Print and assemble

Download the print files

coupon.stl head.stl link.stl link_s3_battery.stl link_s4_esp32.stl link_s5_power.stl tail.stl check_fit.txt (full fit report)

You print nine parts: seven links, a tail and a head. Each one carries an MG90S servo in a cradle at the front and a fork at the rear that the previous servo drives. Eighteen O-rings make the treads, two magnets in the tail and two steel screws in the head make the latch that closes the strap into a wheel, and the electronics tape onto the backs.

Two choices shape every part. First, each link is two blocks joined by a 1.0 mm PETG leaf spring. Each link gives about 7° at a firm push, so the strap has some give built into every segment. Second, the design works with slop. Hobby servos have backlash and clones vary, so the servo sits in its cradle with about 0.3 mm of play, the pivot arm swings free on its screw, and a printed end stop at 89.5° catches each joint on plastic before backlash and leaf flex can push it into the electronics. The simulation is trained with the same hard stop.

The parts are new designs, so print the test coupon first and check it against your servos before you print the set.

What you're printing

The full set is about 139 g plus the coupon, against a 130 g budget. Expect 9 to 10 h of printing. The masses are solid volume × 1.27 g/cm³ × 0.55 (PETG, 3 walls, 15 % infill), and the times come from mass at about 14 g/h, so weigh your first link and trust your slicer's time over these.

The end-stop toe adds 0.8 g to each link. The tail is the heaviest part because of the second pair of O-ring treads under the latch corner, which matches the tread k0 in the sim. The sim's s0 mass of 30 g assumes a 12.5 g tail print; with this tail it should be about 36 g.

Print settings

Print every part belly down exactly as exported (the STL's lowest face, z = -8, goes on the bed), with no supports and no brim.

The lower arcs of the R6.5 groove core and the R8 tread flange (|y| > 33) print as small droops near the bed. They are round on purpose because the O-ring needs them round. Two features are thinner than 0.8 mm by design and your slicer will drop them: the 0.2 mm wedge where the latch-wire groove breaks into the pin seat, and a 0.15 mm² sliver where the rear-tab slot cuts the leaf fillet. Both are fine.

Print two spare links. Fatigue life of the leaf at 14 to 16 MPa in printed PETG is unknown.

Print the test coupon first

The cradle and the horn recess are sized for a datasheet MG90S and a 1.7 mm horn plate. Clones vary by ±0.3 to 0.5 mm, and horns vary more. The coupon (one cradle and one fork, 30 minutes) tells you whether your servos fit before you commit 9 hours of printing.

Before printing anything, measure one servo with calipers. The four numbers that matter most:

1. case thickness
2. tab underside height
3. spline top height
4. horn plate thickness and length

Compare them with the values in MG90S parameters below. If your horn is longer than 15.5 mm from the centre to the tip, cut it down with flush cutters (the recess runs into the crossbar).

Then print the coupon. The servo should drop in with about 0.3 mm of play all round, and the horn should press into the fork recess and stay there. If it doesn't, edit the parameters in cad/make_parts.py and regenerate the parts (see Regenerating the parts). Expect one reprint.

One joint, step by step

A joint is the servo in link N's front cradle, driving link N+1's rear fork.

• Horn arm (+y): the horn presses into a recess on the arm's inner face.
• Pivot arm (−y): turns on the servo's own 2 mm screw, which threads into a boss on link N.

