Anyone have any tips for Sophy races with the Super Formula cars? To start with, I am terrible with the high downforce cars so I'm still learning how to properly use them. But dropping the ECU power down to minimum and adding some ballast (driver weight basically, varies by car) has helped a lot. How much should I stagger the AI cars with power and weight?

Ran one race with this setup so far and there's not much passing going on.

 

[Yard_Sale]

Per the recommendations in this thread, I put a rabbit at the front of my grid for the pack to chase (in this case it was a Super Silouhette detuned to 700pp, with all other cars being tuned anywhere from 660pp to 675pp)

The pack was NOTICEABLY FASTER with said rabbit leading the pack. So much so that I restarted and changed my starting position from 20th to 15th, and I still got dropped!! I increased my PP from 655 to 660, and it’s still a challenge to stay with everyone. For good measure, I built a V8 Supercar-inspired Mustang and Camaro, slapped together a quick tune at about 665pp, and they were the fastest cars on the track besides the Rabbit in front. Gonna do a little more custom BOP tuning on an individual car basis, then I should be good to go.

Information only; was running 18 laps at interlagos with x2 Fuel and x2 tires…making it a no-stop on purpose, to see how the BOP looked like in a longer sprint race, and to see how each car did on fuel and tires. Happy to report that there were enough differences in the cars fuel/tire consumption to accurately mimic what would be IRL differences


PS:
My goal is to give the field enough of an advantage to where I really have to be on my game just to hold position, with the ULTIMATE GOAL of me having to utilize strategy to place in the top 1/2 of the field (extensive previous testing indicated that Sophy doesn’t really stand a chance against the undercut and/or not taking tires at every stop).



PSS:

ULTIMATE-ULTIMATE goal would be to configure a BOP that gave them enough of an advantage for me to be able to use different compounds as part of my strategy

 

[pabloprp69]

I don't see anyone discussing the different types of slipstream in the game:

I've been messing around with slipstream settings (after only playing with "real" for a while..)

Tried a gr3 race with slip set to strong -to try to simulate the custom slip from sport mode- and the race is much closer together. The leader does not pull away like when it's set to "real". I have to admit I had a blast in those experimental races.

The 8 first cars were wildly trading positions back and forth, and in many turns, I saw 3 and even 4 wide!

I had the impression that in general, the race felt easier and I could manage overtakes without much effort. Will have to play a bit more and report back.

As always, boost weak, professional difficulty, real traction loss.

 

[pritchyTim]

The leader does not pull away like when it's set to "real". I have to admit I had a blast in those experimental races.

The 8 first cars were wildly trading positions back and forth

goal is to give the field enough of an advantage to where I really have to be on my game just to hold position

if she falls behind, Sophy will unleash the most aggressive boost increase of all the boost modes (up to ~8%)

Anyone figure out the sweet spot, where a second-sitting Sophy actually pressures/passes you for the lead?

Of course, MGR’s “up to 8%” is a bit of a sliding scale…if me/my car are only 5% better than baseline-100%-Sophy, I can’t expect it to all of a sudden be 3% faster than I am when I take the lead. It’s a little too easy to create that “they stick with me but don’t really pressure” situation, though of course Sophy is great at punishing even slight mistakes.

Still, maybe there’s an answer in combining some of these recently-suggested configurations. Thanks for all the testing and input, all!

If my ‘dummy rabbit’ (runaway redbull, etc.) is 5% better than ME, and I am 5% better than the rest of the field, is THAT a sweet spot? Would a stronger slipstream setting provide the last bit of missing AI-oomph, to force me to defend second, yet still have what it takes to get there in the first place?

Hopefully, something worth playing with.

I’ve been a lousy contributor, recently, hope to follow up on some of these suggestions soon without simply pitching you all more work to do to (in)validate my conjecture!

Also, as long as I’m soapboxing, sorry for throwing some poo in here while catching up on recent chitchat. Grateful for all discourse and friendship among our little Custom Race community. I just prefer to stay on the side of “nobody really knows/proves stuff in here…PD gonna PD.”