Odd behavior in box-box collisions in development version
Posted: Tue Aug 23, 2011 11:07 pm
Hey Norbo, you seem to have completely fixed that deactivation issue I was seeing. Haven't seen it since I tried the development version. Thanks 
There's been one odd thing that I've noticed since I upgraded from the 0.16.2 main release, and I thought I'd warn you about it in case you weren't aware. It can be most easily seen when moving a box around another box by sliding diagonally into it.
Imagine two boxes next to each other from a top down view. Both boxes have LocalInertiaTensorInverse = new Matrix3X3(). The box on the right is dynamic. The box on the left is kinematic. They both sit on a large kinematic floor, as in many of the demos.
If you apply a northwesterly force to the box on the right, so that it collides with the box on its left, it then starts slowly sliding northwards while pressing against the other box. As it should. The weird thing happens when the box on the right clears the box on the left -- you would expect it to start moving left. Instead, it seems to think it's still colliding with the box on its left for a fixed amount of time. So, in extreme cases, the dynamic box could be a whole box length north of the kinematic box before it finally starts moving left.
I've considered the possibility that this could be the StaticFriction of the floor keeping the moving box from establishing westward movement. But it really doesn't seem to be the case -- for a set amount of time (probably a bit less than a second, but more than half a second) the moving box literally doesn't move a single pixel to the left (or... west, I guess I should say). It really looks like the kinematic box invisibly extends northwards. But it doesn't -- you can produce the exact same behavior while moving into (and around) any side of the kinematic box.
Might this be a bug that's crept in, or do I just need to tweak some properties of the boxes, most of which are still set up as they were for 0.16.2?
Alex
*edit* By the way, I'm running my simulation at 60 updates per second , and using the most recent development version at the time of this post.
There's been one odd thing that I've noticed since I upgraded from the 0.16.2 main release, and I thought I'd warn you about it in case you weren't aware. It can be most easily seen when moving a box around another box by sliding diagonally into it.
Imagine two boxes next to each other from a top down view. Both boxes have LocalInertiaTensorInverse = new Matrix3X3(). The box on the right is dynamic. The box on the left is kinematic. They both sit on a large kinematic floor, as in many of the demos.
If you apply a northwesterly force to the box on the right, so that it collides with the box on its left, it then starts slowly sliding northwards while pressing against the other box. As it should. The weird thing happens when the box on the right clears the box on the left -- you would expect it to start moving left. Instead, it seems to think it's still colliding with the box on its left for a fixed amount of time. So, in extreme cases, the dynamic box could be a whole box length north of the kinematic box before it finally starts moving left.
I've considered the possibility that this could be the StaticFriction of the floor keeping the moving box from establishing westward movement. But it really doesn't seem to be the case -- for a set amount of time (probably a bit less than a second, but more than half a second) the moving box literally doesn't move a single pixel to the left (or... west, I guess I should say). It really looks like the kinematic box invisibly extends northwards. But it doesn't -- you can produce the exact same behavior while moving into (and around) any side of the kinematic box.
Might this be a bug that's crept in, or do I just need to tweak some properties of the boxes, most of which are still set up as they were for 0.16.2?
Alex
*edit* By the way, I'm running my simulation at 60 updates per second , and using the most recent development version at the time of this post.