right, I can''t take credits for the idea, it came from a friend who was messing with it, but I''ve implemented it for fun, and wayhay! pretty sweet. another verlet physics demo basically, here is the general concept. It''s a mixture of rigid body physics and verlet particles (yes, another one). But wait, it''s actually a lot lot faster and simpler than the concept of generating hundreds of constraints withing a body. right, first of all, each body has an tetrahedron structure inside, which is basically a verlet particle system (4 particles, 6 constraints). That tetrahedron is used for two things. defining the general orientation and position of the body, and also generating the actual body movement (rotation and translation). Imagine you make the tetrahedron particles as collidable, then the tetrahedron will bounce all other the place happily, like any verlet paricle system. now, the trick is to use the tetrahedron only has a frame to support the rigid body, and use the rigid body''s collision mesh to influence the frame, so the frame rotates and moves upon impact. this is achieved very simply. When a collision between the mesh and the world is detected, you generate a temporary set of constraints from that colliding point, linking the point to the frame (so it''s 4 constraints). What these constraints will do, is push the frame out of the way, effectively rotating and translating the frame. the nice thing is, you can use any types of collision mesh, and you solve constraints only on the tetrahedron (6 per body), and on the temporary collision points (4 extra constraints per point). so in the end, the number of constraints to solve is minimal. In the example, I used a 800 vertices mesh (a bunch of dolphins), and another pretty complex mesh, bouncing on the terrain. The green lines represent the collision constraints, and you can see that there is only a handful of them at a given time. The red thing is the tetrahedron frame, supporting the rigid body. if you can''t download it head straight to my homepage the trickiest part is in the collision detection, which incidently can be done on a per-triangle basis. My next try will be to use something like Rapid (OBB and tri mesh), and try collisions with that. Since I don''t need that much informations (like the normal of collision), and only solve intersection between triangle, that would be quite interesting. the next step, joints. It''s got simple verlet constraints, plus some useful informations, like the actual orientations of the bodies, so very useful for angular constraints.

keys are a, w, s, d capslock, tab for camera movement, and ''r'' to restart.

This idea is already mentioned in the original Thomass Jakobsens paper.



[edited by - bouncer on March 28, 2004 10:27:21 PM]

quote:Original post by Bouncer
This idea is already mentioned in the original Thomass Jakobsens paper.

right, yeah. I re-read the paper, past the point I was mostly interested in before. Anyhow, that''s an implementation of it. And I think it''s worth a demo and further investigations.

about the terrain bug, I''m aware of that. I don''t think I''m doing bound checks on the terrain coordinates.

Good work... I only just discovered you homepage and all the demos posted there.

I just can''t download the demo... I can download others though.

lycos lycos lycos..... Gotta change my provider. any recommends a stable web hosting? no BS, good space (preferably 50 mb, like lycos), and obviously free? Just need something where people can download from with no freaking hassle!