AI: Which game had it right?

[CoolColJ]

Good article about the AI system in "Pure"
http://www.gamasutra.com/view/feature/132313/the_pure_advantage_advanced_.php?page=1

I can see something like this working well for GT6
so that it cater for all levels of play, along with a global difficulty level to preset the challange

I have noticed a similar style of grouping in some races inn GT5 and GT5 spec 2.0+ arcade races.

edit - from one of the comments at the bottom of the article

I am continually surprised every year with the newest racing games that still use the heavy handed and one dimensional rubber banding methods and I have been even more surprised when multiple racing developers I have worked with the past few years have had little interest in correcting the rubber banding limitations that I already knew how to solve. On both Road Rash 3D and Road Rash: Jailbreak (of which I was Creative Director) we did away with the lame one-dimensional rubber banding techniques in favor of Dynamic Speed Management and the results were the best delivery of racing competition in any of the 9 versions of RR or IMOHO of any of the other games on the market since then (sadly we still got beat bad on graphics once the original, market-changing Gran Tourismo came out two months before we shipped)

For RR: Jailbreak the lead AI programmer and I evolved the Dynamic Speed Management system further to treat the pack as a whole like one long accordion which can be expanded and contracted by adjusting target speeds of the first and last rider ahead of the player and first/last behind the player (interpolated for speeds in between). We also heavily limited the best case speed advantage of the opponent bikes to 2% so they will never blow past the player who is at or near top speed.



All the speed management parameter settings would change based on race progression (discreetly for each 1/4 of the race) and on player performance (race place). When play balancing these systems I consciously formulated the Pacing I wanted - one that spread the pack out in front of the player at the start, made it easily to pass each rider one at a time with at least one more opponent in view for the 2nd and 3rd quartiles, then later in the race contracted the pack so there were often 3-5 bikes jockying for position near the finish line when the player is doing well (in the top few places). Because the settings were tied to player place rank we could adjust the settings for various performance and ability levels to increase the sense of excitement and competition towards the finish line but without making it too easy for a player who did really, really bad; my criteria for tuning was that a player who crashes twice in the first 70% or so of the game should still have the opportunity to finish in 1st place, but certainly not every time. The tuning trends of my Pacing structure was maintained from level to level in the racing mode but the competition was increased sooner and more fiercely at higher levels and player mistakes were made more costly. Each of the other modes (Five-O, etc.) had their own unique pacing structure and supporting tuning trends.



These systems proved to deliver exceptional competition game play in the single player racing mode IMO because they ensured the player always saw other opponents during a race (which is harder to do when there are no rear-view mirrors). It also kept the opponents from bunching up into large swarming groups around the player (which will prohibit passing) and it prevented the annoying slingshot of opponents that all of a sudden blow past the player when they are already at top speed (which would otherwise expose an unfair speed advantage). One of our key philosophies was to never make it appear the opponents had any unfair advantages so in addition to capping their top speed targets they would also slide out around turns or crash into cars at roughly the same rate as the player.



I believe strongly that Dynamic Difficulty Systems in all genres are the wave of the future because they allow designers to make a larger sweet spot of fun for a wider range of ability and performance levels AND because they can be leveraged to adjust the Pacing of a game. So despite what you read from Ernest Adams (who can have interesting things to say but the bulk of his experience lies in programming and not so much in design), Dynamic Difficulty has a huge future in racing games and racing developers would be wise to follow Eduardo's lead on Dynamic AI systems in order to strive to improve the quality of the single player experience through the fun of competition in the years to come. I believe the racing genre is far behind the shooter and platform genre in terms of AI development so I look forwards to a time where racing developers start to implement and iterate new types of racing AI. It would be cool to see new AI systems innovations coming from the racing genre instead of only from the shooters of tomorrow.

 

[Dodzzz]

Time to bring out some of LOL worthy GT5 AI vids
To be fair these were in the older versions, AI is much better now

AI Punt

LOL he looks really angry

 

[Tappajakoala]

Gran Turismo Prologue 4 had good AI. Some of license test were racing tactics making decoy move to rival and getting pass other side etc. Rival cars tried to make actual defensive moves. Pity they didn't use similar for GT4 and GT5.

 

[MoLiEG]

GT3 has been the only GT game I've actually felt the AI had some skills...

 

[Samus]

GT3 has been the only GT game I've actually felt the AI had some skills...

Yeah because they cheated.

 

[the_racer100]

I think Need For Speed Most Wanted (the good one) had a "rubber banding" option. If you selected the "catch up" option, if you were to be left behind, the AI would slow down, and, if you were in front, the AI would speed up.

But Forza Motorsport 4 has THE most agressive AI of all times, just look at M.Rossi, P.Muller and R.Sasaki... 98% of the time, I'm rammed off by those three...

 

[Samus]

I think Need For Speed Most Wanted (the good one) had a "rubber banding" option. If you selected the "catch up" option, if you were to be left behind, the AI would slow down, and, if you were in front, the AI would speed up.