1. Centre the servo. Command 1500 µs (see the Build plan) and leave it there.
2. Horn into the fork. Press the single-arm horn into the recess on the inner face of link N+1's horn arm, with the hub facing the servo side. A small drop of superglue on the plate edge is fine. The fork holds the horn axially once assembled, so the glue only stops it rattling.
3. Servo onto the horn, outside the cradle. Hold the servo between the two fork arms and slide the spline into the horn hub along the shaft axis. There is 1.2 mm to spare past the pivot arm, so it goes in square. Choose the spline tooth that makes link N+1 point straight along the servo body at 1500 µs. One tooth off is 15°. Drive the horn screw through the hole in the horn arm.
4. Drop the servo, with link N+1 hanging on it, into link N's cradle from the back. - The rear tab goes into the slot in the front crossbar, and the front tab rests on the post. - The pivot arm goes down outside the pivot boss, and the horn arm goes outside the floor. - The lead leaves the rear of the servo into the leaf-spring gap.
5. Pivot screw. Drive one of the servo's 2 mm mounting screws through the hole in link N's −y cheek. It goes through the pivot arm and into the boss pilot. Stop as soon as the head touches the arm, then back off a quarter turn. The arm must swing freely; it has 0.15 mm of axial play. Tighten it and you add friction the sim does not have.
6. Zip tie (2.5 mm): - Thread it through the slot in the floor just past the servo top (the shaft end), tail end first. - Run it along the recess under the belly, up the outside of the wall at the servo's bottom end, and back over the servo. - Pull it tight with the ratchet head at the shaft end (+y, below the horn), never on top. On top, it sits under the boards on s4 and s5. - Cut the tail flush. The tie sits 0.35 mm proud of the belly and the treads stand 1.0 mm proud, so it clears the desk.
7. Check by hand. Bend the joint down to −69°: nothing should touch except the horn turning. Then fold the joint up by hand: it must stop on plastic at 90° with the electronics untouched. The two toes on the next link land flat on the tops of this link's cheeks. Then press the two halves of each link together: the leaf spring should give about 7° at a firm push.

Measure the leaf spring on your first link. Hang 150 g at 100 mm from the joint and read the angle. 7° is on target. Printed PETG ranges from 1.8 to 2.2 GPa and layer adhesion matters, so your number can differ from the design value.

Check the end stop on your first printed pair. It must land on the cheek tops at 89° to 90°, and the 45° toe must not crack when the servo stalls against it.

The other joints and the O-rings

Build the remaining joints the same way.

• The horn screw and the pivot screw can both be reached later with a 3 mm screwdriver through the holes in link N's cheeks, so you can re-index a horn without taking the joint apart.
• On s4 and s5 the board sits over the servo. Swapping that servo means peeling the board off first.
• Leave a 5 mm S-loop in each servo lead where it crosses its own leaf-spring gap, so the lead does not stiffen the spring.

O-rings: 18 of them, 18 × 2.5 mm nitrile.

• Two go on the front of every link s0 to s7, plus two on the rear of the tail. The head has none.
• Roll each one over the outer R8 flange into the groove. That is a 24 % momentary stretch, easy by hand. Seated, it is stretched 7.6 % on the inside and stands 0.92 mm proud of the plastic around the joint.
• Wipe the rings and the desk with isopropyl before a rolling run.

The magnet latch

The tail carries two magnets in notches. The head carries two steel screws as latch pins. When the strap rolls up, the pins drop into the notches, the magnets pull them home, and each pin closes a sense circuit so the robot knows it has become a wheel.

Magnets (two 6 × 3 mm discs, in the tail):

1. Check polarity first. Stack both magnets on a screwdriver and mark the outward face of each, so both point the same way.
2. Put a drop of superglue in each pocket and push the magnet down from the back.
3. 0.8 mm of plastic separates each magnet from the seat. With the pin pulled home, the gap from pin to magnet is 0.9 mm.

Latch pins (two M3 × 25 zinc-plated steel screws, in the head; not stainless):

• Solder a wire to each screw head before driving it in: left pin to GPIO 27, right pin to GND (see Electronics).
• Drive each one from outside through the outer fork and the inner fork. They self-tap into a 2.6 mm hole. The pan head sinks flush into the outer fork, and the tip ends 5 mm from the centreline.

