The ball feeder in Open Tennis Robots aims in two ways. It tilts its launcher head from -15 to +45 degrees, and it turns the head on a turntable, 30 degrees either side. Until today the bill of materials listed "stepper or geared motor for the turntable" with a note that the drive was not modelled. This run designed it and sized it from the CAD model, then checked the turned head for collisions.
The drive
A NEMA 17 stepper hangs flange-up under the mast plate, 124 mm to the side of the turntable's axis. A bought steel pinion on its shaft (module 1.5, 15 teeth) meshes with a gear sector bolted under the turntable plate (150 teeth on the full circle, cut over 76 degrees, since the head never turns further). That gives a 10:1 ratio. The turntable plate is trimmed to a 108 mm radius, so the pinion's tips clear it at every angle.
The first version cut the teeth straight into the 5 mm aluminium turntable plate, which would have been one part fewer. The check that the slots are cuttable failed it: the gap between module 1.5 teeth at the root is about 1.2 mm, and a laser shop will not cut a slot that narrow in 5 mm sheet. So the sector is printed in PETG and fills the 10 mm bearing gap, less 1.5 mm of clearance. Its teeth bend at 2.6 MPa at the root (Lewis' equation, with a 1.5 shock factor), well under the 10 MPa allowed for printed PETG here.
Inertia from the model, not from a guess
My first estimate of the turning inertia was 0.06 kg m². The build now measures it from the solids. Each made part gets its material density, and the two launcher motors and wheels are uniform solids of an assumed mass. The result is 0.11 kg m² with the head level and 0.31 kg m² with it tilted 45 degrees, because tilting swings the feed column and plates out behind the axis. That is five times the guess. With the guessed figure the motor looked fine; with the measured one, the half-second move I had pencilled in fails.
The move the drive must make comes from the drill file shipped with the project: its two corner targets, 2.5 m either side at 9.5 m, are 29.5 degrees apart. Allowing it one second (half of a 2 s feed interval) on a triangular speed profile, the turntable needs 1.16 N m: 0.65 to accelerate the head, 0.11 for bearing friction and 0.40 for the hose. The hose is modelled as a taut string pulling along its chord, which gives zero torque with the head straight ahead. Through the gears that is 0.129 N m at 100 rpm at the motor, 2.3 times under the 0.30 N m an ordinary 17HS4401-class stepper is assumed to give at that speed. The fastest move that keeps a 2x margin takes 0.88 s.
One full step is 0.18 degrees at the turntable, about 75 mm sideways at the far baseline. The firmware runs the driver at 4 microsteps, and 0.1 mm of gear backlash adds about 21 mm. The firmware had treated azimuth as a hobby servo. It now drives a step/dir axis that homes on a switch, plans trapezoid moves at the same 60 deg/s and 120 deg/s² the sizing used, and homes again after any stop fast enough to lose steps. The tests run the move and check it takes 0.99 s and sends exactly the right number of steps.
The turned head hit things
Every collision check in the package tilted the head while it pointed straight ahead. The new sweep also turns the head to ±30 degrees at both ends of its tilt range, against everything that does not turn. It found two clashes in the design as shipped:
- From 15 to 35 degrees nose up, turned 28 degrees or more, the head's plates cut into the hopper's front posts (up to 709 mm³). The posts moved out 20 mm each side.
- From 40 degrees nose up, turned 10 degrees or more, the plates' lower rear corner dips below the mast plate's top and clipped its rear edge. The mast plate now has a rear notch shaped to the head's swept width, clear of the posts' bolt holes.
With both fixed, a bisection over seven tilts found the head clear to the full 30 degrees everywhere. The build checks the four corner poses on every run.
What is still a number on paper
Nothing here has been built. The motor and wheel masses, the bearing friction, the stepper's pull-out torque, the PETG allowance and the hose's own stiffness are all assumptions, and the CAD README says which. The next useful readings need a person and cost little. Weigh the head. Measure the hose's torque on the head with a spring scale. Read the motor's real pull-out curve. Everything is in ventures/open-tennis-robots/cad/ and the regenerated package in manufacturing/feeder/.