[Balance] AI unfair cornering, traction, speed

I’m too lazy to dig up discussion about how Friends list influenced Drivatar car selection possible other consequences. As I like to keep my networks separate, I don’t have any friends in Horizon, which is also convenient in sense that’s one less modifier to worry about. That also means I don’t have any personal experience what possible influence friend list drivatars may have to Drivatar speeds and behaviour, to say for or against.

What I can say however is that based on certain baselines to eliminate modifiers, bone stock cars is one, no custom blueprints, for difficulty easiest and hardest are only ones usable, as for what I have heard sometimes in gaming industry to have planned “normal” difficulty named “above average” in final product for end users as psychological trick, as perceived experience feels better when most play that “above average”. We don’t know how this is in FH4 so for end users like us, so it’s unknown target which makes it a null point.

For hundreds of races I made for each class in Street Scene, seasons are modifier, but I think my sample is big enough, 10 cars from D to A and 10+ for S1 as there was separate discussion if Track Toys are somehow underdogs. All RWD’s and raced on every 21 Street Race events game has, no props, no wall riding, against whatever Drivatars game throwed in lineup. I actually finished all that in December and determined “The Peel” effect in D-class and about PI-threshold where it doesn’t matter anymore but didn’t ever came to post it. I probably will tomorrow or coming week.

Without friend list possible influence to Drivatar car selection etc. Ultimately races aren’t about Drivatars but time, beat the clock, win the race. For most part FH4 rubber band works very well between about D 407+ to S1 900 creating illusion of race and helps player getting faster by presenting reasonable challenge even on Unbeatable difficulty. I base this to fact that lower PI car can win higher PI Drivatar. Most races I did couldn’t be winnable without rubber band, lower PI RWD starting from 5th row (Street Scene, note) winning over higher PI Drivatar AWD(s) starting from first row.

On 3 out of 21 Street Scene events it could happen on S1 that rubber band didn’t compensated enough for player car against some Drivatar lineups. Routes were Ambleside Ascent, the Highland Charge and the Monument Wynds. I don’t know for sure if Monument Wynds got some sort of patch or has it been just dumb luck, but I haven’t encountered super fast AWD’s starting from first row when using RWD myself there on Unbeatable since, perhaps after November update. There is also Otleydale Dash where I got inconsistent results, but can’t say for sure if it’s about my skills as player that is a bottleneck or something else.

Furthermore, I encountered two additional anomalies on Highland Charge and Coastal Charge where getting faster times led into situation that boosted, not only one but two Drivatars, creating a situation where player getting faster didn’t result better position, winning the race, but losing positions. I recall there was something weird like that also on Glenfinnan Chase So there are situation where rubber band can “break” and starts punishing player for better performance. That only happened with '16 Dodge Viper ACR (S1 893) [Track toys] so as far as my tests went, extremely rare scenario.

What comes to lower classes, besides “Peel Effect” there are some problematic scenarios on lower classes too, they are rare, but exist. On B-class North Coast Rush Can-Am Maverick could finish in time equivalent to ~6% of the world Rivals leaderboard (at the time). LOL

One more anomality specific to Street Scene is “no traffic” issue. It happened to me only one day during my test runs, but for some reason Drivatars appeared to get huge speed boost if there were no traffic at all, it was not just one race, but few. IIRC at the same time my connection to Xbox servers, or whatever they are was choppy. I don’t understand what Internet connection has to do with single player events though. I recall I had one race on the Marathon sometimes during last summer or spring, where there were no traffic at all, but I don’t recall Drivatars getting any sort of speed boost then.

For AI besides rubber band, for my experience it appears to reward clean racing not being overtly aggressive if player isn’t and Drivatars can mess up pretty big at times, colliding with traffic, or with each other (latter is rare) and sometimes they can get in a bend too fast and slip off from road, that is extremely rare. They ability to react player moves is limited, sometimes they ram player car behind simply because way they are programmed to corner and / or they are in pack with limited options for racing line and braking earlier would cause other Drivatars collide with them. Still annoying. I rarely encounter them blocking on purpose, most of the times it happens because they appear to be somehow aware what’s behind them and predict other, faster Drivatar moving to their line, they are not always perfect at executing that. Many of these things could be considered as flaws, but I think errors Drivatars do makes gaming experience better, more lifelike and immersive.

