Why the cable goes down the middle

Put a cable on the rotation axis of a joint and it only ever twists. Move it 30 mm off the axis and every swing of the joint also bends it, stretches it and drags it across whatever it touches. That’s why industrial robot arms wear dress packs and cable guides on the outside, and why they still wear out.

A humanoid has no room for any of that. Loops of cable at the shoulder snag on shelving, catch on people and get pinched when the elbow folds flat. So the standard answer is a hollow-shaft actuator: a frameless motor whose rotor sits on a tube, a reducer with a clear bore through its centre, an encoder that reads around the hole rather than across it, and an output flange with a matching opening. The wiring for everything downstream passes through the joint like thread through a bead.

Unitree’s G1 is the textbook case. A teardown by harness maker JONVER found that all motor power and signal cables run through the centre of the joint shafts, with the motors in each limb connected in series through internal channels. Power comes in on a unified XT30 connector, data runs on CAN, and the main board in the torso fans out to every limb.

The bore is a budget

A hole in the middle of a joint isn’t free. The reducer’s wave generator, the motor rotor and the output bearing all have to grow around it, so a hollow unit carries more metal at a larger radius than a solid-shaft one of the same torque. The trade is visible in Harmonic Drive’s catalog, the closest thing the industry has to a reference design. Across the SHG-2UH hollow-shaft range the bore holds steady at roughly a third of the outside diameter.

Those are industrial units, heavier than what goes into a production humanoid. But the ratio carries over. A wrist-sized joint around 70 mm across leaves you about 20 mm of bore; a hip-sized one can offer close to 40 mm. Motor makers design for the same space. Kollmorgen’s TBM2G frameless motors, pitched at robot joints, offer rotor bores from about 18 mm in the 60 mm frame up to 44 mm in the 115 mm frame, explicitly to pass “encoders, cables, hoses, shafts, tools”.

Four wires, not forty

Twenty millimetres is plenty for a thin bundle and hopeless for a fat one. What keeps the bundle thin is the network, not the mechanics. Each joint on a modern humanoid carries its own servo drive, so the cable crossing a joint doesn’t need three motor phases and an encoder lead for every actuator further out. It needs a shared DC supply and a shared bus.

Menlo’s open Asimov humanoid shows how far that goes. Every actuator has two electrically identical XT30 (2+2) ports carrying exactly four conductors: BATT+, BATT–, CAN_H and CAN_L. The actuators daisy-chain along six CAN branches from the body – legs, arms, waist, neck – and the assembly guide has builders thread single-ended cables through the hollow shafts at the hip roll, knee and shoulder roll before soldering the joins. Thread it first, join it second. Get the order wrong and you take the joint apart again.

The cost of sharing one supply is voltage sag when several joints pull peak current at once, as they do on a jump landing. The G1 answers that with distributed capacitor buffers at the supply end of the outermost joints, according to the JONVER teardown. Buses with more bandwidth, such as EtherCAT, add conductors and shielding, which is one more reason bore size and network choice get decided together.

Twist, not bend

A cable through a hollow shaft is clamped at both ends – one on the fixed side, one on the part that moves. When the joint swings, the length between the clamps twists. It doesn’t rub, but it does fatigue, and the fatigue depends on how much twist each metre of cable has to absorb.

Industrial cable makers rate this directly. In July 2026 igus launched a robot bus cable, the CFROBOT8.PLUS family covering CAN, Ethernet up to CAT7, PROFINET and DeviceNet, rated for ±360° of torsion per metre, designed for 10 million cycles and tested past 15 million. Those figures come from six-axis industrial robots, not humanoids, but the arithmetic transfers.

That’s why free length is a design parameter, not packaging slack. Shorten it to save space and every cycle costs more of the cable’s life.

When the joint never stops: slip rings

Twisting works as long as the joint has end stops. Take them away and no cable survives. Boston Dynamics did exactly that with electric Atlas: of its 56 degrees of freedom, most can rotate 360°, and both the head and waist turn all the way round so the robot can work behind itself without stepping. The company hasn’t published how it carries power and data across those joints.

The usual tool for continuous rotation is a slip ring: rotating contacts that pass current across a spinning interface. Industrial options already cover the two things a humanoid needs. Mercotac’s liquid-metal contact rings, sold for robotics, carry up to 30 A per conductor in the eight-circuit Model 830. For data, ROTOCON markets brushless EtherCAT slip rings listed with the EtherCAT Technology Group and claims zero packet loss in rotation. Supplier Servotecnica lists the problems that remain for humanoid joints – size, weight, wear, electrical noise, power handling and cost – and each one grows with the number of circuits, which is one more argument for four-wire buses.

What to check on a joint module

• Clear bore through the whole stack – motor, reducer, encoder, brake and flange – not just the gear. The smallest hole sets the limit.
• Conductors per joint crossing. Local drives on a shared DC bus and CAN or EtherCAT keep it to a handful; centralised drives multiply it.
• Joint range against free cable length. Work out degrees of twist per metre and compare it with the cable’s torsion rating.
• Connector serviceability. Atlas builds joints as swappable modules; Asimov needs cables threaded before they’re joined. The routing decides how fast a joint can be replaced.
• Continuous rotation. Any joint without end stops needs a slip ring or a contactless link, with its own mass, noise and lifetime.

FAQ

Sources