The Last Garage Hands-On: 1000Hz Physics Clarified, AI and Online Discussed

The Last Garage Hands-On - 1000Hz Physics Clarified, AI and Online Discussed RD.jpg
At the recent Sim Formula Europe event in the Netherlands, OverTake was able to go hands-on with The Last Garage sim racing project and find out further details from its creator.

Marcel Offermans caused a stir recently by announcing a brand-new sim racing platform from his startup – The Last Garage.

Indefatigable in the quest for driving physics perfection, the sim racing veteran not only has a working prototype of his latest project, but some intriguing ideas about the future of online competitions. Then there’s the debate as to how the end consumer, i.e. us, will be able to one day experience it at home.


See direct capture gameplay and hear our initial The Last Garage hands-on impressions by watching our Sim Formula Europe event recap above.

Initially touted as a platform for third-party companies to develop upon, the affable Dutch developer seemingly softened during the Sim Formula Europe event. In our experience, visitor reactions were overwhelmingly positive following brief hands-on sessions in a light, rear-wheel drive, car.

“It hasn’t been decided yet, but I would like to do both,” explains Offermans to RaceDepartment when quizzed about its final form.

“I have lots of ideas about how to build a full simulator out of this myself, probably too many ideas, so I probably need to scratch a few of those.

“But I’m also open to others licencing the technology and using it in their projects, whether it is B2B or B2C. I don’t even mind people competing with me using this engine, because I think there can be many different things you can do with it, and I already know I can’t do all of them.”

The Last Garage simulator, classic single-seater
The Last Garage at Sim Formula Europe 2023 – classic single-seater. Image: The Last Garage

1000Hz Physics, All The Time​

Tantalising. Especially considering it feels so telepathic to drive like the venerable Leo Bodnar wheel base used to showcase the technology was somehow affixed to the end of our arms.

Detailed – almost every surface imperfection around the undulation Bridgehampton test venue was relayed back to us – yet also naturalistic. Kicking the rear out of the Escort-like car upon corner exit to create an exuberant slide was satisfyingly straightforward.

There was also a classic formula-style car (depicted above as a 3D render in Blender by Mauricio Leiva) which delivered an intense sense of speed. We even grabbed a little air over some of the ‘yumps’.

During the initial reveal last week, one of the bullet point claims was a physics system running at 1000Hz which caused a debate.

The Last Garage classic car interior, Sim Formula Europe 2024
The Last Garage classic car interior. Image: The Last Garage

Offermans clarifies with alacrity: “The whole physics engine is running at 1000Hz all the time, that includes all the vehicle dynamics, all the tyre modelling, etc.

“All those calculations are done within one millisecond and that goes for all the cars that will be in the sim.

“The smallest step you take, the more detailed samples you can take off the road and if you’re driving at speed, even within one millisecond, you move quite a considerable step.

“So, the smaller steps you take, the more detail you get when driving over kerbstones and other rough surfaces, and that feeling that helps you drive the car better. The smaller you can go there, the better it will be for the feeling.

“That’s why we go at 1000Hz. If we could go 2000Hz, I might even do that, but that’s not feasible with today’s hardware.”

While the baseline technology has been newly created – with the Godot 4 platform in use for the graphics and sound – elements of the physics engine are based on an open-source project called Project Chrono, which The Last Garage’s lead is actively participating in the development of.

Ultimately, there is a modular framework, and if you really would like to dig further into the details and individual components, you can.

“My primary concern was to first get the physics right for this new engine, because if you don’t get that right, it doesn’t matter what else you do, you’re never going to make a good game out of it,” enthuses the former Managing Director of rFactor 2 developers Studio 397.

Marcel Offermans testing The Last Garage, sim racing, Sim Formula Europe 2024
Marcel Offermans testing The Last Garage, sim racing, Sim Formula Europe 2024. Image: RaceDepartment

‘Robust’ Online Multiplayer​

As the diminutive team is in the relatively nascent stages of development, the quantity of content included isn’t a primary concern currently. Now is the time to create the engine, dial in the physics and gather feedback.

When the question of mods is raised, the Netherlands native is open to their potential, but only if “a way that avoids issues with unlicenced and ripped content” is found.

We are intrigued about elements such as AI performance and online connectivity, the latter an area where the Luminis Technologies alumnus has a unique perspective:

“I want to make sure that the online play becomes more robust.

“[If you have] an architecture where everybody connects to a single server, if something happens to that server or the connection, you are immediately in trouble. That’s not something I think that’s acceptable in the modern world where we have plenty of bandwidth and connectivity.

“I want to make a more robust system where you can have a redundancy. If one server fails, everybody just moves to the other one without disrupting the race in any way – it should be seamless.”