I’m not always sure if everything attributed to Drivatar speed is about AI, but Drivatar physics. Speed, rubber band is relative to some sort of estimated player performance based on player car PI and difficulty level at least. On higher classes, most notable on S2+ on tarmac races at least, Drivatar super grip in rainy conditions is misleading players. People learn a lot by intuition and good gaming experience relies a lot on that and here’s a problem, learning by following Drivatar lines on S2 rain, isn’t always intuitive experience but opposite simply because player has different set of physics to deal with. There are things that favours player, weight transfer, alternative racing lines that plays on strengths of player car but they do not apply to every situation. It isn’t always a matter of S2 or rain either.

On Ambleside Ascent for example, on S1+ there are very few stock RWD cars in game that can out torque Drivatar acceleration on last third or so. Practically race is often decided in segment where’s Ambleside Approach Speed Zone or in some cases, S-Bends Drift Zone segment, who ever gets there first wins and there’s nothing else player can do about it. This creates additional issue with players having very different, yet genuine experiences. Ambleside Ascent is great route on lower classes with stock cars, it also favours certain kind of tuning, leading to some players having great experience there and others figuring out how to heck to win it, even if they used the same car, on same difficulty but different tune. I think issues like this lead players here often talking past each others and if there is much that could be harvested from this topic to build better product, I’m not always seeing it. PG contributed themselves for this issue to exist and If they ironed some things out, this topic and peer support in general would work so much better.

Personally I think that a force field is programmed around your car which means that the Ai in front of you are affected by your own speed, and the cars behind you catch up with you better. You might call that rubber banding. The aggressive behaviour might come from other players, it might come from yourself, it might be random.

Back to the force field, it just means that your acceleration can accelerate the other cars, and on Unbeatable it can make the Ai crash off the track on some bends. I have seen that happen a lot if I am catching up with the car in front he may plough off the bend.

Also if you touch the grass, or a fence the Ai will accelerate when you are slowing down which is like a double punishment.

Also if you go into a skid the car is programmed to remain in a skid until you relax the controls a bit which has a problem that you don’t get control of the car sometimes when you should.

Rubber Band, Force Field, I guess it comes to definition of rubber band, but in case of FH4 in context of this discussion, effect is the same. What I have seen a lot in Street Scene,

What comes to car control, I think it’s a problem with “virtual” clutch. Virtual in sense that even when not using manual with clutch, you can notice that if you stop on uphill section and then give your car a bit of revs, your car starts moving backwards, downhill even if you have 1st gear, or actually any forward gear on. It’s not a bug, but how real car would work if clutch were applied disengaging the engine from powered axles. The situation with slides, which can be actually pretty bad on dirt, I have been thinking it comes to that we clutch isn’t applied to release locks. It also affects turn in response. That’s why some players have wrote about how shifting on lower gears helps controlling car through curves. In real world, we could take some bend by coasting on 3rd gear (there is one good example of this on Waterhead Sprint since it’s right after downhill section) by clutch, keeping revs up and be back right on powerband once we would be on stable racing line by just releasing. It’s good example because you can still make it with right racing line, I think because it’s probably lower than 90 degree angle that turn looks like.

In reverse, this is why Ambleside Scramble is so difficult with stock RWD’s. There’s that one bend uphill to left. Player can get there fast, but on manual without clutch at least, needs to shift to lower gears not to make car understeer in turn. This wouldn’t be so bad if player could keep revs up, but that’s impossible without clutch. So losing speed uphill and losing momentum, because non optimal gearing. Engine braking might work, in a way that puts car to slide might work, but uphill corner on dirt, I would estimate that would be very tricky to get it right there.

I don’t know anything about programming games but perhaps it’s been common practice to enhance car gaming experience by introducing some sort of hidden clutch, a bit like portion of automatic transmission, except it’s invisible to players. What all sort of functions this hidden clutch we have in FH4 has, I don’t know. I could imagine softening gearing, even if players doesn’t notice it and in case of extremely powerful cars, to make them drivable at all, not to mention drifting those cars. So maybe it’s about that and not about normal racing why it’s like it is.

The phenomena you told about, I know it from Cross Country but can happen on other settings too. It’s very noticeable on AWD’s when using handbrake in certain scenarios, where all forward just momentum disappears. That also means you can’t trigger power glide with handbrake and I haven’t found proper technique to make it work at all. Instead it really feels like some sort of scripted sequence. I’m inclined to think though, that it might rather be because some sort of tradeoff for drifting physics.

