Automobilista 2 V1.4.3.3 & Racin´ USA Pt3 Officially Released!

Absolutely not.

They could appear realistic from the outside while having simpler physics. You don't have to drive them, so many details are not important at all.

For instance, AMS2 is known for following a tire model based on actual physical simulations, while statistical approaches are also possible (like AC does). I don't care if the AI uses a simpler statistical approach, if they have a similar tire falloff/grip loss on certain parts of the track/wet surface than the user with the complex simulation of tire thread.

Btw, if you expect AI physics being the same as user's, I think you should quit simracing for a while. Contrary to some claims in the forum, this is not done in current sims. Moreover, it would be completely pointless to do, because then each AI car should be essentially a self-driving vehicle (with accurate physics,more or less just as in real-life), what is still a research open (or at least "state of the art") problem.

 

I don't mean to sound pushy, but do my comments get in your way? Follow yours too friend, this is just a game, not a life goal.

 

Yeah it's clear enough you dont want any help, just to complain endlessly.

 

I was hoping someone was gonna tell him he needed to try the uninstall and reinstall method and see how he felt about that haha Since a simple FFB adjustment is too much I imagine his head would explode.

 

This "advice" part from you is very entertaining because of its patronizing form.
I dont think I have told you what you should do - right?

But its a bit ironic that some of my postings in this forum does point exactly to the trivial fact that the socalled physics in as example AMS2 is just some more or less simplified algoritms whoes outcome on the screen sometimes has a resemblance with what can happen in the real world - but often dont have.
But every time I have touched this subject in this forum I have been told that AMS2 does simulate the real world physics - and can therefore not be reduced to some more or less simple algoritms.
Hehe and now you are telling me that I have misunderstood this simulate physics vs simple algoritms thing.
And if I still believe this then I should take a break from sim racing.
Im not sure its me who have misunderstood things

 

Sometimes I swear I cant understand half of what you guys are saying. Wake me up when december dev update is up.

 

Is it possible that some of the wishes of having the AI on the same physics as the player comes from the AI being a little too good at correcting mistakes and driving line changes from collisions? So, not a question of physics in some cases, but reaction times and the use of inputs perfectly calculated for the situation. It's veeery rare to see a random AI struggling in the gravel after contact or a driving error and it could be in part because they always catch whatever slides they get into.

On a related topic. The addition of the programmed errors was great, but it still looks a little bit robotic and there's not much variance in the types of error that I can remember, as it's usually a mid-corner left-right-left squiggle. It could use some further refinement and additions when there's time for it. Lockups in braking zones, squiggles when driving through puddles, long slides ending in a spin with cars that have that tendency, that sort of thing. Tied into simulated components of the game like track conditions etc. where possible.

 

Take into account most of us are not even native English speakers. This means it is not always easy to manage the tone of what we write. Sorry if I sound pedantic or condescending, it was not my intention at all. I think most of the hot discussions in this kind of forum are related to that.

This is not black and white. AI has a simulation of physics going on, of course, they are not just as the ghost car that just follows a list of positions (I bet they are implemented as that). What I claim is that AI physics don't need to be as accurate as the human driver's. Moreover, this is a sane decision for practical reasons (performance), and AI limitations are not related to that approach. Furthermore, I also claim iRacing AI, (which is the best of the market in my opinion) is implemented exactly that way, based on what I've read in the iRacing forums (and in my experience).

So, as a summary, do AI cars need to have the same physics as the player? No. Does that mean they don't have any kind of physics simulation behind them? Also no! Therefore, they are still tricky.

 

It would be nice if at least the AI was affected by puddles on the road like the player is, or if they lost grip like the player does if they get wheels on the grass, or if the physics of cornering g forces and grip out of corners was the same as the player, as right now they've got some crazy braking abilities, crazy acceleration grip out of corners and in many instances they seem to be able to take corners faster than is possible by the player driving the same car.

