Anatomy of a humanoid robot
What a humanoid is actually made of — five systems, 38parts. Here's what each one does, and why, on Valumech, the parts are the depreciation.
Every humanoid robot is built from the same major subsystems: the actuators that move it, the sensors that let it perceive, the compute and AI that decide what to do, the power system that keeps it running, and the hands and structure that let it act on the world. Here is what each part does, and why it matters for what a robot costs new and what it is worth used. Component functions are general engineering knowledge; representative makers and figures are cited.
Actuation & motion
The motors, gearboxes, and actuator assemblies that convert electrical power into joint torque and limb motion. Actuators are typically the single largest share of a humanoid's bill of materials, and their wear (gears, screws, bearings) heavily influences durability and used value.
Electric motor (BLDC / PMSM)
A brushless permanent-magnet motor that turns electric current into rotational torque - the prime mover inside nearly every humanoid joint.
Why it affects value Torque density (Nm/kg) sets how light and powerful each joint can be; the motor is one of the costliest items in an actuator, and actuators dominate the BOM.
e.g. Kollmorgen
Frameless (housingless) motor
A motor supplied as just rotor and stator - no housing, bearings, or shaft - so it can be embedded directly into a joint to save weight and volume.
Why it affects value Integrating the motor into the joint raises torque density and compactness, but tight integration makes the actuator harder and costlier to service or replace.
e.g. Kollmorgen
Strain-wave (harmonic) gear
A compact, near-zero-backlash speed reducer using a flexible toothed cup deformed inside a rigid ring to achieve very high reduction ratios. The classic choice for high-torque joints like hips and shoulders.
Why it affects value A large fraction of a rotary actuator's cost; its low backlash sets positioning accuracy, and flexspline fatigue is a key durability concern on used units.
e.g. Harmonic Drive SE
Planetary gearbox
A reducer with a central sun gear, orbiting planet gears, and an outer ring gear that shares load across many teeth - high torque in a compact coaxial package at low-to-moderate ratios.
Why it affects value The workhorse of lightweight, backdrivable quasi-direct-drive actuators; its simple, robust gearing tends to wear gracefully, helping used-unit reliability.
Cycloidal / RV reducer
A reducer where an eccentric drives a lobed disc against pins, transmitting motion through many contact points for high reduction, low backlash, and high shock resistance.
Why it affects value Carries high torque and shock loads compactly - valuable for heavy joints - and its rolling, multi-tooth contact wears slowly, supporting long service life.
e.g. Nabtesco
Linear actuator (ball / roller screw)
Converts motor rotation into linear push/pull via a screw, used where a joint is better driven by a sliding strut (knee, elbow) than a rotary module.
Why it affects value The screw is a precision, load-bearing wear item; high-quality linear actuators are expensive, and screw wear is a direct indicator of remaining useful life.
e.g. Tesla Optimus
Degrees of freedom (DoF)
Each independently controllable axis of motion is one DoF; a humanoid spreads dozens across legs, torso, arms, and hands, each driven by its own actuator.
Why it affects value DoF count maps almost directly to actuator count - and since actuators dominate the BOM, more DoF (especially dexterous hands) raises both build cost and the number of wear points affecting resale.
Quasi-direct-drive vs geared (tradeoff)
A design choice between QDD actuators (high-torque motor + low gear ratio) that are light, backdrivable, and force-sensitive, versus high-ratio geared actuators that are stiff, precise, and torque-dense but not backdrivable.
Why it affects value Dictates which costly components each joint needs and shapes the safety/compliance profile that determines a robot's suitability - and market value - for working around people.
e.g. MIT Cheetah
Sensing & perception
Exteroceptive sensors (cameras, depth/LiDAR, microphones) perceive the outside world while proprioceptive sensors (IMU, encoders, force/torque, tactile) let the robot sense its own body and contact forces. The richness of this suite is a primary driver of autonomy - and therefore capability tier and value.
RGB / 2D cameras
Standard color cameras; vision is the dominant perception modality for object recognition, scene understanding, and navigation, often feeding end-to-end neural networks.
Why it affects value Vision is the backbone of autonomy; a camera-first design (vs relying on teleoperation) is a key differentiator for higher-value robots.
e.g. Boston Dynamics
Stereo / depth cameras (RGB-D)
Two or more cameras (often with an IR projector) that compute per-pixel depth, producing a 3D point cloud; RGB-D adds aligned color for object size, shape, and distance.
Why it affects value Depth is essential for obstacle avoidance and grasp planning; its quality and range affect how reliably a robot can act autonomously.
LiDAR / time-of-flight
Laser sensors that measure pulse return time to build precise 3D maps, working well in low light or low-texture environments where cameras struggle.
Why it affects value Adds robust long-range 3D mapping for navigation; its presence (vs a camera-only design) is a cost/capability tradeoff.
e.g. Livox / Unitree
IMU (accelerometer + gyroscope)
Measures linear acceleration and angular rate - the robot's inner ear - to estimate body orientation and motion.