One more theory I came to think relating clutch/locks. What if Drivatars has all the goods? Practical effect would be like clutch for each tyre. Practical effect would be comparable to having Drivatar cars every powered wheel being powered by individual electronic motor, each wheel, left / right being independent of each other and with AWD’s front and rear too, absolutely a monster. That would explain a lot though.

Physics, if it’s a tradeoff, don’t know how it could be fixed without influencing leaderboards. Drivatar force field / rubber band, unwinnable scenarios they tend to happen on routes where player starts to slight uphill and Drivatars starting from frontlines start downhill. There’s exception, the curious case of Edinburgh West End, which is winnable but I think it’s so because it heavily favours alternative racing lines. First slight S bend can be done really fast with really direct line that Drivatars don’t risk, following wide curve to right can be taken really fast if you can take inside, some cars can even use sidewalk. Couple of more opportunities, last being when turning right to motorway. Don’t’ know how all or any of those work on S2 stock though.

Some addendum to previous post.

Clutch / locks theory, regardless if it’s wrong or correct, there are other things regarding how Drivatars interact with environment vs human player. Good example can be seen at Castle Cross Country Circuit if it’s raced with D-class vehicles. Circuit includes a jump leading in castle courtyard from the dunes. In lower D-class player car may not be able to make jump to courtyard, yet Drivatar cars with same PI, even same car, often can complete that jump successfully. This wouldn’t be possible unless game physics worked with Drivatars differently than players.

Considering many players experience, that Drivatars can be “like unmovable objects” when rammed and their ramming tends to spin player car easily, even in scenario with similar car weight and speed, it could be speculated that Drivatars have more mass, but less weight than player vehicles. This would also explain, how in some scenarios, higher class races, Drivatar cornering ability/performance appears abnormal for players.

Other

“Peel Effect” threshold for D-class (amount of PI where Peel vehicles stop appearing in Drivatar lineups) D-407. More details here.

I’ve played many racing games, the AI in Forza Horizon 4 is the WORST…
Either their car maybe overpowered ( let’s face it some cars in FH4 is utter rubbish even if it’s the same Class as yours , it’s a fact there’s literally no freaking Balance to this game at all ) which it shouldn’t be.
At times it’s clear that the AI car is faster but you can easily catch up with it on the straight…

Also the AI is like heavily scripted…it will take the racing line no matter what…even if you’re on the inside lane, it will just ram your side and push you continuously and you will end up not able to turn the corner or get ram off the road…

The worst thing of all if you’re playing Team vs AI, the AI will go crazy for some reason…
FH4 maybe a DECENT racing game, but it have TON of flaws…yet people claiming to be " D " BEST of all racing games lol !!

1 Like

You never played NFS games then.

I have. Not much difference, really. Same cars on rails, same “super drivers”, same traffic spawn routines to screw you over, same catch up, same laziness.

I’ll give you your due though because most of these tropes in arcade racing AI have originated in NFS itself. So Forza just went with the flow. In FM6 you could ram your way to victory, which made them change things for the next few games (regardless of studio).

It has flaws, but it is still the best… games programmers aren’t usually scientists.

Drivatar ramming can be really annoying with some cars that have certain characteristics to make them delicate to handle in some situations. Mazda RX-7 Savannah in stock configuration for example. Nothing wrong with the car, but getting rammed while exiting curve, when on the edge if it’s going to drift a little getting hit from behind can easily throw it to full drift and in some scenarios on unbeatable it is very difficult to win that race after that.

I have wondered if there’s some sort of boost mechanic involved with Drivatars in last positions causing it, as it doesn’t feel as much a problem in mid and leading packs, even though relative speeds should be about the same.

Another thing is with Drivatar hits is that sometimes they sort of get stuck for awhile. With RWD’s effect is losing speed, like rear is being lifted from ground.

Problem with these discussions is that it’s rare that anyone posts about what would make Drivatars better. Also sometimes it’s not the Drivatars per se, but certain routes with certain cars. What alternatives there are?

