Analysing the AI behaviour and performance in GT6

  • Thread starter Samus
  • 129 comments
  • 12,278 views
KamilCader
Read my posts, this is exacly what I mentioned, the AI itself is already decent but is being slowed down by the handicap system with is turned down in the S class events.
Click to expand...
No the AI is not decent, not now not ever not in S mode not in wonderland they always frigging suck.

I suppose I'll have to finally get around to making a video or two of just how slow they really go through some corners.
Samus
Not being funny but are you new to the GT community? People have asked for, amongst numerous others, better AI for years and PD just haven't done anything about it. Asking is all we can do but it's unlikely PD will listen.
Click to expand...
@KamilCader - Yeah this is about as far from a new request as it gets, and it's just these few simple things, really.
I've been discussing it myself since GT4 life, GT5P, GT5, GT6 have all come and none have any better options for racing the AI, it's the exact same career mode being hashed out over and over and over and over and over and over and over.

I had over 100 friends from GT5, more like 200. (I often had to delete and re-add people to keep the number at 100, and also had a second account for different friends/racing).
About 5-10% are still playing any GT game at all.

I kind of wonder if PD/GT isn't cooked at this point, TBH. They held themselves up as unreachable gods (because they were at a point) long enough that they didn't even notice the party crashed and everybody went home. Everybody.
 
All we need is the race editor. Even if the difficulty can't be set high enough you could always take on ballast or detune your car. Just no more rubberbanding and I'll be happy.
This would also address the issue of weather and tirewear only available in S events.
While racing online is fun it could be difficult to find people that want to run the same race as me. I like 30-50 minutes races with weather, time progress, tire wear, damage ...
 
If I have some time I'll analyse some S-Class races, see if there is anything really different going on.

Oh and as for the "Slow when ahead, fast when behind AI rubber band mechanism" if you ever needed proof of this run the S-Class special event which is a Ford GT showdown on the ring. First run it the best you can, pass the AI as soon as possible and try to get the gold. Note how far behind the AI car is at the end and what time it does. Then do it again but this time sit behind the AI the whole way, never overtake it. It's been a while since I tried that but I do recall the first instance it was fast enough to get gold itself, the second instance it was at least a minute slower, maybe closer to to two.
 
Just try the infinite single race mode for around 15 laps with an super gt500 or 300 or whatever you want. The game will create an even field of silimar cars. At first there will be a handicap to make you catch the first two cars but then it will turn of. At the limited number of laps AI would let you pass at the end of the race but there is no end so the handicap wont occur. If PD would make an option to disable handicap algorythm completly, every car would race with the max pace from the begining. Just try it and tell me if that would be challenging enough for you.
 
I have tried it before and they're still not fast enough IMO. There is more potential in the cars the AI don't come close to using. The biggest thing really is their breaking zones, way too early which generally results in you running into the back of them several times a race.

But in the interest of science, i'll gather some data for those sorts of races as well at some point.
 
Johnnypenso
It appears as if the AI saw that the player wasn't going to catch him because the AI ran a 2:01 second lap and player 2:04, so the AI slowed up 10 seconds per lap off his potential place to let the player catch up.

Only two data points, but it appears as if the AI calculates after a couple of laps or a % of race distance, whether you'll catch him or not and adjusts his pace accordingly or not, as the case may be. In the race I watched, the AI was brutal in lap 3, after decent laps 1-2. All of a sudden in lap 3 he's slowing dramatically for corners, double braking, driving on the grass at the apex etc. He obviously deliberately slowed to let the player catch up.
Click to expand...

That theory got my interest and I tried to test it in an S-class race. I think my test & data don't confirm the theory though.

I chose 10 laps Ascari and entered the race with a McLaren MP4-12C at 585PP on sports soft, which should be an appropriate car judging from the description. Now I ran 2 races vs. the same AI grid without leaving the race screen. I also wanted to check if the AI looks only for your laptimes or for the cars power and tyres too, so I tried the following:

In the first race I entered the car at 585PP SS, but I stayed behind the last AI car for 5 laps and started pushing only after the pitstop at the end of lap 5. Laptimes were about 2:28-2:30 behind the AI, so I lost ground vs. the AI lead for the first 5 laps until it was 46 sec back before my pitstop. At the beginning of lap 8 I took the lead and ran away. The final lap was very poor because tyres were done already.

