A Gateway to the Future: Neural AI in Simracing

The article is interesting but it covers a lot of other stuff and only halfway through there is something about the topic. Also, can you give us some example of neural AI? Otherwise this thread will be the usual 'my AI is better than yours'.
An example of AI is AlphaZero, which smokes any human world champion at chess or go. I believe the term "neural AI" is not really used, but programs like AlphaZero do "learn" like described in the article.

The real challenge for simracing AIs is not to make them drive perfectly (they easily would) but human-like. My guess is that will be achieved shortly as a big part of the technology is already there, in real-life autonomous cars.
 
AI is just really complicated stuff. I think its mainly down to finding the AInstein (ooff) who is the cleverest at programming something that is low enough on CPU usage but human enough in behavior.. AI is all hip and trendy and I'm sure lots of cool things are happening, but at the end I feel the main part still is the cleverness of the programmer here..

True @Neils_at_Home , but aren't we just on the cusp of a huge transition in programming where the GPU becomes the heart and mind of a PC probably more so than the CPU (hence Nvidia deep learning, Intel returning to the GPU market).

Isn't that going to usher in an AI revolution?
 
Just watched the video about the MotoGP 19 game. All very promising. Still, that is a promotional video to sell a game that is not out yet. It is not a documentary on neural AI applied to simracing. I would be a little bit cautious. In any case happy to see someone (Italian like me) doing something ambitious.
 
I rather the AI be used for vehicle tuning first. I want a virtual "team" behind me as a driver.

One one hand the sim is trying to replicate real life. On the other hand, I'm the driver, mechanic, engineer and strategist for a GT3 team which is absolutely not close to reality at all. Those functions should be AI driven and combined with audio feedback as a driver would give. "I have understeer in T1 and oversteer in T14" etc.
It sounds like you are under the naive impression that a RL pro driver dont have to bother with setup knowledge.
No driver would be hired in as example F1 if he was only able to give some borderline useless info like "I have understeer in T1 and oversteer in T14".
As an example Kevin Magnussen was prefered by the McLaren team against some other junior drivers because he was able to give very precise useable feedback to the pit.
And you can be sure that even the drivers in the bottom end F1 teams is able to discuss and understand setup parameters on a reasonable level.
Otherwise they would be without a job.
Only taxi drivers can be hired without knowledge about what is under the hood.:)
 
AI is just really complicated stuff. I think its mainly down to finding the AInstein (ooff) who is the cleverest at programming something that is low enough on CPU usage but human enough in behavior.. AI is all hip and trendy and I'm sure lots of cool things are happening, but at the end I feel the main part still is the cleverness of the programmer here..
Speaking of which...Whacha up to Niels!?
 
Computer is not intelligent. It's just algorithms. Back when I was a kid the other cars were "computer controlled", and they always will be.
AI lack racecraft, which is where AI learning can be better applied. The technology for fastest race line, even on the fly, is already present, but choosing when to pass, when to block, when to stay locked with a drafting partner, when to pit for tire change, and looking out for cars re-entering the racing surface is where AI are blind as bats.
 
Geoff Crammond did it best with Grand Prix 4, some 17 years ago. :inlove:

I hate banging that drum, but in several respects that is so true. The AI provided an awesome race, it was like watching a real race. Also, another thing that was mentioned here was it had built in telemetry, at least in GP2. I used to use it to set ride height, but I was young and didn't know better then. Now it would be cool.

The caveat is its only one type of racing, one particular car. AC and PCars are hamstrung by their variety. But I'd also see smaller games that are more focused. Build an engine, leverage that engine across several "packs" (like AMS did) that offer complete experiences including well-tuned and calibrated AI for the cars and tracks. How much better would pC2 had been if it weren't a giant unoptimized mess. I would have taken 75% of the content had it been more cohesive.
 
I rather the AI be used for vehicle tuning first. I want a virtual "team" behind me as a driver.

One one hand the sim is trying to replicate real life. On the other hand, I'm the driver, mechanic, engineer and strategist for a GT3 team which is absolutely not close to reality at all. Those functions should be AI driven and combined with audio feedback as a driver would give. "I have understeer in T1 and oversteer in T14" etc.
If you can still find a copy of it, Setup Developer was made back in the rF1 days and essentially acts as an active guide towards better setup makings. Tell the software that the car is doing and it will give you a solution as to how to make it handle better. It was made by/for Race2Play, but good luck finding a working download link ever since R2P itself shut down a few years ago.
 
Don't need neural AI but the research is nice.

You just need a talented driver to set a handful of demo laps, and some good base AI code.

Recording my own AI lines for NASCAR 2003 Season has pumped a ton of new life into the game on tech that would now be considered archaic.
 
Great article! I would love to have Neural AI for Assist where the AI would teach / correct me in real-time and kick me in the butt when am not getting the best lines and fastest laps :)
 
Last edited:
I rather the AI be used for vehicle tuning first. I want a virtual "team" behind me as a driver.

One one hand the sim is trying to replicate real life. On the other hand, I'm the driver, mechanic, engineer and strategist for a GT3 team which is absolutely not close to reality at all. Those functions should be AI driven and combined with audio feedback as a driver would give. "I have understeer in T1 and oversteer in T14" etc.
Indeed, it would be really interesting to feel as a driver working with an "intelligent" team behind. And it's not that impossible. There was a title which had this feature : total immersion racing (which also included the best fictional tracks ever in a game and unfortunately never remade in our beloved sims). The engineer was adjusting your car each lap during the practice time. It was a slow process but it was quiet amazing at this time.

Any serious sim should provide this feature and allow the player to use it or not, to use it and to be able to manually override some parameters. Any player would get his own sim experience.

Total immersion racing had also, in theory because I don't remember if it really worked, an AI which could remember the player behaviour of the mlayer during the championship, i.e. the more agressive you were with a specific opponent, the more agressive this one would become.
 
I think I understand what you're trying to say in this article, but it seems clear you don't know much about the subject or the technology. Also, if you change the picture and the end of the article to another sim, such as rFactor or PCars I might be inclined to take RD a little more seriously.
 
Don't need neural AI but the research is nice.

You just need a talented driver to set a handful of demo laps, and some good base AI code.

Recording my own AI lines for NASCAR 2003 Season has pumped a ton of new life into the game on tech that would now be considered archaic.

Indeed it's a good way to set a basic lap, an average reference. The issue is how the AI is going to react when something unexpected occurs, which was not included in the driving line.

This neural AI thing is interesting, but I'm wondering how this AI will stop learning? If it becomes too effective, it isn't any better than the current AI which makes mistakes, sometimes realistically, sometimes not.

Will this AI learn only during the development of the sim or will it be able to learn while running with the final consumers (and so becomes too good)? If its learning process is stopped after the development, it means it will never be able to handle the situations it hasn't met before. Which means a buggy AI...

How will this perfect AI be able to choose to make bad decisions when in a situation it has already learned before? A too perfect AI won't be a better thing than an old AI on rails.

I understand the emergency to develop this technology for the championship (current or in preparation, I don't know) with only cars driven by AIs, which is the competition leading to the future of consumers cars, but for a game, I'm not sure the cost of developing such an AI, and developping programs to limit it to make it more human, is anything good as an investment and as an enjoyable experience for the player.
 

Latest News

What's needed for simracing in 2024?

  • More games, period

  • Better graphics/visuals

  • Advanced physics and handling

  • More cars and tracks

  • AI improvements

  • AI engineering

  • Cross-platform play

  • New game Modes

  • Other, post your idea


Results are only viewable after voting.
Back
Top