Here’s some footage of a car driving around in the Shiva game engine, using the ‘raycast wheels’ approach. I have found that this gives much nicer results than the demo car that comes with Shiva, which treats wheels as constrained spheres. The problem with modelling wheels as spheres rotating on one axis is that this assumes rigid bodies for the wheels; actually, wheels are not rigid.

The basic idea with raycast cars is that you cast a ray from each wheel hub, and calculate the distance at which a surface is met (if at all). This defines the suspension’s extension, and then you simulate spring behaviour by applying increasingly strong forces as the suspension gets more compressed. There’s also some subtleties such as anti-roll and varying spring strength depending on whether its compressing or extending; but these are mostly tweaks applied to the core idea.

I also found that at high speed the default Shiva ‘spherical wheels’ car would suffer from occasional glitches and speed wobbles, rendering it fairly useless for most games. Sometimes it seemed to stagger sideways at high speed, presumably due to errors on the axis constraints allowing the spheres to role sideways. I assume these problems are caused by over-relying on the accuracy of the underlying physical simulation in order to obtain the required behaviour.

In the video clips shown here, the behaviour of the car in all ways other than spring response has been results-driven, i.e. the desired behaviour (skidding, drifting, tendency to roll, etc.) has been considered and this is used as a starting point for programming the calculations – rather than starting from an accurate physical simulation and expecting realistic/fun results as an emergent behaviour.

This represents very good progress towards what I’d like to have for a major new Darkwind project (Darkwind 2?) that I have started thinking about. Really the ineffective default Shiva car is what had stopped me considering this before now.

Last year I also wrote Musclecar Online. In this the car was entirely results-based, having no accurate underlying physical simulation from which the behaviour emerged. Musclecar Online is actually a 2D driving simulation with 3D models (hence, no hills or bridges) – and it was therefore entirely possible to define the behaviour I wanted first and then to write code to directly produce that behaviour.