In the second race I detuned the car to 550PP and used comfort soft, but attacked from the beginning. My laptimes were mostly defined by how well I could cleanly pass the dirty bots, resulting in times between 2:17 and 2:21 during the first 5 laps. So 10 sec a lap quicker this time. At the end of lap 5 I pitted directly behind the then leading FXX and took over because of less refueling.

The top 3 AI cars were the Ferrari Enzo (sport medium), Ferrari FXX (sport hard) and McLaren MP4-12C (sport medium).

Finishing order and total racetime in 1. race:

1. me 585 SS 24:06 min
2. Enzo 24:30 min
3. MP4-12C 24:32 min
4. FXX ?

Finishing order and total racetime in 2. race:

1. me 550 CS 23:50 min
2. MP4-12C 24:15 min
3. FXX 24:26 min
4. Enzo 24:47 min

The laptimes of the AI were quite similar in both races too. Differences of only 2-3 sec, mostly explained by how much they got held up by traffic. The Bugatti started on pole and did less errors in the 2. race. The big dropoff for the Enzo in the second race happened in laps 8-9, when I was long gone in both races.

So at least in this S-class race I cannot find the pattern @Johnnypenso described. There can be huge differences in AI performance there though and I don't know yet what determines them?
 
Last edited:
I've don't a few races at Ascari - and posted loads of AI lap times for the races.

https://www.gtplanet.net/forum/posts/9584003/ (And above)

It seems much of the speed is programed in. The grids are a rolling one but the cars spread out quickly - so the leading cars pull some nice times. - I would guess this is to space the cars out - so that the processor load is lighter. After 3 or so laps they just drop the ball. Maybe the old B Spec "Hot/Cold" mechanic is in play and without any stimulus they just forget how to go fast.

Also as the AI pitstop, sometimes they get held in the pits, but mostly a large engine powerful car will stop for a long time to fuel up. They don't goof up as much as in the Willow Springs event where the AI can run out of fuel and need a final lap pit stop.

Good luck finding stuff... but lets hope the next version cranks up the AI driver abilities.
 
Z1-AV69
That theory got my interest and I tried to test it in an S-class race. I think my test & data don't confirm the theory though.

I chose 10 laps Ascari and entered the race with a McLaren MP4-12C at 585PP on sports soft, which should be an appropriate car judging from the description. Now I ran 2 races vs. the same AI grid without leaving the race screen. I also wanted to check if the AI looks only for your laptimes or for the cars power and tyres too, so I tried the following:

In the first race I entered the car at 585PP SS, but I stayed behind the last AI car for 5 laps and started pushing only after the pitstop at the end of lap 5. Laptimes were about 2:28-2:30 behind the AI, so I lost ground vs. the AI lead for the first 5 laps until it was 46 sec back before my pitstop. At the beginning of lap 8 I took the lead and ran away.

In the second race I detuned the car to 550PP and used comfort soft, but attacked from the beginning. My laptimes were mostly defined by how well I could cleanly pass the dirty bots, resulting in times between 2:17 and 2:21 during the first 5 laps. So 10 sec a lap quicker this time. At the end of lap 5 I pitted directly behind the then leading FXX and took over because of less refueling.

The top 3 AI cars were the Ferrari Enzo (sport medium), Ferrari FXX (sport hard) and McLaren MP4-12C (sport medium).

Finishing order and total racetime in 1. race:

1. me 585 SS 24:06 min
2. Enzo 24:30 min
3. MP4-12C 24:32 min
4. FXX ?

Finishing order and total racetime in 2. race:

1. me 550 CS 23:50 min
2. MP4-12C 24:15 min
3. FXX 24:26 min
4. Enzo 24:47 min

The laptimes of the AI were quite similar in both races too. Differences of only 2-3 sec, mostly explained by how much they got held up by traffic. The Bugatti started on pole and did less errors in the 2. race. The big dropoff for the Enzo in the second race happened in laps 8-9, when I was long gone in both races.

