I am trying to use Bullet to get physically accurate simulations by setting up the simulations through Blender (2.73a or 2.75a). It seems to be very unreliable with respect to the physics_step_sub property.
Most annoyingly, a collision is often missed, and the object gets stuck. I hope someone spots an obvious mistake.
I set up a minimal working example:
- a sphere (rigid body, collision_bound_type: SPHERE) that's given an initial velocity, and
- a wall (actor, static, BOX), that it bounces back from.
- no gravity
- elasticity 1 for both objects.
The outcome changes based on the substeps parameter. Which would be ok, but which is closest to actual physics? I read, that it's not good to have small objects, so I enlarged the scene, the outcome is similar: tests.zip
The minimal blend file can be found here: scene.blend
I tried adjusting the magic "game.radius" parameter, but it only changes anything, if it's too large and the objects immediately start to distance themselves.
Collision margins don't seem influence this behaviour.