Driving Demo (2025-07-06)Link

Try the Driving Demo for yourself.

Strangely, I feel like I have very little to say about the driving model I've been developing. It's my obsession, I can't turn myself off from it, and I want to share and shout about it, but there's no singular element that I really want to crow about. It's at a state I'm relatively pleased with, doing things fairly stably. I'm not blown away by how it feels, it doesn't quite return to centre as quickly as I'd like so the car tends to wander just a fraction more than I'd want, and it's not like even after all the time driving it I can place it wherever I want. But the generic test car is fun enough to throw around, predictable, and doesn't do anything stupid. I assume if I was working on some of my favourite games, I'd pick holes in the physics all day. Just how I am, without a defined target, everything is a perpetual improvement project.

I'm happy with some of the lessons about car development I've discovered. The car is far away from the limit, there's easily more lap time in it from more brake torque and a bias tweak, but that also makes it harder to drive. Brake bias should be set further forward than you'd expect, because when you start cornering, the front inside locks instantly, and then the rear inside follows if it's at the limit of braking. This also means ABS, or some way to control each individual tyre's braking, is a huge advantage in combined braking that I never fully appreciated until now. Lots of diff locking in decel is similarly advantageous.

There's also a lot to learn about discrete numerical simulations. I've learned to hate all the bugs that get introduced from timesteps. Every fraction of a second is time for two variables to jump past each other when you'd want them to meet. Not to bring up braking again, but that was a fun problem to solve. "Apply a torque opposing the wheel direction" sounds great until you realise you've got to account for when the wheel is nearly stopped and suddenly has all this torque appled that wants to launch it way past stationary. There was a similar issue with the clutch, which at the time of writing (or V0.0.11), has a sort of hack that I'm fairly certain is indistinguishable from how it should work, but I'd like to investigate more to be sure.

My next future plan is to introduce a better magic formula. In theory, Pacejka answers everything I could need, but I like the idea of just setting certain values for peak slip or the initial slope instead of having to run some loop to find that information. I already have a basic system, but the initial slope of that is fixed and I'm hoping to create yet more variety. The setup I'm envisaging uses an initial slope as a 0-1 variable for if it goes vertical or diagonally to the peak, some way of controlling what percentage of the peak is linear, and then some way of drawing a curve from the end of the linear region to the peak. Post peak matters less to me, I'll probably keep the sin function I'm already using for that. I don't care about copying tyre data or replicating curves I've seen elsewhere, I want some predictable modifiers I can use to try different kinds of setups.

That's enough rambling. Hopefully you enjoy the demo!