So at least in this S-class race I cannot find the pattern @Johnnypenso described. There can be huge differences in AI performance there though and I don't know yet what determines them?
Click to expand...
Maybe I'm misreading something but this is what I see.

Your time total in the second race is only 16 seconds faster or 1.6 seconds per lap.

You won the first race by 24 seconds and the second one by 25 seconds.

You were 16 seconds faster in the second race, the AI was 15 seconds faster as well.

The pattern is almost perfect. What did I miss?
 
I'm interested to hear how he ran 10 seconds a lap faster
resulting in times between 2:17 and 2:21 during the first 5 laps. So 10 sec a lap quicker this time.
Click to expand...
And only finished 16 seconds ahead of himself...


I'd also like to point out, as I mentioned before, how terribly slow this is showing the AI.
A 550PP on CS against unrestricted 580-640PP supercars on sports tires, and he's dominating them and playing games with them.
 
Johnnypenso
Maybe I'm misreading something but this is what I see.
Your time total in the second race is only 16 seconds faster or 1.6 seconds per lap.
You won the first race by 24 seconds and the second one by 25 seconds.
You were 16 seconds faster in the second race, the AI was 15 seconds faster as well.
The pattern is almost perfect. What did I miss?
Click to expand...

The total time look very similar yes, but in between it were 2 very different races. In the first race I was 46 sec behind the second at the end of lap 5. There were also no big differences in AI laptimes when I was close (and it was almost like racing on the CS). So the AI could only have adapted to my delta over the whole race, for which they must be able to see the future? ;)
 
CSLACR
I'm interested to hear how he ran 10 seconds a lap faster

And only finished 16 seconds ahead of himself...


I'd also like to point out, as I mentioned before, how terribly slow this is showing the AI.
A 550PP on CS against unrestricted 580-640PP supercars on sports tires, and he's dominating them and playing games with them.
Click to expand...

As said I messed up the last lap on the ss, I could have been much closer. The point was I was running almost 25sec a lap below my potential on the ss for the first 5 laps only to make that up later.
 
Z1-AV69
The total time look very similar yes, but in between it were 2 very different races. In the first race I was 46 sec behind the second at the end of lap 5. There were also no big differences in AI laptimes when I was close (and it was almost like racing on the CS). So the AI could only have adapted to my delta over the whole race, for which they must be able to see the future? ;)
Click to expand...
We need to see each of your lap times in each race, along with the lead AI laptimes in each race to figure out if any rubber banding took place. It's also possible rubber banding isn't as strong or is maybe non-existent in this particular race.
 
Z1-AV69
As said I messed up the last lap on the ss, I could have been much closer. The point was I was running almost 25sec a lap below my potential on the ss for the first 5 laps only to make that up later.
Click to expand...
It's my mistake, I forgot the first race was on SS, and therefore you easily gained most of the time back.

Regarding the AI, I never timed an "S" race, however, in my eyes those races are automatically handicapped by the AI over-pitting tremendously.
I mean seriously, you can win Ascari in a damn S2000 just because you stay out and don't pit ffs.
 
Johnnypenso
We need to see each of your lap times in each race, along with the lead AI laptimes in each race to figure out if any rubber banding took place. It's also possible rubber banding isn't as strong or is maybe non-existent in this particular race.
Click to expand...

Maybe I find the time tomorrow, now I have to got to bed. But it wont help without the whole replay. There is a lot going on with crashing AI and stuff.
 
CSLACR
It's my mistake, I forgot the first race was on SS, and therefore you easily gained most of the time back.

Regarding the AI, I never timed an "S" race, however, in my eyes those races are automatically handicapped by the AI over-pitting tremendously.
I mean seriously, you can win Ascari in a damn S2000 just because you stay out and don't pit ffs.
Click to expand...