Latch-sense wires (in the tail):

• Lay a bare 0.5 mm tinned copper wire in the groove across the lower front of each pin seat, then up the groove on the outer face of the block. Glue the ends on the back.
• Join the two notch wires with an insulated wire across the tail's back. Use one wire per notch: a single wire straight across the middle would pass through the head's inner forks.
• When a pin is pulled home by the magnet it bears on its wire. The wire stands 0.15 mm proud, which gives a 0.07 mm overlap with the pin, so test continuity on your build.

Pull-off test: follow the step in the Build plan. The magnet pull through 0.8 mm of plastic plus the 0.25 mm gap is an estimate, so this test is where you find out. If the latch is too strong, shim each magnet deeper with paper before gluing. The seat funnel opens at 45° to ±4.5 mm, so the pin is steered in from that far off-axis. Watch the first few closures: the fit check covers the seated pose, and the swing into the funnel is something you confirm on the real robot.

Where each board mounts

Everything sits on the back, which is the inside of the wheel. Use 1 mm foam tape, zip ties and hot glue, with no extra screws.

No board may touch both halves of a link. Fix each one to one block, the front or the rear, and keep it at least 2.5 mm clear of the other, checked with the leaf flexed ±10°. Never tape wires across a leaf spring.

Wiring and soldering for these boards are on the Electronics page.

Why it is laid out like this

The back is the inside of the curl. At +90°, a neighbouring link stands up as a wall at x = 8 and x = 40, so each back has a 32 mm window. Everything above is placed to fit that window with the joints folded.

• Battery standing up. A pack lying flat (25 mm along the body) on s3 cannot share the window with the ESP32 on s4 when j4 is at +90°. Standing on edge, it takes 12.5 mm.
• Power board 20.5 mm deep along the body (x 19.0 to 39.5). It has to clear the ESP32 on s4 when j5 is at +90° and s6 when j6 is at +90°.
• Mid IMU at the rear of the module can, kept below y 23.8. That keeps it clear of the battery's y band when j4 is at +90°, the INA219 at j5 +92°, and the servo-plug wires at j5 +90°.
• XT30 and fuse on top of the pack, not on the s3 front block. The front-block spot (x 29 to 39, y −33 to −8) lies inside the ESP32's y band, so it is hit from j4 = 75°. The battery would also hit it as s3's leaf flexes. If you see the front-block position anywhere else in the docs, use the on-top position given here.
• Wire channel on s6 (and the other plain links) at x 30 to 34. At x 27.5 to 31 it would meet the servo-plug wires (z 29) on s5 at j6 = 87°.

Keep the stacks low

The electronics clear the end stop by only 2.5° to 3° on j4 to j6 (rigid links). Anything taller than the dimensions above eats that margin:

• Battery size. Packs sold as 2S 500 mAh range from 47 × 25 × 10 to 56 × 28 × 13. Anything thicker than 12.5 mm in x moves the front of the battery past 28.6 mm and eats the j4 margin. A longer pack overhangs further at −y, which is fine up to y = −40.
• Foam tape. Real foam tape ranges from 0.8 to 1.6 mm. Thicker tape raises the s4 stack and costs about 2° of j4 margin per millimetre.
• Servo wires. Keep the plug wires on s5 bent flat by z 29.
• Wire bundles and hot glue. The bundle thickness in the channels, slack loops squashed inside the curl and hot glue are not in the model. The fit of the 45° snap lips on the wire channels is a best estimate too.

After mounting, fold each joint to the stop by hand again and look at j4, j5 and j6.

Servo lead lengths

With the header on s5, the MG90S lead taken as 240 mm, and 15 mm of slack per joint crossed. Measure your leads; these numbers assume 240 mm and the routing above.

Design reference

cd cad
python make_parts.py     # parts -> out/*.stl, prints masses
python check_fit.py      # all fit checks, PASS/FAIL table, out/check_fit.txt (about 25 s)
python export.py         # out/assembled_*.glb/.stl and out/render_*.png