Latest Roadmap Update - February 2018!!!

I guess it comes down to the fact that rF2 offers AI that behaves somewhat human like in certain situations and offers a more consistent experience. Offline racing in rF2 is, where other people do pickup racing in other sims. I got to say I enjoy both alot especialy when I want to race a special car and track combo. And a very big advantage is, that rF2 AI doesn't drive unfair most of the time and that I don't get t-boned by some idiot in T1 everytime. Quite often it offers the more plausible way of racing if you ask me. Atleast when we're talking rF2.

 

RF2 ai best in the biz.

nuf said!

funny thing is it hasn't been touched in years lol!

 

I don't see a predictability problem with rF2 AI. The problem is its inconsistency between different mods in order to get a realistically fast AI for a given strength. It seems to be very difficult for track modders to create good AIW.

Some cars brake at full throttle turns giving usually very good opportunities to undertake in a not very realistic way.

The algorithm that controls car inputs to follow a determined path as fast as possible is obsolete. It hasn't changed since rF1 times. Todays AI techniques have improved a lot. Just see Deep Mind's Alpha Zero revolution in chess.

The track learning approach doesn't seem to have helped. Besides none of them provide a new AIW to be used which would really prove how effective they are when trying them with other cars or track conditions.

Besides all that and as I have raised in other occasions, modders are provided with the so much driveable Skip Barber in dev mode so that we create fast paths to be used with very different cars...

Furthermore, it seems impossible now to use an official car to do this task since all of them are encrypted and gen files are no longer accessible. We have to rely on old or 3rd party content to do this task.

No wonder AI behaviour is stuck in progress.

 

It'll be interesting to see how AI development evolves and I am very much looking forward to new possiblities with hopefully improved AI within rF2, especialy if there is stuff like offline championships coming with the new UI. There is also a quite interesting to watch video by Kunos talking about AI and considering that they will use their AC AI as a base for ACC I see some very big issues. But yeah, development in that area has been stagnating for a couple of years now and it feels as if rF2 was the last of them getting it somewhat right, but even within this sim it has been an underrated front. I am also looking forward to iRacings implementation. I guess, people often forget that it is the first step to create racecraft before you hop onto an online server and I am a big fan of the concept of training with AI, may it be FPS games or whatelse.

 

Although rf2 AI is very good it would be interesting to see google self-driving car software (AI) implemented in modern sims and tuned for racing behaviour.

 

Formula E hosts an AI racing competition. For what I have read, the state of the art of this technology is currently behind human skills. The human driver beat the AI by 7 seconds.

NASA also organised recently a drone flying competition with similar results regarding human being notably faster. On the other hand AI was more consistent.

For what it seems the main difference is that humans learn much more efficiently than machines. However, this is bound to change in not so much time. Let's hope that video games benefit from this technology as well.

It should be noted that in the same way as for tire model, a too much complex AI algorithm could also limit the number of AI using it.

 

Does anyone have any idea of the reasons why ACs AI uses the same physics as the player and RF2 doesnt?

Is it something to do with the empirical tyre model and their car number limit? Really would like to know.
In RF1 the AI also doesnt use the players physics.

 

AI uses TBC tyre model which is more scripted and faster to manage computationally. Human uses TGM tyre model that requires accessing big lookup tables. Using this techniques for all AI drivers would be too heavy for the CPU.

 

But then AC also has a limited number of cars at 30.
In the new game that is going to be a problem for example with grids with more than 50 cars in real life.

Is the deformation lookup table processing really that much heavier to limit the car number to something like 5 cars?

But RF1 also doesnt use RF1 player physics.

 

Also rF2 AI runs rF phisycs...which is still the base of most Sims out there...so...

 

I would suspect that AI managing is not their bottleneck. We have seen in rF2 much bigger grids in people's tests. I remember someone putting around 100 cars at Nordschliefe.

 

Kunos said that AI is the bottleneck in the racedepartment videos.
And that the bottleneck is caused by the choice to use the same physics as the player.
Dont expect anyone to watch it because it is way too long so you are going to take my word for it but here it is anyway:


Just something that may be interesting as an option for RF2 in the future.

 

if rF2 AI used player physic you'd run like 3 or 4 AI max. You can have a good AI without using player physic
AC does a mistake by using the same as the player IMO. That also says alot about the said physic...
They'll have to change that anyway if they want a Spa grid. I'm almost sure I can run 60 AI in rF2 with my i3 xD (in DX9, and in DX11 once they fix whatever bug caused the AI to use too much cpu)

 

I thought ACs physics were lighter due to their use of an empirical tyre model. Like the ones Niels explains in his articles.
But I agree that it may be a mistake for ACC because then you just cant simulate real life grids.

 

'roadmaps and deadlines or AC'.
I don't think that's too much to ask especially seeing as there is a forum for it: Off-Topic / Other Games

 

Well that worked , tumble weeds now .

 

Ok, I will talk about the February roadmap...what is the expected date of delivery of these updates that you outlined?

 

Sooner than ACC (that game with crappy physics)

 

I would love to have an estimate (2 weeks / 1 month / 3 months).

I am like a lot of users very enthusiast and ... I must admit it, I am very impatient !

Nevertheless, I do not think @Marcel Offermans will do it, but I hope so.

Their logic of communication follows a strategy which they consider the best from their point of view, there are very high commercial stakes :

For a refractory simracer at RF2, the arrival of the new UI and the performance can appear like a " new simulation ".[/USER]

 

I just came across this message dated 28/02 on a Dutch IT site :
https://gathering.tweakers.net/forum/list_message/54438049#54438049

Translation :
The update for Zandvoort I expect the end of March. It is not possible to precisely determine but this coming week we are testing a large update internally. How it goes from there depends on the amount of bugs we find.