But Forza Motorsport 4 has THE most agressive AI of all times, just look at M.Rossi, P.Muller and R.Sasaki... 98% of the time, I'm rammed off by those three...

have you played Ferrari Challenge? The AI can be pretty fast in that game but their 'defend' algorithms could do with some work to say the least.

 

[SavageEvil]

GT5 has the best AI out of all GT games bar none. Problem with GT5 AI is that it has two different modes which you can plainly see in GT Mode and Arcade mode. Arcade Mode has the ability to set the AI to their maximum difficulty, which depending on your car you'd be hard pressed to win the race in under 4 laps on real courses, the created courses like Cape Ring the AI are weird on probably because of the giant loops which the AI tends to never push the car hard on.
GT5's AI also doesn't seem to have the ability to understand the length of races either, they don't exhibit urgency when nearing the final lap they drive the same pace as long as their tires allow it, they also drive the smoothest was possible to keep tire wear low and this goes for whether or not tire damage is on and also the AI never ever drives the car without some pressure from an anomalous entity(you) very hard. They never run the car hard, they don't hurry up and don't attempt to reclaim their lost position. Now let's couple this with them being unable to tune their cars to match your PP and you have a recipe for lack luster AI.
In Arcade Mode the AI always tends to drive on whatever stock tires the car is locked with when you purchase it in game, sport hard. I have raced the AI countless times in Arcade mode on professional difficulty, they are far less friendly and back off pretty much never only if you took the line away but they'll be on your tail but won't try to out brake you to take the position back. I hate the fact you can't force them to tune their cars so I can have a level race PP wise. I'd choose my 550PP Mustang Cobra R 517hp and 1324kg with rear spoiler on sport hard tires and end up in a pretty ok race on Laguna Seca or any course that has a smooth realistic layout. Depending on the course the chances of winning changes, problem is all the cars you race against are in stock form running the same tires though. It's a winnable race no matter the course, but it does show that the AI is still a mix of level 40 and down to about 30 or lower. If you drive in a bone stock ZR1 you'd find yourself in an impossible to win race simply because the AI has a much faster and higher rated car than your own. PD needs to add a few more algorithms to GT5 AI one where they will actually try to win the race and try as hard as possible to keep their lead by pushing their car and to take into account the length of the race so they can go all out on shorter races and dial back on lengthier races. They should also add rival type emotion from some AI depending on how you attacked and passed them and if their car is up to it, no more than 2 and they should also be able to have rivalries with other AI drivers too. Would create some pretty interesting replays to view as two cars attempt to battle for position and a third just creeps in and takes the position away and then watch as they then try to hustle and team up to regain the lost positions.
AI can be improved and the extra easy AI in GT Mode needs to be done away with, stop treating us like idiots who don't get better with time.

 

[Dodzzz]

Yeah because they cheated.

Yeah I remember this clearly..
I was leading by 20+ seconds and the AI came out from the pit, and suddenly drove in an insane speed (like it was fast forwarded 1.5 times ignoring any kind of physics) and came right behind me.. that pissed me off to know that such cheat exist...

 

[Samus]

Yeah I remember this clearly..
I was leading by 20+ seconds and the AI came out from the pit, and suddenly drove in an insane speed (like it was fast forwarded 1.5 times ignoring any kind of physics) and came right behind me.. that pissed me off to know that such cheat exist...

Yep, it's called rubber band AI. Drop behind them and they start going slower, pull away from them and they start going faster, sometimes much faster than they could possibly do in terms of physics.

 

[Ills1999]

I think the problem is that game developers are trying too hard to keep the racing close.

Who says you can't have fun lapping the field?

 

[SilverArrow]

Why aren't the AI in seasonal events as good as the professional difficulty AI in arcade mode? They're so slow and at times completely unaware of your existence. They drive me off the circuit then sharply brake (coincidently to the same speed as I can go on gravel) making it impossible to get back onto the track until they slowly accelerate away.

 

[Eunos_Cosmo]

NFS shift had decent AI. They were too aggressive, but they actually behaved like human beings and made mistakes occasionally.

 

[nostuner03]

If they could program the GT6 A.I. to act like GTR2's A.I, then we'd be cookin' with gas. It knows how to block, overtake and, adjust to random situations from what I've experienced.

 

[CoolColJ]

If they could program the GT6 A.I. to act like GTR2's A.I, then we'd be cookin' with gas. It knows how to block, overtake and, adjust to random situations from what I've experienced.

But also bad in others, they are terrible with chicanes and other oddities

 

[MoLiEG]

Yeah I remember this clearly..
I was leading by 20+ seconds and the AI came out from the pit, and suddenly drove in an insane speed (like it was fast forwarded 1.5 times ignoring any kind of physics) and came right behind me.. that pissed me off to know that such cheat exist...

For real???

I never experienced that... or maybe I don't remember it... It's been about 5 years since I touched GT3 (sold my PS2)...

 

[nostuner03]

But also bad in others, they are terrible with chicanes and other oddities

I honestly barely ever have problems, with chicanes and such. But, it has happened before.