It’s not the first inverse mouse-ball drive system.

They use an OTS IMU, which grealty simplifies the process.

The controller might as well be the subject of an undergraduate student thesis project or MAYBE a homework assignment for a masters level controls class. It’s a simple LQR loop with a small inner experimentally-tuned PI loop to correct for frictional effects. Of course, input dynamics have been omitted, and that’s one of the major reasons why this controller works well to stabilize the ballbot on most surfaces but can only follow straight lines on carpet.

This does not deserve the attention it’s getting. It’s a 2D inverted pendulum. People need to get over it.