Thank you very much. The jitter you see when using higher values for gravity is normal. It is a problem inherent with every physics engine, and there is no solution which works fine in every case.
Here you could simple decrease the timesteps used when advancing the simulation time, for example like this:
self.mgr.doPhysics(dt, 20, 1.0/180.0)
The default values are up to 10 timesteps each 1/60 second. Setting the timestep to 1/180 will reduce the jitter significantly.
Then there is damping. The sample doesn’t apply linear and angular damping.
Finally, Bullet offers an experimental method to apply some additional damping, but this method is not exposed so far. I will try to fine a way to do so for the next release.
About the BallInMaze sample: The maze has it’s orientation set explicitly each frame, depending on the mouse position. The ball falling through the floor might happen if within one frame the “floor” moves from below the ball to above the center of the ball. This can be avoided easily, for example by using a box for the floor (instead of a plane), but this would mean that I have to modify the maze.egg file. I don’t want to, since the intention of this sample is to show how to automatically set up much of a scene from “standard” Panda3D collision solids.