It looks very cool.

Physical

Wow, if that works as well as it sounds... impressive!

I wonder if he's using that Jakobsen stuff...

Looks like it is. You build objects from meshes. If that's the case, that would be dog slow for more than fairly simple things. Same reason Gish requires a 1GHz computer.

How do you figure? They used it in Hitman (a 3d game.) Plus the paper is 4 years old. The most expensive part is the square roots, but he includes a fudged sqrt function for that.

Leisure Suit Lurie Wrote:How do you figure? They used it in Hitman (a 3d game.) Plus the paper is 4 years old. The most expensive part is the square roots, but he includes a fudged sqrt function for that.

How do I figure? Because the Hitman game had very few constraints used. And I would guess that it doesn't have very many objects being simulated at once. This Physical thing has you create models that use what looks like hundreds of constraints for even simple models. I'd expect the physics to work fairly poorly for rigid body stuff. As you might notice, in Gish things often get turned inside out, and nothing is really ever rigid.

That's why I gave up on SPK, because objects never really had a set shape I couldn't use fast collision detection methods. You have to deal with segment-segment collisions, and be careful that things didn't get tangled or other weird problems. It works great for ragdolls, rope, and cloth, but it's frustrating for anything else.

I did do a 2D vehicle over a height field using these techniques. (I faked the wheel rotation)

It was plenty fast where I could have done more.

-Jon

That thing looks pretty cool, I would have loved it when I was 8. You could probably make somthing similar using Newton, its a 3D physics engine but its been used for 2D simulations before and handles constraints.

Maybe. I kind of like the simplicity of Jakobsen's approach.

aarku Wrote:I did do a 2D vehicle over a height field using these techniques. (I faked the wheel rotation)

It was plenty fast where I could have done more.

-Jon

The collisions aren't really the problem. The problem is that you have models built using hundreds of constrains that must run several iterations each frame. When you have a few constraints, you can use few iterations to relax them. When you have many, you need to use more iterations to relax a model sufficiently.

And even when you do use lots of iterations, Jakobsen physics work poorly for rigid models. Like I said, awesome for ragdolls, string, and cloth, but not so great for other things.