Charlot Lab · Physical AI × Material Science × Additive Manufacturing
A readiness map for building a humanoid entirely by additive manufacturing — every actuator subsystem, what AM can do today with hard numbers, and the one wall it can't cross.
Everything in additive manufacturing for robots splits into two worlds. A fully-printed humanoid lives in the gap between them — which is why no one has built one.
Metal AM. Every subsystem — cores, windings, magnets, flexures — is individually demonstrated at high performance.
Polymer AM. The entire robot already prints in one shot — including its control logic as printed fluidic transistors.
Each actuator subsystem, the AM processes that reach it, the best result demonstrated to date, and an honest technology-readiness estimate. TRL is shown as a nine-layer deposition stack.
The standard radial BLDC is optimized for machining and winding, not printing. These are the print-native alternatives — ranked by how much of each one AM can actually make.
You cannot print a power MOSFET, GaN, or SiC die, or an MCU. These require crystalline-semiconductor fabrication; printed and organic thin-film transistors cannot switch motor-level power. Printed electronics reaches the board, interconnect, passives, and even PCB-trace encoders — but not the active power and logic silicon. This is the only subsystem AM genuinely can't cross. There are two ways through it.
Trade "printed" for "ours." A self-fabricated open die (SKY130 / GF-class) is a coherent — arguably stronger — sovereignty claim than pretending a printed transistor can drive 40 A. The lab already fabricates silicon.
Printed fluidic logic already controls whole actuated robots end to end — diodes and pressure-gain "transistors" printed in a single run. Zero silicon, at the cost of speed and power density.
In the rigid regime, every piece is proven but:
In the soft regime, the whole robot prints in one shot but can't carry a biped's loads.
The track's thesis is the bridge: rigid-regime force density, soft-regime one-print integration, and printed-or-self-fabbed control — accounted in joules per joint, released CC0 as prior art.
Four capabilities, four separate research communities. No group occupies the intersection — multi-material AM + morphology/control co-design + print-native actuation at humanoid load. The nearest institutional analog is a wind-turbine program.
No single machine spans structural polymer + soft-magnetic steel + copper + magnet + elastomer. It's a process portfolio. Here is the optimal process per subsystem, the frontier that consolidates them, and the sequence to build.
The minimum viable stack is three machine classes: (1) polymer powder — MJF or SLS — for structure, gears and flexures; (2) multi-material metal LPBF for the electromagnetic actuator; (3) material-jetting or direct-ink-write for soft actuators and sensors. Metal LPBF is the capability-defining, expensive one.
The single machine that collapses the actuator core: Aerosint's selective powder deposition prints 316L steel + CuCrZr copper in one LPBF build — exactly the soft-magnetic-core-plus-winding pairing a motor needs. No platform yet spans metal + polymer + functional inks in a single system. Owning that convergence is the manufacturing-research frontier — and the moat.