Very true. The S-class races are fun when not knowing them, but you get very strange results when you try to build hard races. I did the Brands Hatch race with the same 550 PP McLaren on SH. For 11 laps I'm dead last, 10 sec before being lapped when the AI pits. At lap 15 I win by 20 sec having passed 3 cars in the rain I could never catch on dry. Just because they pit again after 2 laps back on slicks. It's time trial.
 
Last edited:
Well, the events in GT6 are more like challenges than actual races. It's more of a "pass-all-of-these-cars-within-three-laps"-thing than a "let's see who will win this race"-thing.

It seems like that's what the events and AI are designed for.
 
It should be already known by the sound that the AI uses some sort of "rubber band" handicap system. The way of how it works is determined by the length of a certain race you choose. I want to point out, and make you to look more closely at the maximum possible racing pace of the leading AI at any track, and think how it would be if "rubber band" wuldnt take place.

I've decided to run a test myself with the single races infinite laps.
The track is baturst, car SLS GT3 from the rentcar menu, no aids, no weather, opponets are silimar GT3 cars, racing hard tyres for everyone. At the start it didnt took me long to clean pass the entire field (approximately 1.5 lap), and after that I've started to run away to ~5.8 sec at the lap 6, but after that the 2nd AI started catching me very rapidly at the lap 7 and overtook me at the lap 8( I didnt tried to block him ) then he gained advantage to around 3 sec till the lap 10 where I quit the race. Since i dare to say I'm not fast but quite consistent( my laps were around 2min.10sec ), I ask about how much fast the AI can really be without hanicap rubber?
 
Last edited:
eran0004
Well, the events in GT6 are more like challenges than actual races. It's more of a "pass-all-of-these-cars-within-three-laps"-thing than a "let's see who will win this race"-thing.

It seems like that's what the events and AI are designed for.
Click to expand...

It seems incredibly odd to me that Polyphony spent so much time perfecting their physics engines, collaborating with suspension/tire companies, accurately modeling tracks/cars, "transforming" gamer's into racers with GT Academy, taking part in the 'Ring 24h etc. only to funnel players through a "career" mode that has little relevance to any of it. It was kind of cute in GT1 through 4, but that was only because it suited the arcade physics of those games. Now, it has become another one of those unfortunate legacy items like the sounds, standards etc. that make Gran Turismo seem somewhat anachronistic.
 
machschnel
It seems incredibly odd to me that Polyphony spent so much time perfecting their physics engines, collaborating with suspension/tire companies, accurately modeling tracks/cars, "transforming" gamer's into racers with GT Academy, taking part in the 'Ring 24h etc. only to funnel players through a "career" mode that has little relevance to any of it. It was kind of cute in GT1 through 4, but that was only because it suited the arcade physics of those games. Now, it has become another one of those unfortunate legacy items like the sounds, standards etc. that make Gran Turismo seem somewhat anachronistic.
Click to expand...

In one way it's like the entire career mode is like a big "coffee break challenge". And nothing wrong about that, that kind of challenges can be fun as well. But it's definitely missing some realistic races.

I guess somehow they came to the conclusion that the average gamer wouldn't want races longer than 5-10 minutes and then they designed the events based on that:

Step One: Stick to the 5-10 minutes mark - 3-5 laps per race
Step Two: Make sure that the player has a good chance of winning even though the races are so short - make the AI slower than the player
Step Three: Make sure that the player doesn't overtake everyone in 1st corner - spread out the field about a kilometer ahead of the player
 
I have made a second test run on the nurburgring gp track with nissan GT3 from rent car menu.
Same conditions: no aids, no weather, racing hard tyres, numer of laps set to infinite. After 15 laps I ended up 3rd with +16.680 to the lidering Audi R8 with had the best lap record of 1:59,227; my lap record was 2:02,017. The last car with was Nissan GTR GT3 had a gap to the lider of +50.990 and its lap recod was 2:04,174. I ask again: How fast the AI in GT6 can really be without hanicap "rubber band"?
 