I recall my all time favourite NFS game, Porsche 2000 and that’s the game where I started driving stock just because after awhile even hardest difficulty wasn’t challenging enough on most routes, also hated how cars started to feel the same with upgrades/tuning, which is realistic to the point, but not always very entertaining. NFS went to other direction after that, let’s say it took the road and I took the highway and I don’t know how it has been since. In comparison for the definite simcade of the era, Gran Turismo, NFS Porsche 2000 was the only multiplatform game that could challenge it.

I recall one terrible Test Drive game, years before TDU duology. It turned out to be one of those games that gave so called Rubber Band AI its bad name. How it was implemented caused all the disdain and it was deserved. There were no limits in AI car boost. I recall doing perfect runs for 5/6th and was going for victory with about max speed and then AI car blew past me, like in some sped up film. Better you got, worse your odds were and it wasn’t only game like that. Design punished players being too fast by creating impossible to win scenario, it was a ridiculous hack and deserved all contempt it got.

Horizon’s Drivatars are IMO somewhere in the middle. Odd scenarios aside in Street Scene, Dirt Racing and Cross Country can be something else when raced with RWD’s. The big issue is that seeing Drivatars doing something which is impossible with similar PI vehicle tells player just that, not WHY it’s happening and what player should do to improve. No wonder AWD swapping is so common.

S2 on wet surfaces is something else with stock vehicles. Some can be good some can be terrible and I just read this from Forza Reddit. Don’t know if there’s anything to it: Reddit - Dive into anything

You can’t really explain what makes a drivatar better to people that don’t program computers. Ai can learn what other drivers do, and copy that behaviour, and it can learn to get round a course faster, and faster without cheating. It can learn overtaking techniques, and it can learn how to drift when it needs to. The problem is that the human has to learn how to program the Ai to do those things.

There are more factors here than Drivatar AI. What I really meant was what would make Drivatars better experience in game. If there really is rain penalty like in Reddit post I quoted and if supposed penalty doesn’t apply to Drivatars we would have explanation why S2 Trials tend to get negative feedback about every time they appear on Playlist. I actually quoted it because I remembered one of your posts about psychological trick, making cars slide to make them feel faster. So we could speculate that problem that programmers tried to solve with rain penalty was that cars S2 cars didn’t feel fast enough compared to S1 cars. Well, according to feedback if that’s the case, their method sucked. It’s doesn’t work well with the Trial (important for selling the playlist and related services), it’s counter intuitive for players to learn the game in two accounts 1. It doesn’t help with learning curve which is made worse that it’s inconsistent with experience with lower classes. 2. it emphasises player experience that Drivatar racing lines are “magical” leading to whole new issue which is breaking the player experience of immersion.

So let’s say if rumour were true, we have a scenario where programmers supposedly solved a problem and in a process created new ones. Really impressive, oh wait… LOL

I don’t say bad management doesn’t happen because it does, but there are also other scenarios. I find it hard to believe that some sort of super AI would appear and be implemented in FH4, I sure it would easy to assemble team which claims they could do just that if there were money involved. It doesn’t matter then if they were super autistic team trying to solve right problem in wrong product or just scammers disguising behind all sort of mumbo jumbo after money and supposed status. What matters is that MANAGEMENT is involved and committed towards enhancing the product with solutions that can be actually executed.

The psychological sliding trick is due to XBox One not having power to actually increase the car speed any higher. When the new XBox comes out you will see better car physics, and better Ai. You will probably get better sliding collision too off the trees. When you skim a tree at the moment you usually stop dead when in fact you should slide off it, it can be confusing, but that is down to processing speed. Sliding collision off every tree is an expensive test for processing speed, but on the new XBox they could shatter bark off the tree, and all sorts.

Where is this trick documented as being a thing?
FH4 is part of the Forza franchise which is a tuning game. “we” would notice if cars were sliding more than they should.

yes, and I noticed.

I agree!

AI’s needs an overhaul - like total re-write the code. I know this isn’t something you do over night, so I hope in the next Horizon game AI’s have been changed for the better - oh an also, more complex engine sounds but yes, that’s for another post and not this.

1 Like

I’m not sure with whom you are agreeing with. My take on this is that certain could improve how players experience Drivatars. Say supposed “wet penalty” were true, if that effect were toned town, players would probably feel they were more competitive against Drivatars. Is that an AI thing? No, it’s a physics thing.

