I do scaling of the reaction force because I don't use Pacejka. I don't understand why you need to divide your reaction force by the load if you use Pacejka. With "magic" formula you are suppose to get and exact amount of reaction force specifically for a specific wheel with a specific load on wheel.

I'll be adding description of friction model I use for the tank into documentation, can drop a link when it done. Otherwise I'm not sure how I can help here, I mean you can't just randomly change minus to plus in your formulas and expect to get some proper result. It doesn't work like that.

You told that you divide it. Well, if I don't divide it, it goes even more crazy. The wheels seems to be turning fast in random ways. Nothing works as expected. One of my problems is actually understanding where to use minus, but where not to use it...

I divide it by total mass of the vehicle, as my friction is calculated out of the mass of the vehicle, relative velocity and number of contact points with the ground. Load on specific point has nothing to do with it. It's just a simple rationale that friction would be proportional to the mass of the object on which it has effect if it's calculated using its mass.

Yeah... But load is the part of the car's mass multiplied by gravity that's located on wheel(at least I thought so). So if I divide it by gravity, I get the mass on wheel. Hm... But pacejka has some squared load functions.

That's has nothing to do with the scaling of reaction force... again, if you are using Pacejka you shouldn't scale reaction force but get your units and variables right. Do it on paper first for few time steps. When wheel rotates with the same linear velocity as car is moving and reaction force is zero, then wheel slows down by 1 radian a second and your reaction force should react to it, eventually accelerating wheel back to the zero relative velocity with the ground.

Hm... And how would I calculate it from reaction torque?

By the way - if wheel is free rolling, than it's propably not gone tend to slow down(except in case of braking or if it's locked with drivetrain which adds some inertia).

I just have no idea how to calculate this. I see that the reaction torque must be propably clamped in this case(if it is bigger than 1 rad/s multiplied by wheel inertia, than it must be 1 rad/s multiplied by wheel inertia). Am I right about it? Is clamping method right at all or are there better ones?

Forget about clamping and yes, the wheel will slow down if you push brakes or if you have some mechanical friction of the transmission which naturally slows done wheels + there is rolling friction which slows down wheels. Regardless. The most important is that moment of inertia doesn't change velocity by itself.
Moment of inertia is angular mass equivalent to a normal mass. It tells you how much body will resist rotation when torque is applied. In this case moment of inertia is used only to calculate angular acceleration and by knowing angular acceleration and time step you can calculate new angular velocity.

1) apply some torque to the wheel by using it in Pacejka formula or by calculating friction
2) from step one you get Reaction force or friction force which you convert into reaction torque
3) now you can calculate new angular velocity as
w += T/I*dt

Just this thing should accelerate your wheels if car is on the slope.

In Pacejka you apply slip ratio which is defined by:

slip_ratio = (w * R - V) / |V|

w = angular velocity

R = radius

V = longitudal velocity of the wheel

In this case you must already calculate angular velocity before applying Pacejka calculation. Than the result is friction force. And to convert in torque it's multiplied by R. However, if the wheel doesn't apply any torque, it will be turning 0 at beggining. And than it gets only turned by this torque. But what if, for example, this torque is very big(e.g. slip ratio is around 6%), while the car is moving at very little speed. It will result in the wheel to be turned at opposite direction to the car's motion(due to resistance) with strong torque. This will make car change it's direction. Although this would slowly become smaller and smaller and tend the slip ratio to become 0, however, it would cause shaking and many angular velocity changes in wheel. And I don't know how to get around this issue.

Bored Engineer, I've got a question - in case of slipping clutch, do I calculate engine delta angular velocity this way

deltaav = (Tengine - Tclutch) / I * deltatime

and than apply to the engine? Or is it done else-way?

Yes, but you don't need a minus sign there. Torque from clutch will already have a proper sign so I would change it to plus.