IMO there is only two runners and the others are blockers
if you do a monotype race like redbull standard, it's clearly visible that:
- extra speed
what ever you do,
you exit a corner just behind the ai (runners) and going on a straight line with the drag on your side,
ai (runners) outperform you

- they have extra aids not accessible for the player
i often see cornering at the limit for me and the ai
i'm inside, ai (blockers) outside and is able to move in an unnatural way on your driving line to make you a brake check
or even worse, ai just don't accelerate at all

it's just done like that to make you a "challenge"

Ai is not dumb or slow to react, it's just programmed to act like that

"off-topic"
what is really slow and badly programmed are the hands animations for shifting gears
take the kart shifter, you are already in the 4th gear and animation is only going to shift in 2nd gear :D
take the enzo and drives a bit with too low or too much revs, the hands are already in position to shift the gear ...
 
Samus
The biggest thing really is their breaking zones, way too early which generally results in you running into the back of them several times a race.
Click to expand...
Agreed. As someone has said, it appears to be a main cause for their driving into the player too (after dodgy hit detection of course). I didn't agree initially but after some more time I can see what was meant. They can't react quick enough to you suddenly appearing up the inside after braking so they just assume their normal line. If that was fixed, along with the rubber band being snapped, I'd be happy enough. Still want "racier" AI though. And no, I don't mean with sexy suspenders and 'Carry On' jokes.

By the way, it's "braking," although I'm pretty sure that a) the AI is broken enough to warrant using "breaking" instead, and b) you know that and it was a typo.

Anyway. I like this thread. Keep up the good work guys. I'm only really posting here to get alerts when there's a new post. :D
 
Surely analysing the faults of the AI in GT6 is a job for PD? And it's also their job to listen to their disgruntled fans complaints re: the terrible AI in the first place; instead of virtually ignoring them for more than a decade.
 
Last edited:
eran0004
In one way it's like the entire career mode is like a big "coffee break challenge". And nothing wrong about that, that kind of challenges can be fun as well. But it's definitely missing some realistic races.

I guess somehow they came to the conclusion that the average gamer wouldn't want races longer than 5-10 minutes and then they designed the events based on that:

Step One: Stick to the 5-10 minutes mark - 3-5 laps per race
Step Two: Make sure that the player has a good chance of winning even though the races are so short - make the AI slower than the player
Step Three: Make sure that the player doesn't overtake everyone in 1st corner - spread out the field about a kilometer ahead of the player
Click to expand...
I think that's PD's exact checklist. :lol: (I really do)
 
eran0004
In one way it's like the entire career mode is like a big "coffee break challenge". And nothing wrong about that, that kind of challenges can be fun as well. But it's definitely missing some realistic races.

I guess somehow they came to the conclusion that the average gamer wouldn't want races longer than 5-10 minutes and then they designed the events based on that:

Step One: Stick to the 5-10 minutes mark - 3-5 laps per race
Step Two: Make sure that the player has a good chance of winning even though the races are so short - make the AI slower than the player
Step Three: Make sure that the player doesn't overtake everyone in 1st corner - spread out the field about a kilometer ahead of the player
Click to expand...

Probably quite accurate. And this is one of the things that baffles me about GT6. I mean, it is really very little work to put together this 'career mode' we have now, with the framework of the game already in place. All you need to do is add an event instance, and adjust the parameters to what you want them to be. 5-10 minutes work per event.

It could SO easily be so much more, if they just put someone on the job to REALLY concentrate on it for a while.
 
There's all kinds of great things they "could" do, but they've never done or attempted anything other than their current (and only) career mode.

I fully expect GT5.5 when GT"7" is released
 

Similar threads

_MJ
Strange 'AI behaviour' - custom champs
Replies
1
Views
923
Chris30
Chris30
Chikane_GTR
AI in the rain are a mess how would you fix it?
Replies
9
Views
2K
Meythia
Meythia
D
BUG: AI difficult depends on the number of cars
Replies
8
Views
1K
pedrocor
pedrocor
N
GT 7: AI pace changes according with tyre usage?
Replies
7
Views
2K
Nicktrix
N
B
AI broken at Spa 1 hour after 1.31 update
Replies
14
Views
3K
DextroCat
D

Latest Posts

Back