The demo we tested, in theory, is already compatible with LAN multiplayer, with the online components in active development. AI should follow later, explains Offermans:

“Right now, I don’t have any AI yet in the sim. I must start somewhere and [multiplayer] is where I like to start, I think that makes sense.

“AI is always a little bit difficult because if you have such an advanced physics system, it becomes a little bit harder to create 40 or 50 AI cars that use that same system. That’s way too heavy. You almost always see AI cars using simpler physics, so you can have more of them running on the same machine.

“That is an area where I’m working with a few universities to research simplification techniques for vehicle models so I could create, let’s say 100, and comfortably run them with those models being almost identical to the player car. We want to get as close to that as you can.”

Marcel Offermans at Sim Formula Europe 2024
Marcel Offermans at Sim Formula Europe 2024. Image: RaceDepartment

The Year Ahead​

Following 18 months of working in the shadows, the Dutch Sim Formula Europe event in Maastricht was an opportunity to see if others agreed with the platform’s direction – something pivotal to Offerman’s development process:

“At some point, you start wondering, ‘where am I with this stuff?’ and ‘what do people think about it?’, so I wanted to make sure that at some point I went to an event and got it as much feedback as I could.

“I’ve made a little list already of things that that I should still work on. One common element I’m hearing, especially in the formula-style car, is that the sound is not great, with too much tyre noise.

“So that’s something I need to fine tune and there are a few other minor things that I’ve noted down that we’ll work on the next couple of weeks.

“I think, add another year, and I will have more details in the physics engine like dynamic roads, tyre, wear, tyre temperatures… those will be implemented, for sure.”

To see direct-capture gameplay footage of The Last Garage, watch OverTake’s roundup of the recent Sim Formula Europe event. Let us know any questions you’d like to ask Marcel Offermans in the comments below, and we’ll keep you updated over the coming months.

You can also submit suggestions
in our The Last Garage forum
About author
Thomas Harrison-Lord
A freelance sim racing, motorsport and automotive journalist. Credits include Autosport Magazine, Motorsport.com, RaceDepartment, OverTake, Traxion and TheSixthAxis.

Comments

If your talking about the player benefiting from thousandths/hundredths or even tenths of a second here, then it really matters only to those sim drivers that compete at a higher level, most sim drivers have trouble trying to string together half a dozen laps within a second or two, many of us 'compete' at a level where it doesn't matter that much.
Just go to an online AC practice at Nords and see how many Lamborghini's are buried in a fence as you drive past, or,find a server where 20 odd cars are racing and see just a couple doing well an the rest are sometimes 'laps' down.
I'm happy if I do 5 laps without some sort of major excursion into the kitty litter!
 
FFB doesn't matter/AI doesn't matter/Multiplayer doesn't matter/graphics dont matter/Sound doesn't matter/VR doesn't matter/Mod support doesn't matter

If we get a good vibrator that vibrates with the game's actions
 

Crawdad97:​

Yes it does matter. I don't have time to wait on games in a lobby. I have 3 kids and finding time to play is difficult. I would love a great AI within a game. Not everyone wants to play online. To be able to pause a game to take care of an issue with children is important.​


True. I'm 40 and I have 1-2hours of free time every couple of days. Racing with AI is something simracing titles are really bad at. In every sim I tried, Max AI, be it ACC, AC, AMS2,... After 1h of practice, I will be 2-4 second faster, they all have "one corner problem", where they lose a second there.

I was never simracing on consoles. But if GT introduces Sophy so that you can run any custom combination, and they are as fast as me, I can see myself switching.

So that's one of the reason why iRacing is so popular. You log in, and you can race *anytime*.

I wish Kunos would hire some students from university, and create proper ML AI. That can be their phd research project. It would be win:win:win situation. Students would learn, their company would have something special, and consumers would be happy. I was laughed out when I proposed this. They said that their AI is at the edge of what's possible. Sometimes, one corner problem can be fixed by only changing their line. But they didn't bother.

AI is an afterthought in sim racing currently. And I really wish it would change.
 

Crawdad97:​



True. I'm 40 and I have 1-2hours of free time every couple of days. Racing with AI is something simracing titles are really bad at. In every sim I tried, Max AI, be it ACC, AC, AMS2,... After 1h of practice, I will be 2-4 second faster, they all have "one corner problem", where they lose a second there.

I was never simracing on consoles. But if GT introduces Sophy so that you can run any custom combination, and they are as fast as me, I can see myself switching.

So that's one of the reason why iRacing is so popular. You log in, and you can race *anytime*.