 

after all, at the time of the development of PC1 and later of PC2, the Madness Engine was the first graphics and physics engine in the field of SIM-Racing that supported multythreeds calculations. Even with "Martin Griffihts" one of the Render Coders behind the Madness Engine
experiments were carried out to exploit logical cores as well as not only the physical ones, precisely because at the time of pCARS1 the average of cores per CPU was 4 physical and 4 logical, and to make the most of the high simulation field of the pneumatic model we also tried to make the most of the logical cores.
I remind everyone and I hope that one of the Reiza developers will confirm that the pCARS2 pneumatic model used a simulation with 128 contact points to the surface as opposed to ACC, which when they switched to 3 contact points seemed to cry out for a simulation miracle.
Not only that but the Madness Engine with the "SETA" tire model simulates the heat transfer between the tread surface, to the tread, and subsequently between the tread and the carcass, and consequently the elastic properties of the entire tire are regulated and consequently the footprint on the road. There was even talk of how the brakes warm up, heating up the rim and subsequently the air inside the rim, subsequently heating the tyre.
They are a significant amount of parameters that are taken into consideration in the mathematical model, and that the Madness Engine itself has managed to scale well on the various cores to calculate the simulation.
I remember that AC, and RRRE himself had Limited CPU problems because their graphics and physics engine used only one core in the simulation, even if they had less stuff to calculate on the mathematical model, just because 10 years ago the CPUs didn't have the same capacity of IPC.

Finally what I would like to say is that looking at the considerable amount of simulation both in the field of transmission, but above all of SETA, it would be almost impossible to replicate it on the AI, apart from having a graphics engine that would be able to process an enormous amount of data per second for each car, but also owning a CPU that could handle everything in real time, even with current CPUs I don't think they would be able to run in real time without problems.

 

And to add that they brake in corners where there is no need to do so, coasting would be more appropriate or just go flat out.

 

I have HWMonitor on all the time, as I want to have the information at hand of temperatures, usages, voltages and so on, whenever I see something suspicious at perfomance, and sometimes just for knowing that everything is working fine, specially now that we are going through a heat wave in Argentina.

An application that uses proper multithreading will see CPU usage go up to 100% overall. This is because it loads up all cores and puts them all to work. So far, I've yet to run a game which does that, % is always much lower because it leaves cores unused. It could be that I've failed to check this at all times with every single piece of software I own, so I should do my homework, run some specific tests with Automobilista 2 and report back, but I guess that this is the case.

Bear in mind that even the Madness Engine, as good as it is, still keeps a legacy of physics that go all the way back to rFactor 1, which in turn has its legacy go back even further, back towards the Sports Cars GT game from 1999. AFAIK, there are no truly new engines in simracing developed from the ground up for this. Some devs are using Unreal engine, which I think it does support it, IF the devs create the proper threads for it. But here is where my knowledge stops short. I have read that multithreading takes a lot of work to implement from a coding side, but I don't know if the reason why this is not massively implemented, is because that on itself, or the fact that pushing the quality of the simulation upwards with it would leave out a lot of users. Specially on a hobby comprised of members that value every single cent spent, and want for their investment to last for a lifetime, both on hardware and software.

 

Saw this after my previous post, it seems like I have to stand corrected in several things I've said! TY.

 

Excuse accepted my good man
But to change the direction of this player car physics vs AI cars a bit then the reason I never use AMS2´s socalled AI cars is that its so obvious that eventhough these virtual cars are Artificial they are absolutely not Intelligent.
Here Im talking about the virtual AI drivers decision making and skills.
But when it upon that also after max 5 sec use of these AI opponents become obvious that they are not only unintelligent but also are using quite more simple physics algoritms than the player car - hehe then I start the AMS2 Time Trial mode up instead.
Allways

Why am I saying this?
Because it could be fun and challenging if the racing sim I prefer had some advanced AI drivers which was able to simulate human drivers decicion making and skills.
And was not dependent of how the Reiza devs did tweak the rather simple AI algoritms of today (to try) to get rid of the most idiotic and unintelligent behaviour of these moving chicanes.

What Im talking about is something like Sonys Sophy AI.
Which is an AI based on socalled deep learning algorithms and neural networks - where the artificial intelligence inside a given boundary is teaching itself how to get fastest around a virtual race track.

That could be fun and challenging.
And could maybe make me start AMS2 up in a non TT mode.

Sony's 'Sophy' AI, Trained on Thousands of Races, Bests Humans Champs

 

I feel like I'm hearing again the very arrogant tone of this guy who used to end all his speeches with "by the way".

 

You can have 2 or 4 bot max with more complex model instead of 25 with simplified one??.
I can stand for...

 

Who also complained about forum members ‘hiding’ by using a different name on Steam.

 

I agree, and personally I'm more bothered by this than by the AI being too fast certain places. This could be class-specific of course, and I almost exclusively mess around with the Vintage and Retro stuff, but it is rather a shame that you can so effortlessly win SP races because the opponents slow down in corners you as a player can easily take flat or by just lifting a bit.

I've reported the particular corners in question a long time ago, but perhaps I should give the good people at Reiza a heads-up again.