In Street Scene, perhaps for some routes Drivatar lineup could be scripted in a way that prevents certain odds happening. Perhaps Peel vehicles could be scripted to use less upgrades that contribute to insane acceleration. Perhaps certain B-class (or was it C?) buggy could be tweaked in a way that it would stay within top speed limit stats says it should achieve.

Things like those fascinate me because they might be actually implementable. Those actually wouldn’t do anything to Drivatar AI in general, driving lines, etc but instead fix few other oddities. Yet players might perceive it like Drivatars were improved and have better experience.

I don’t have as much to this game as I used to, but I run tests some new and some older cars. Raced them bone stock vs Unbeatable on every Street Scene race there is. New Toyota Sprinter Trueno, Rover Vitesse, Lexus LFA, won every event, so those are all Street Scene safe. So is Mazda RX-7 Savanna I mentioned earlier, Ford SVT Cobra R is also Street Scene safe and very easy to handle compared to Savanna but Savanna has character, kind of experience I look forward. Encountered oddity with '73 Pontiac T/A not surprisingly on the Monument Wynds, which I managed to won, but after needed several retries and didn’t beat the best time AI could set in that lineup, even though it was amazingly good on other Street Races in Edinburgh.

I’m doing another Street Scene run now with Caddy Limo for joke, but so far I have won 11 of 21 races. Monument Wynds is probably going to be something really interesting with it though.

S2 might be it’s own thing for classes. CC and Dirt sure are for race type and compared to Street Races there are some power routes, Water Head Sprint being most infamous I guess, that can be really difficult on Unbeatable and Custom Blueprints with power sections can be even unwinnable on Unbeatable with stock, but overall I haven’t found anything to support thinking that FH4 AI is somehow universally terrible.

It would be for the best if we could get some hard evidence.

I don’t race nearly enough S2 to say anything about it. S2 difficulty is bit of a curious thing though and one way to measure this might be running some tests on Road Racing track with stock cars very close in PI.

Finished my Street Scene run with Cadillac Limousine, all wins. Inspired by this topic tried '09 Pagani Zonda Cinque Roadster (S2 908) and it was really great even bone stock. Had difficulties on Monument Wynds but that’s mainly because new LEGO Bugatti Chiron may be broken there. I was able to win the race when it crashed to something but despite starting from about middle of grid it was setting times around 2:07’s while my best was in 2.09’s maybe 2.08’s. Won against McLaren 720S with 2:09.313.

Other than that it was easy run through. It was raining only in couple of races though, as it was Summer in Horizon.

For my runs, experiment with Cadillac Limo was proof how good FH4 AI can be adapting to almost* whatever player takes on route on tarmac races, experiment with Zonda is as good example of how inconsistent it can be sometimes. '09 Pagani Zonda Cinque Roadster would be about perfect starter car for S2, except if one car (Lego Chiron) appears in Drivatar lineup, in one race (Monument Wynds).

  • Cases like stock VW Baja Bug where lot of PI comes from its lateral grip, which isn’t ideal feature for Street Racing in particular case as speeds in D-class tarmac races never get high enough for it to really benefit from that lateral grip.

HI everyone , am I the only to find that our cars are far to have the same grip , acceleration or jump reaction than drivatar’s ?
In every race i never see drivatars slipping with their car , or on a jump with the same car the same speed the same angle and trajectory we never land the same way than them the always land further with no rebound .
Sometimes i think that devs really don’t equilibrate the chances between players and drivatars , opponent have very rarely random cars in their trajectory , the always seems to have better grip , acceleration and stability than you .
I know we need some difficulties but sometimes it seems that the game is making fun of you

1 Like

There are some different physics involved. It’s easy to notice at Castle Cross Country circuit if you race it in D-class. If player car PI is low enough, jump from dune to courtyard isn’t possible, no matter what, yet Drivatar with same or similar car can often do it just fine.

Someone made really great post about tyre heat and grip but I can’t find that post. I have come to think though, that tyre grip doesn’t really matter for Drivatars in all situations. Drivatars can have perfect stable lines in Road Racing right after start, while player, depending of car, needs lap to get tyre heat up for best grip. In Street Scene I see Drivatars sometimes losing grip and going all over the place sometimes, so I’m not sure what to think.

On most jumps I overshoot the Drivatars and land beyond them. Annoying is when I land on them.