I wish Kunos would hire some students from university, and create proper ML AI. That can be their phd research project. It would be win:win:win situation. Students would learn, their company would have something special, and consumers would be happy. I was laughed out when I proposed this. They said that their AI is at the edge of what's possible. Sometimes, one corner problem can be fixed by only changing their line. But they didn't bother.

AI is an afterthought in sim racing currently. And I really wish it would change.
It isnt an afterthought, Marcel brought it up earlier the decisions that have to be made. It takes a considerable amount of resources to get your "dream AI"

Something like Sophy uses a cloud infrastructure as well as beefy hardware. Even the PS5 can only handle what 4-5 cars running Sophy?

You got laughed at because its just not feasible for this market... but its not a dumb idea.
 
For anyone who says "AI doesn't matter", obviously it doesn't matter TO YOU and that's fine, but consider this: even iRacing, the king of online multiplayer racing sims, now has AI. And even crazier is that supposedly it's the among the best out there (I don't know, I haven't tried it). So, if even iRacing puts time and resources into developing AI then I'm gonna go out on a limb and say it matters.

I feel like this AI vs Multiplayer argument, like somehow it's not valid for both to exist in the same sim / game is like saying we like listening to the same band, but your favorite member is the singer and I'm into the guitarist, so one of us has to be "wrong". Call me delusional (as I'm sure someone will), but it's entirely reasonable to have different parts of your audience/customer base buy into your product for different reasons - OR completely ignore you if what they're interested in isn't there.

Anyway, to at least be partially on-topic, I wish Marcel and his team the best on this project and hope that it evolves into something really special. The sim racing community is a tough crowd, and rivaling and/or hopefully exceeding the established players in this space will be no small task.
 
No AI = No Buy!

Seriously though it's good to see Marcel continuing on in the landscape of a harsh niche simulator culture, of course away from us goons sat at home in our pyjamas pretending to be Senna there is another world of potential private interest and as the news item suggests the possibility to license out the work.

Another one to add to the "Follow List", good to have you back Marcel I wish you all the best with this project :)
 
Premium
Thanks @MarcG, as you know it's still early days (I have not even defined, announced or named a product yet). I wanted to get feedback from the community early and I'm hearing strong sentiments for AI, multiplayer without having to wait in lobbies, triplescreen and VR support and many more things so I have my work cut out. But I'm not in a rush, I want to make sure I get it right.
 
Premium

Crawdad97:​



True. I'm 40 and I have 1-2hours of free time every couple of days. Racing with AI is something simracing titles are really bad at. In every sim I tried, Max AI, be it ACC, AC, AMS2,... After 1h of practice, I will be 2-4 second faster, they all have "one corner problem", where they lose a second there.

I was never simracing on consoles. But if GT introduces Sophy so that you can run any custom combination, and they are as fast as me, I can see myself switching.

So that's one of the reason why iRacing is so popular. You log in, and you can race *anytime*.

I wish Kunos would hire some students from university, and create proper ML AI. That can be their phd research project. It would be win:win:win situation. Students would learn, their company would have something special, and consumers would be happy. I was laughed out when I proposed this. They said that their AI is at the edge of what's possible. Sometimes, one corner problem can be fixed by only changing their line. But they didn't bother.

AI is an afterthought in sim racing currently. And I really wish it would change.
rFactor 2's AI now is quite good, although for reasons that only rF2 itself would be able to explain if it were a person, the exact same combo and conditions, in some PCs (even very good ones) seem to produce wildly different results, with 100% murderous AI.

But those 2/4 seconds faster than the AI won't happen in many quality combos (even with mods).
If you join S397's Discord and/or the Forum, you should be able to get enough information from other single player focused users and have a very good racing against the AI experience.
 
Premium
Thanks @MarcG, as you know it's still early days (I have not even defined, announced or named a product yet). I wanted to get feedback from the community early and I'm hearing strong sentiments for AI, multiplayer without having to wait in lobbies, triplescreen and VR support and many more things so I have my work cut out. But I'm not in a rush, I want to make sure I get it right.
You can rest assured that rabid criticism and wild manifestations of insatisfaction are guaranteed ...
 
