Optimal scale for simulation?

[Alic]

Hey, I've been reading over the docs and examples for the past day. This looks like a really great creation.

I just tested out adding BEPU into my game, because my collision detection system has been getting on my nerves, and it worked great. I created a very large kinematic box for the floor, and created dynamic boxes for my creatures. Then, I started trying to get jumping to work properly, and I had to turn the directional Momentum I was adding on jump up to around 5000, with gravity at (0, -200, 0), in order to get the arc of the jump to be satisfyingly video-gamey. I definitely haven't played around with the settings enough yet, but since I'm going to bed, I thought I'd ask if having the gravity set so high, with momentum being applied in such big numbers, would be a less-than-ideal way to use the engine? And, if so, what range of settings performs the best with BEPU?

Thanks for the great physics engine!

(p.s., I was using linear dampening set at .85 on my creatures. Tomorrow, I'll try to turn up the friction on the floor and see if I can turn the linear dampening down.)

[Norbo]

I thought I'd ask if having the gravity set so high, with momentum being applied in such big numbers, would be a less-than-ideal way to use the engine? And, if so, what range of settings performs the best with BEPU?

It could indeed cause problems. Various collision detection systems do not take kindly to excessively sized objects and can result in numerical problems. This is partially because of the limits of single precision floating point numbers. Some pairs of objects, like box-box, box-sphere, and sphere-sphere, have special case collision detection systems which handle larger sizes much more robustly. If only those types were used, large sizes probably wouldn't cause any issue.

However, if you find continuous collision detection to be necessary, using regular scales would be advisable. Its quality will suffer from excessive size in almost all pairs, though it doesn't need to be as precise as regular collision detection.

Generally, sticking around 0.5 to 10 units per dimension per collidable object is a good rule of thumb. This applies to individual convex objects as well to individual triangles within a mesh. Going outside of this range a bit is usually just fine as long as it's still on the same order of magnitude, but as things get bigger, you will begin to see more numerical hiccups.

Thanks for the great physics engine!

Your welcome and thanks