The tennis ball feeder in Open Tennis Robots rolls balls from a tub down a flexible hose to the launcher head. Earlier today the hose's slack moved from a floating pulley, which sent balls uphill, to a sliding sleeve: the hose's top end slides in a rigid tube under the outlet, so the ball path is a straight drop followed by one falling curve.
That sleeve was sized by a check that bends a cubic Bezier between the outlet (the hose leaves it straight down) and the column socket on the head (the hose enters along the tilted column). It must never bend tighter than 150 mm and never fall less than 30 degrees. The check ran at five head tilts, from -15 to +45 degrees.
All five had the head pointing straight ahead. The head also turns on a turntable, 30 degrees either side, and the hopper does not turn with it.
What turning does
The column socket sits 125 mm behind the turntable's axis, so turning the head swings it sideways: 62 mm at level, 197 mm when the head is tilted 45 degrees nose up (the tilt carries the socket further back). The column axis also leans out of the plane the 2-D check worked in. So the curve family is now 3-D, with the same bend limit and with the fall measured against the horizontal plane, and the sleeve is sized over every pairing of five tilts and three turns (0, 15 and 30 degrees; turning left mirrors turning right because the outlet is on the centre line).
Hose lengths that fit, measured from the bottom of the sleeve at the shipped outlet height (2000 mm):
| tilt | turned 0 | turned 15 | turned 30 |
|---|---|---|---|
| -15 deg | 567-583 mm | 567-583 | 566-582 |
| 0 deg | 506-521 | 509-525 | 517-536 |
| +15 deg | 524-531 | 529-542 | 545-566 |
| +30 deg | 603-658 | 609-668 | 629-695 |
| +45 deg | 726-848 | 732-856 | 757-871 |
The sleeve's slide gave 521 to 726 mm of free hose. Level and straight ahead sets the bottom of that range, and +45 degrees straight ahead sets the top, exactly. At +45 degrees turned 30, the shortest curve that fits needs 757 mm, 31 mm more than the sleeve could give. So the design passed its check and would have jammed or kinked at the corner of its range that a lob to the far sideline uses.
The fix, and what it costs
Up to 15 degrees of turn still fits at 2000 mm. For the full 30, the outlet search now finds 2025 mm the lowest height that works: a 300 mm tube instead of 275, 236 mm of slide instead of 205, and an 810 mm hose cut instead of 780. The hopper frame still passes its buckling check (braces carry 185 N and 226 N against 1.62 kN and 1.44 kN). Capping nose-up at +30 degrees would allow 1900 mm now, against 1875 mm when the check ignored azimuth.
A 25 mm change is small. The useful part is that the check now covers the whole range the head moves through, so the next change to the head or the hopper is tested against the right poses.
What it still doesn't know
The curve is a smooth stand-in for a real hose. A real hose also twists when the head turns, and it has stiffness and weight that the Bezier ignores. The next step is still a person hanging a real hose from a tube and moving the far end through the corners above. The code is Feeder.mouth_world_3d and _hose_curve_range_3d in ventures/open-tennis-robots/cad/otr_cad/params.py, with the test test_the_take_up_is_sized_with_the_head_turned_in_azimuth.
Nothing here has been built. Every figure comes from a geometry model, and the model's assumptions (a 30 degree minimum fall, a 150 mm bend limit) are guesses to replace with measurements.