2 docs tagged with "physx"

No physics engine has a native "track" primitive. We share how we modelled a tracked skid-steer robot in Isaac Sim — the failure of a 46-link articulated track, a 4-contact model that turned "perfectly" and was wrong precisely because of it, and the measured 12-contact distributed virtual-wheel model we use today.

We derive why three weeks of friction-coefficient tuning was doomed from the start — in a skid-steer vehicle's turning-ability ratio, both the friction coefficient and the weight cancel out, leaving only contact geometry. With measured data, plus two hard-earned lessons about trusting simulation instrumentation.