The 3DOF motor does seem simpler and makes more sense, and it does exactly what's needed (drive the relative transform of two objects physically rather than kinematically). Damn it!
The application I was thinking of specifically was how to unify rigid and deformable bodies; in the former case the constraint geometry (joint anchors, contact points) is fixed/constant -- during constraint solving only the body transforms are changed. If instead you let the solver modify both the body transforms *and* the constraint geometry, this gives you a way to model deformable objects. Basically, instead of the perfectly rigid anchor point transmitting all of the impulse to the rigid frame in which it's embedded, part of the impulse is "absorbed" (deforming/moving the anchor point's local position) and the remainder is felt as a rigid impulse.
I actually have this "frame in a frame in a frame" system working, but as you've pointed out it's sort of uselessly complicated.. unfortunately I haven't really figured out how else to model rigid and deformable (esp. plastic deformation) objects with the same model unless you take such an approach. Simulating constraint geometry in local space lets you add this extra layer of degrees-of-freedom to the model; rigid shapes are simply those whose constraint geometry is infinitely massive/doesn't move wrt the frame in which it's embedded.
Possibly it's just a mistake to consider deformable objects to have "a" rigid frame in which their geometry is embedded, and simulating everything in worldspace (i.e as is usual; the shape is just a bunch of linked particles) is the better model. I was basing this off of the shape matching papers where an undeformed frame is extracted from the current configuration in order to measure the current deformation -- I figured "why not just simulate the undeformed frame explicitly".. sigh.
Login
Register
FAQ
Search
