fix for bug in btRigidBody::appplyImpulse() #272
Conversation
|
As far as I know, the effect of an impulse on the linear velocity of the center of mass is the same, no matter where on the object you apply it. Can you provide a reference or derivation of the true linear component? |
|
Sigh... yeah, you're right. Thinking it over again this morning has made me realize that this happens to be true for the special case I was considering, but is not true in general. It is odd that a unit impulse applied at an offset actually introduces more energy into the system (changing both linear and angular velocity) than one unit applied at the center of mass (changing only linear velocity by the same amount), but I guess the paradox is resolved when the responsibility of conserving of energy is transferred to whatever code computed the impulse in the first place. |
|
Yes, it is counter intuitive, as described here: https://www.physicsforums.com/threads/apparent-paradox-with-angular-and-linear-motion.623383/ |
|
You're right. In fact, that is the exact error I made in my calculations, which is why I had to apply the impulse along the direction of the relative offset vector. This morning I've redone the math, modified my matrix, and reverted to the correct applyImpulse() implementation and things are correctly stable. I'll try to find out what it takes to fix the wikipedia article. |
|
I fixed the wikipedia article. Interestingly, the reference at the bottom of the article (whose images were copied for the wikipedia illustrations) had the correct equations. |
There is an inaccuracy in btRigidBody::applyImpulse() that causes an instability in some experimental custom constraints I've been working on. This is a proposed fix.
When an impulse is applied at some relative offset from the RigidBody's center of mass it should be split into its linear and angular components. The bugged implementation of applyImpulse() does not compute the true linear component and instead applies the full impulse at the body's center of mass. The fixed implementation applies only the component of impulse that is parallel to the offset vector.