Reaction-Wheel Cube

A cube that holds itself upright on a single edge using reaction wheels on three orthogonal axes — the same momentum-exchange principle a spacecraft uses to point itself, running against gravity on a desk. Accelerate a flywheel one direction and the frame is pushed the other, and if you can do it fast enough and precisely enough you catch the cube faster than gravity can tip it.

The structure is a pre-tensioned aluminum space frame on CNC-machined corner blocks, because the wheels need something stiff to react against — any flex in the frame shows up as lost control authority. The flywheels are brass, sized in CAD to hit a target inertia rather than picked and hoped for. Drive is brushless with field-oriented control, and a Teensy 4.1 closes the loop against an IMU.

Before committing to hardware I wrote an RK4 plant simulation and a sensitivity-sweep tool in Python, to find out which parameters actually matter. Two do: wheel torque and cube mass. Everything else is noise by comparison. The sweep also put a number on latency — every 10 ms of delay in the control loop costs about 10 % of the recovery envelope, which is the margin you have to catch the cube before it is past saving. That is the argument for the next step.

Which is a custom board. Three off-the-shelf driver modules are doing the work now, and replacing them with one motor-control and flight-computer PCB removes the latency the sweep says is expensive. The FOC driver stages, inline current sensing, SPI encoders and a Wi-Fi telemetry link are selected; schematic capture is next.