Thanks @MarcG, as you know it's still early days (I have not even defined, announced or named a product yet). I wanted to get feedback from the community early and I'm hearing strong sentiments for AI, multiplayer without having to wait in lobbies, triplescreen and VR support and many more things so I have my work cut out. But I'm not in a rush, I want to make sure I get it right.
Yep they're all the standard items I'd want to see in a racing sim by default, what I would like to see is a Dev (any dev not just you) take some of these things to the next level. So you've already talked about moving to a second server almost seamlessly if the first server packs up, so in essence with that the next level for the two of those items I'd say is;
  • AI - Neural or whatever the next big thing is in creating an AI that reacts like a human, not a "bot" like in every other racing sim. RaceRoom have touched on the Adaptive Learning, expand on that, take AI learning "on the fly" properly, have them learn a track through practice, wet conditions, become enemies with another driver, give then human attributes that are dynamic and not locked in a .txt file. AI by parameters has been done to death and still other Devs are struggling to get them acting like a human in many scenarios.
  • VR - Cockpit of course but go beyond that, allow the player to walk around the car, open the doors, open the bonnet, wander down pit lane and check out the competition, watch from the stands, have a pit garage with screens of data to walk up to and read, walk into a room for the Race Stewards Meeting in an online lobby etc etc
Next level stuff, I know I dream but if you look at racing sims today and what's coming out this year they're all practically following the same design from 20+ years ago. This hobby doesn't necessarily need a dramatic change, but it could do with a Dev going to that Next Level or Next Gen as the kids call it these days! No pressure :D
 
Premium

Crawdad97:​



True. I'm 40 and I have 1-2hours of free time every couple of days. Racing with AI is something simracing titles are really bad at. In every sim I tried, Max AI, be it ACC, AC, AMS2,... After 1h of practice, I will be 2-4 second faster, they all have "one corner problem", where they lose a second there.

I was never simracing on consoles. But if GT introduces Sophy so that you can run any custom combination, and they are as fast as me, I can see myself switching.

So that's one of the reason why iRacing is so popular. You log in, and you can race *anytime*.

I wish Kunos would hire some students from university, and create proper ML AI. That can be their phd research project. It would be win:win:win situation. Students would learn, their company would have something special, and consumers would be happy. I was laughed out when I proposed this. They said that their AI is at the edge of what's possible. Sometimes, one corner problem can be fixed by only changing their line. But they didn't bother.

AI is an afterthought in sim racing currently. And I really wish it would change.
It's not an afterthought, it's just not easy. If you see what's happening in the AMS2 beta forums, a lot of the updates/testing is around AI. They are putting a lot of work into it, and it is getting better, but its a very time/resource consuming task.
 
Premium
If the goal is to have AI which behave like humans, i mean like on LFM for example, it may be better to do without it.

The problem with humans is , that there is everyday more of them behaving like selfish /crazy / doom/ persons.

We need an AI which behaves like "normal" and reasonable drivers just hoping to not die during the race !

It is then not perfect, but a lot more enjoyable in the end for those not having the possibility to race only against trusted friends online
 
Last edited:
For me Ai needs to be "aware" more of the player, and react to the player in a realistic way. But I presume that computationally that is resource intensive.
 
Premium
More realism, more illusion, more of everything.
Personally, it seems more and more like the smartphone market.
There can no longer be real innovations, so we tinker with the microscopic details in nice marketing language. Look here, you believers in the telephone.
We've refined our case curves by 0.00000001 nanomillimeters and our display can now achieve 64K resolution for an even finer image.
This is about software that hardly costs more than what a 14 year old gets in pocket money per month. I don't know why the big racing teams in Formula 1 use simulators that cost several million euros when entertainment software can supposedly do it much better.
Likewise the armed forces, NASA, etc.
Why do they all buy such expensive equipment?
I think most people have forgotten what it's actually about.
Have fun.
Every little thing is pushed as the absolute superlative.
Haven't you all had any fun so far?
Do you think you'll have more fun if you jump on every "improvement" pushed by marketing strategy?
A home computer with software for 50 euros plus a steering wheel and all that stuff will never come close to the performance of a professionally used simulator.
Even the simulator for the driving students at the driving school here is far superior to any pocket money simulation. The cost for this was just 500,000 euros.
But in the Hi-Fi sector, cables are also responsible for the sound of the speakers and not the combination of recording, amplifier and speakers.
Learn to race again just for fun.
I'm sorry, but that's how I see it
 

Latest News

Article information

Author
Thomas Harrison-Lord
Article read time
6 min read
Views
7,915
Comments
80
Last update

To join the OverTake Racing Club races I want them to be: (multiple choice)

  • Free to access

    Votes: 93 91.2%
  • Better structured events

    Votes: 16 15.7%
  • Better structured racing club forum

    Votes: 16 15.7%
  • More use of default game content

    Votes: 11 10.8%
  • More use of fixed setups

    Votes: 28 27.5%
  • No 3rd party registration pages

    Votes: 34 33.3%
  • Less casual events

    Votes: 9 8.8%
  • More casual events

    Votes: 33 32.4%
  • Other, specify in thread

    Votes: 5 4.9%
Back
Top