Something has to change in ranked MP - this is'nt fun anymore.

I just got dq in Imola by "earning" 30 incident points and i am pissed as fuck.
To be honest, 2 points for leaving the track and 4 points by touching on of the opponents car are absolutly my fault, the rest is done by other cars bumping and crashing in me.
I think it's time to talk about a different way to allocate penalty point. It's not fair to get points for mistakes or bad behavior of others.
I would love to share the replay file of the race, but it's to big to upload (even packed). I bet my ass, if you watch this you won't believe i was the one who gets disqualified.

I am not a programmer or AI specialist, but it must be possible to find a solution for this problem.

 

How? If you can come up with a way for an automated system, with no human interaction, to correctly determine exactly who is to "blame" for every incident, then you stand to earn some serious money... because nobody else has managed it so far. It has nothing to do with AI, it's to do with the fact that an automated system simply cannot ever apply human judgement in it's calculations.

The problem isn't the system, it's that people are expecting far too much from it.

 

https://forum.sector3studios.com/index.php?forums/leagues-communities.70/

 

The easy answer is qualify or start better and then you will be ahead of the carnage or start last and wait until the noobs get DSQ to move up.

 

I think starting from position 4 in a 27 driver field of GT1 and GT2 is pretty ok. I am a very defensive driver who tries to survive the first corners/first lap of a race, even if i loose positions. But most of the others try to win the race in the first lap, because they have no patience. I started simring with GPL ans Nascar Racing 2003 later, both games where you HAVE TO clean racing. On touch and you are out of the race. Sometimes i think nowadays all others started their racing carrer with Gran Turismo or Forza Horizon...

I think collecting data for a statistic model might be a first step and helpful.
Let's have a look i.e. for a typical crash in the braking zone of a corner. B is in front, A is missing the breaking point or tries to overtake. Lets say you watch 100 replays of such crashs, you will find out in ~75% A is making the mistake. So why not split the incident point 75/25 for driver A/B?
Another example what i often experience - driver A is loosing the car and left the track. Instead of waiting (for the field to pass) he tries to re-enter the track without watching the traffic and crashs in another car. Again, watch 100 replays and you will find out in ~100% driver A did the mistake. So split the incident points 100/0 or A/B.
Typical WTCR situation. A and B lapping the track door by door, banging wheels, having contact, but both stay on track and just having a good time. Why allocate penalty point at all if nothing happens?

I am totally aware of the problem that you can't code an ingame race director, but splitting incident points 1 on 1 is'nt the king's path for all incidents.

The replay file I am talking about: https://drive.google.com/file/d/16BqtKhbJb9lf2pTR-9SEFdrAE8SqO1CO/view?usp=sharing

 

How about an AI trained algorythm that learns based on a human making the determination initally and then human corrections of its decisions initially untill its trained.

A bit like how Nvidias DLSS algorythm learns based on a refrence image and then upscales a low-res image to a high-res image based on the "16k" reference image.Initially this had to be done on a per-game basis and the system learned a particular game, now that the AI has learned enough games it can be applied globally to any* game becaus the AI has run many billions of simulations its learned whats common between all games etc etc...

A human would initially tell the AI "this is driver As fault" it doesnt have to specify a reason or anything. after thousands of inputs the AI can recognise a pattern in what is whos fault. then the ai can train its self and a human can make corrections to further refine it by studying human race replays etc. eventually it would be perfect.

Im not saying this would be possible for S3 to implement or cheap to do, i dont know, im just sayin "come up with a better way" - there it is... it would be a good way to do it, that would over time get better and better. To the point where AI can now recognise cancers in scans better and much faster than a human oncologist.

 

Until we can all afford super computers the system will have to remain rather simple, this I can understand but for the love of Satan why oh why are track re-entry destroyers not immediately struck by lightening and toasted is beyond me. If Sir Lewis just waltzed back on track killing 4 people his super licence would be shoved so far up his own arse it would never see the light of day again. So why can Johnny go quickly, kill 4 hop out and go do exactly the same thing again and again and again? It's time the rankings actually stood for something, if re-joining the track cost 50% of your entire ranking and a weeks ban, it would stop overnight, there is no jeopardy for anyone as things stand, you know, perpetual beta, maybe time to have an adult discussion on how to go forwards, as things stand it's hard to tell the difference between a ranked race and a Wreckfest race. Some people need to do some practice before jumping in the ranked servers for starters and of course, will the player base ever allow for split servers. One bit of advice I can give, don't bother writing each lowlifes name down, as there just aren't enough trees on the planet.

 

The big difference between all those impressive examples you're citing and trying to apply this to the problem of deciding the "who dun it" in an online racing scenario is the introduction of humans.
A running AI, a picture or a cell, they all have one thing in common: They don't try to "cheat" an observing system, because they are unaware of its existence.

Any system that distributes the penalty for an incident unevenly has to work flawlessly. If said system fails for whatever reason and you end up receiving the greater share of the penalty despite being the victim, wouldn't you think that's even more unfair than sharing the blame equally? I certainly would.
And with humans in the equation you can never fully judge their intentions. An AI might be able to judge normal racing incidents but as soon as you add malintent (brake checking f.e.) things get very fuzzy.

