Movie introduction of dVC3d (using Bullet interface).

[ngbinh]

I've been working on dVC3d ( http://cgi2.cs.rpi.edu/~nguyeb2/doku.php?id=dvc3d ) for a year or so.

The movie: http://www.youtube.com/watch?v=Yh5qKwK8FcQ
( It's best to view in 1080p , especially if you want to see what happen inside Blender)

Short introduction:

dVC3d is intended for scientific/robotics/machine design simulations as it strongly focuses on accuracy. It's backed by strong convergence (to underlying physical models) theories and verified in some difficult (to simulate) experiments (look at my publications if you care) .

Currently, dVC3d uses (slightly modified) Bullet (2.77) for collision detection and programmable interface. So basically, you can throw dVC3d in anywhere Bullet works.

Some (maybe subjective) pros/cons:
Pros:
+ Accurate to (inelastic) rigid body, Coulomb friction model.
+ Deterministic ( does not rely on rand(), so results are reproducible )
+ Stable
+ Bullet's API. dVC3d ran on all Bullet's demos.
Cons:
+ Slow ( 0(n^3) instead of O(n) ). But usually fast enough (real time) on simple scene. All the scene in my movie was run in real time.
+ Rely on a closed-source but free solver ( PATH http://www.cs.wisc.edu/cpnet/cpnetsoftware/ ). But I'm working on a new solver that should be faster ( could be comparable to Bullet ) and will be open source.

A little FAQs:
* Will dVC3d replace Bullet?
No, dVC3d has different objectives. If you are happy with Bullet/ODE/PhysX/Havok/... then there is no reason to switch.

* Then why spend your time on dVC3d?
I believe there are intended audiences for dVC3d. Anyone who want to simulate a small scene and require high accuracy and stability should find dVC3d useful. We've been using dVC3d on some projects ( my publications if you care). One excellent candidate for dVC3d is robotics grasping simulation ( watch the movie to see for yourself ).

Also, dVC3d is our (RPI Robotics group) attempt to study different aspects of physical simulation. Some of the recent results are a new method that could handle non-convex free space and use a completely different collision detection ( simple and parallel friendly) and new Newton-based solver (again, my recent publications).

At its current form, I doubt gaming/graphics community find dVC3d attractive but the two results mentioned previously could make dVC3d interesting ( fast, simple and highly parallel).

Binh Nguyen.

[Dirk Gregorius]

Nice work. Well done!

[bone]

Looks good, and I'm sure we're all interested in hearing more about your new solver!