AXF-1: one design, scaled per joint
Four sizes of the same printed axial-flux design (only Do changes) mapped onto a 30 kg humanoid. The budget delivers the thread's blunt finding: a full 25-DoF fully-printed bonded-magnet humanoid does not close: 18 kg of actuators is 60% of a 30 kg robot. It closes as a lean 14-DoF walker at 34%, and comfortably as a sintered-leg hybrid at 25%. The sovereignty tax is now a number.
The sizing skeleton
markers scaled to actuator size · lean config shownKnees = M · hips/ankles/shoulders = S/XS · wrists → tendon (◯).
| size | OD | peak | cont* | Kₘ | module |
|---|---|---|---|---|---|
| XS | 75 | 28 | 14 | 0.49 | 0.54 |
| S | 90 | 49 | 24 | 0.71 | 0.77 |
| M | 110 | 89 | 44 | 1.06 | 1.16 |
| L | 130 | 146 | 73 | 1.48 | 1.62 |
Does the mass close?
actuator module mass ÷ 30 kg robotReal humanoids run 30–50% of mass in actuation. Config A blows past it because low-torque-density printed bonded magnets multiplied across 25 joints simply weigh too much: every joint is correctly sized (all margins > 1.0), the count × mass is what fails. Cutting to a 14-DoF walker and letting the legs use sintered magnets both bring it home.
Joint assignment: lean walker (Config B)
smallest size clearing peak, with margin| Joint | × | DoF/side | req peak | size | peak | margin | mass |
|---|---|---|---|---|---|---|---|
| LEGS | |||||||
| Hip (pitch + roll) | 2 | 2 | 45 | S · 90 | 49 | ×1.08 | 1.55 |
| Knee | 2 | 1 | 55 | M · 110 | 89 | ×1.61 | 2.31 |
| Ankle (pitch) | 2 | 1 | 40 | S · 90 | 49 | ×1.21 | 1.55 |
| ARMS | |||||||
| Shoulder (pitch + roll) | 2 | 2 | 25 | XS · 75 | 28 | ×1.12 | 1.08 |
| Elbow | 2 | 1 | 18 | XS · 75 | 28 | ×1.56 | 1.08 |
| HANDS → tendon-driven (LEAP / ORCA class), outside the AXF-1 family | |||||||
| TOTAL | — | 14 DoF | — | — | — | — | 10.2 kg |
Torque targets are design intent for a squat-capable walker with mild dynamic margin (~1.5–2 N·m per kg of robot on the demanding leg joints). Knees are the one joint forced to M, the D³ torque wall. Hands are a separate problem (tendon drive), already flagged in the capability map.
The four knobs on the budget
① fewer DoF: the biggest lever
25 → 14 DoF drops actuators from 18.0 to 10.2 kg. Dropping hip-yaw, ankle-roll, shoulder-yaw and wrists (to tendons) costs some dexterity and turning grace but is the difference between infeasible and buildable. The Berkeley-Lite lesson: minimal DoF is a printability strategy, not a compromise.
② hybrid magnets: legs go sintered
Let the leg joints (knees, hips, ankles) use sintered NdFeB: ~2× torque density → ~0.6× module mass, while arms stay fully printed. Total falls to 7.4 kg / 25%. The full-print sovereignty tax is +38% actuator mass. A clean, quantified choice: pay 38% for rare-earth-independent, fully-printable legs, or don't.
③ torque target: don't over-provision
Torque ∝ D³, and mass ∝ D², so a joint spec'd for backflips instead of walking jumps a size and gains ~50% mass. Size honestly to the gait. Every unnecessary N·m of peak is paid for in kilograms across 14–25 joints.
④ robot mass: the weak lever
Raising the robot to 40 kg lowers the fraction, but required joint torque scales with mass × length, so the actuators grow too: it mostly chases its own tail. Better to attack the numerator (DoF, magnets) than inflate the denominator.
A 14-DoF, 30 kg biped: knees on M (110 mm), hips/ankles/shoulders on S–XS, hands on tendon. 10.2 kg of fully-printed actuators: 34%, inside the humanoid norm, with a drop-in path to 25% the day you allow sintered legs. This is the honest envelope of a fully-printed bonded-magnet humanoid: lean it, or hybridize it. A dexterous 25-DoF fully-printed machine at this scale waits on either a print-native higher-flux magnet or the SRM / rare-earth-free variant, which is exactly why that branch stays alive in the build order.
Nothing here moves until MEAS-1: the whole family's torque column rides on printed-magnet Br. Measure the coupon, then this table is real instead of intended.