 

I agree initially it wold not work, but thats not when it should be implemented/used. It should be used when fully trained to be agreeable to a human 99.9% of the time... even humans make mistakes and referees get a lot of stick for it.

brake checks and things like that can be totally accounted for with AI, anything can really with enough itteration. you would just have to feed the AI 100 or 500 examples of what a brake check is, then tell the AI to tell you what a brake check is, then you correct what it got wrong, and repeat that step a few times. eventually the AI would be able to detect a brake check with close to 100% accuracy. how it does it is not really relevant or even understood by even the people that program these systems, they just have the correct outcome and thats all that matters.

accidental or intended is not realy relevant either IMO. if you made a move that fails and ruined someones race then you should be penalised, even if you had no intention of doing that.

 

you can actually do AI simulations on a 'normal' computer. all be it a lot slower, its not neccesary to have a supercomputer to build your model, then you can send off/ pay for the model to be itterated bilions of times by someone with a supercomputer. Like simulating a billion races that could never be done by a human. its not as far-fetched as it sounds or it would have sounded even 2-3 years ago.

 

The fuzzy bit isn't the brake check itself, it's the various other scenarios that could look like a brake check but are actually racing incidents, or the following car braking too late on purpose in order to get rid of the car ahead.

By now we've seen hundreds of incidents where even we as humans couldn't decide with absolute certainty who was at fault or who intended a crash.

And as said above, imo 99% accuracy just isn't good enough when it comes to a system that is supposed to determine who caused a crash and deal out uneven penalties based on that decision.

All of this is not to say that it might not be worth investigating such possibilities and in general looking for ways to improve the system we have currently. It's far from ideal, we all agree, but anything that's supposed to replace it has to work better without introducing borderline cases.

 

There will always be borderline cases, VAR over here in Blighty is causing all sorts of stink inside footy, as even when the outcome is 100% correct, it's not the outcome some agree with, let's say it lacks nuance but then why should it have any, it's surely black & white but VAR has definitely thrown this concept under the bus.
Surely the game knows if a car is off track and when it comes back on track, it also knows if it made any contact on it's return to the track, so why the hell can't it also return it directly to the pit with a huge penalty?

 

What if you do re-enter as you should but another driver decides that he wants to give you a friendly nudge, knowing that you'll be the one who'll be DQ'ed for it?

So it would have to have a concept of what a safe re-entry is. You cannot expect everyone who left the track to wait until every following car has passed him. You can re-enter the track safely while there's other cars around, away from the ideal line f.e. It certainly wouldn't be impossible to create such a system, but again there's very many potential scenarios to consider and it'd be quite a task to make such a system work without having borderline cases while keeping CPU load justifiable.

 

Yes i understand what you mean but what im saying is an Ai can account for any scenario as long as you have examples of it to train it, and can correct and inform the AI eventually you will get the same accuracy as a human without the human.

There is no such thing as 100% in this kind of situation, not even a human (that you are trying to replicate) would not be correct always, however, the race director is the authority so agree or not thats what happens.

But ignoring all that theoretical AI implementation i think some small change are merited.

A car behind hitting a car in front who broke at the correct braking point for the corner should suffer more penalty points than the car in front. The exception would be if the car in front broke before the braking point at that corner that they statistically usually did in previous laps and only then if there was no car braking in front of them. in the latter case the initial car in front that caused the accident by braking X% too soon should recieve the penalty for brake checking and causing a pile up to the 2 cars behind. (yes it may have been an accident and under pressure they broke too soon, but they caused the accident!)

Pretty hard to program with IF THEN statments but i dont find it fair that i can "utilise" some of my incidnet points to just push a slower but defencive driver out of the way and scrape down the side of them, push them off the track and they suffer more penalty points than i do (for contact and for leaving track). things like that could do with some attention.

Again, im not saying "just code something, its easy" - far from it, its a ball-ach im sure, but working on it rather than not is progress. test some things and see. but i dont know anything im not a programmer so ignore me

 

Good point!

On real tracks there is something called "race accident". If no judge can decide who's fault it was no one gets a penalty. Maybe something like this could be implemented? In generall i have no problem to have such a accident, something like this happens in racing. I've made mistakes also and i stand for it. But getting incident point for every little accident (no matter who's fault it is) is not ok. 28 points in 14 minutes of race, thats ridiculous for someone who is in simracing for 23 years now.

 

i feel for the devs i do, its so hard to do. as it is now, its not "bad" its fine, but there is always room for improvment.

 

could there be curated race directors as actual players who are vetted a bit like forum moderators, community members who actually join and watch the race and can make a human decision?

I know u can dish out penalties in the dedicated server but how about something like that for the ranked. the current system is in place as normal but a director can override the penalty if they found it incorrect.

or is that something that is definitly a bad thing?

just thinking out loud. not really thought about it much

 

I think that would require far more manpower than S3 could afford.

 

The current system is not flawless either but you decided to stick to it nonetheless. And I think the reason is quite clear, if perfection can't be achieved, you can settle with something that's "good enough". Why do you think 99% successful system isn't good enough? Do players think the same? I mean they are who have to deal with what the system produces, so their opinion should matter...

That's what's happening right now, no? If an accident is not you fault, you get more penalty than you deserve.

I don't think that evenly distributed penalties matter to people all that much, it's rather fair vs. unfair that matters. And in that case, the current system is often unfair. So, if we have two systems that both aren't 100% correct, but one is correct in more cases than the other, then it means you can have a meaningful choice which one to use.

What are we supposed to do right now when see a "bad player"? We report them. Can't we do the same with a new system too?