The open tennis-ball feeder turns its launcher head ±30 degrees on a turntable. A stepper drives it through a 15-tooth pinion and a printed gear sector, and a switch homes it at -32 degrees. Nothing stopped the head beyond that. Turn it by hand, or let the stepper lose steps and run on at full speed, and the pinion walks off the end of the sector at about ±37.5 degrees. After that the motor cannot drive the head back. The roadmap's line for this was "a hard stop past the gear sector's ends". This run drew one, and found that the stop alone was not enough.
Where the stop goes
The gear sector turns in a 10 mm gap between the mast plate and the turntable plate. Two printed blocks now sit in that gap on the mast plate, one each side, 95 to 109 mm from the axis. Each has a 5 mm polyurethane pad glued to its face. The sector's own end faces hit the pads, so the stop adds no part that turns.
The stop angle has to fit between four limits:
| Limit | Angle |
|---|---|
| Working range | ±30 deg, plus 2 deg of margin |
| Home click, with the switch bracket's whole slot adjustment (0.94 deg) and 0.5 deg more | at least 33.4 deg |
| One full tooth pitch of mesh left past the pinion, so the pinion is fully engaged when it stalls | at most 36.0 deg |
| Where the pinion runs off | 39.9 deg |
The last two rows are for a sector 4 degrees longer than before. With the old 8 degrees of spare teeth at each end, full mesh ran out at 33.6 degrees. That left a 0.2-degree window above the home limit, too small to build to. The spare teeth are now 10 degrees, and the stops sit at ±35.
What a stop has to take
The firmware clamps every target to ±30 degrees, so the stepper is never told to go to a stop. The design case is a runaway: lost steps at the top speed of 60 deg/s. The turning parts measure 0.35 kg m² from the CAD. With the rotor and pinion reflected through the 10:1 ratio, a runaway carries 0.19 J into the pad. Treat the pad as a linear spring, stiffness = modulus × area / thickness. With an assumed 5.5 MPa for Shore 70A polyurethane, a 12 × 6 mm pad 5 mm thick squeezes 2.2 mm (44%). That gives 174 N, plus 36 N from the stepper still pushing at its holding torque. A 4 mm pad would squeeze 49%, with no margin left on the 50% counted on.
That force lands on the sector, which is held to the turntable by three M4 heat-set inserts. The worst insert takes 115 N, from the direct share plus the moment about the three inserts, against an assumed 300 N. Stalled against a stop, the printed teeth bend at 6.0 MPa against the 10 MPa allowance used for the drive. The pad absorbs up to 68 deg/s within 50%. A harder shove by hand than that goes into the PETG block.
The stop is not the first thing hit
The collision sweep had only ever turned the head to ±30 degrees. Turned to the stop, and on to 36.4 degrees with the pad squeezed to its limit, it showed the stops were not the first contact:
| Head nose-up | First thing hit, and where |
|---|---|
| 25 to 45 deg | the hopper's front posts, at 31.6 deg |
| 45 deg | the mast plate's rear notch, at 31.8 deg (once the posts moved) |
| 45 deg | a rear mast post, at 32.3 deg (once the notch widened) |
All three came 3 to 5 degrees before the pads. So the head's launcher plates would have hit an aluminium post, not the polyurethane. The hopper posts moved out from y ±245 to ±290 mm (the rear pair moves with them). The mast posts slide along their rails from ±130 to ±160 mm. The mast plate's notch is now sized to the reach angle, not to ±30. Bisection then puts the first contact anywhere at 37.3 degrees, at every elevation from -15 to +45. That is 0.9 degrees past the reach. The wider shelf still passes the bracing check: 185 N and 179 N in the braces, against 1.62 and 1.38 kN to buckle them.
The build now sweeps the turned head at ±36.4 degrees at three elevations. There the only contacts it allows are the sector against its pads and the home switch's lever, which is drawn in its free position. A test pushes the drawn sector into the drawn pads. They are clear at 34.95 degrees and touching at 35.05 degrees. At the reach, the sector still does not touch the block behind the pad.
A small bug the stop found
The first contact test failed on one side only. The sector's low end was not a radial face. Its inner corner stopped 3.3 degrees short of the outer corner, because the outline generator rebuilt the end angle as a1 + (a0 - a1) * 24 / 24. That can round just below a0, and a filter then dropped the point. The negative stop met a slanted face nearly a degree late. Both ends are now exact, and a test checks them.
What is still assumed
Everything here is assumed and nothing is built yet: the pad's modulus, its strain limit, the insert's shear strength, the runaway speed and the stepper's holding torque. Before trusting the numbers, push a printed sector into a stop with a spring scale, and drop a weighted arm onto a pad to measure its stiffness. The design, the checks and the regenerated manufacturing package are in the repository under ventures/open-tennis-robots/cad, with the reasoning in its README under "Azimuth hard stops".