Why it affects value Foundational to dynamic balance and fall prevention; without reliable inertial sensing a biped cannot stand or walk.
e.g. Bosch Sensortec
Joint encoders (absolute / incremental)
Position sensors at each joint reporting angle and velocity - the basis of proprioception. Absolute encoders give exact angle without homing; incremental count pulses.
Why it affects value Accuracy and type set motion precision and repeatability; high-accuracy absolute encoders raise capability and cost - a clear value tier.
e.g. HEIDENHAIN
Force / torque sensors
Six-axis sensors measuring the three forces and three torques at a joint - typically at wrists (manipulation) and ankles (ground reaction for balance) - letting the robot feel contact.
Why it affects value Enables compliant, safe interaction and force-controlled manipulation (not crushing fragile objects), distinguishing capable contact-rich robots from rigid position-only ones.
Tactile / touch sensors
Fingertip- and skin-mounted sensors detecting contact, pressure, and texture; vision-based optical tactile sensors image gel deformation across many sensing points.
Why it affects value Fine touch enables delicate in-hand manipulation and slip detection - an emerging, high-value differentiator over grippers that rely on vision alone.
e.g. GelSight + Meta AI, Figure 03
Microphones / audio
Microphone arrays that capture speech and ambient sound for speech recognition, sound localization, and voice interaction; paired with speakers for output.
Why it affects value Enables conversational, voice-driven interaction - raising value as a human-facing collaborator rather than a silent industrial actuator.
e.g. Figure 03
Compute, AI & software
The robot's brain: onboard edge-AI processors run perception, reasoning, and real-time control, while a layered software stack - from kHz motor loops to whole-body control to vision-language-action models - turns sensor data into coordinated motion. Software and AI are a fast-growing share of value and can extend useful life via over-the-air updates.
Onboard compute (edge AI SoC / GPU)
A power-efficient processor that runs perception and AI inference locally, so the robot acts with low latency and without depending on a network.
Why it affects value A major and rising BOM cost that gates which AI models a robot can run; more capable silicon directly expands autonomy and resale appeal.
Low-level motor control
Deterministic, high-frequency (often kHz) loops that command each joint's torque, velocity, and position from encoder, IMU, and force feedback over a real-time bus.
Why it affects value Underpins safety and smooth motion; foundational firmware that everything above it depends on.
Whole-body control & balance
The mid-level controller coordinating all joints to keep the robot balanced (managing its center of mass) while walking, standing, and manipulating on uneven terrain.
Why it affects value A core capability differentiator: a robot that walks and recovers reliably is far more deployable - and valuable - than one that doesn't.
Vision-language-action (VLA) models
Multimodal foundation models that take camera images plus a natural-language instruction and output robot actions, letting a humanoid generalize across tasks instead of running hand-coded routines.
Why it affects value The fastest-growing and most defensible part of a humanoid's value; analysts liken connected robot software to a high-margin services layer on top of hardware.
e.g. NVIDIA GR00T N1, Figure Helix
Imitation & reinforcement learning
How skills are acquired: imitation learning copies human/teleoperated demonstrations; reinforcement learning discovers policies through trial-and-error reward optimization, heavily in simulation.
Why it affects value These pipelines and the demonstration data they consume are a key competitive moat; continuous learning of new tasks raises a fleet's long-term value.
Sim-to-real trainingemerging
Training policies in massively parallel physics simulation (with randomized dynamics and terrain), then transferring them to the physical robot - reducing costly real-world trials.
Why it affects value Lets developers scale skill development cheaply; faster, more reliable transfer compresses development cost and accelerates capability gains.
Teleoperation
A human remotely operating the robot - used both to gather high-quality demonstration data and as a fallback when autonomy can't complete a task.
Why it affects value The autonomy-vs-teleop balance is central to economics: many 'autonomous' demos still rely on teleop, and reducing the human-in-the-loop ratio drives margins and per-robot value.
e.g. Whole-body teleop
Over-the-air (OTA) updatesemerging
Secure remote delivery of new firmware, control policies, and AI models to a deployed robot or fleet over a verified channel.
Why it affects value Lets a robot gain new skills and fixes after purchase - extending useful life and supporting resale value, much as it does for connected cars - and underpins recurring software revenue.
Power & thermal
An onboard battery (usually lithium-ion) managed by a battery management system feeds power electronics that drive the joint actuators, while a thermal system sheds the heat dense actuators and batteries generate. Limited runtime is one of the central bottlenecks for industrial humanoids, which is why hot-swap packs and fast charging are emphasized.
Battery pack (lithium-ion)
The onboard energy store - usually lithium-ion cells - powering actuators, compute, and sensors, increasingly integrated into the torso as a load-bearing member.
Why it affects value A primary used-value factor: lithium-ion capacity fades with charge cycles and age, so remaining battery health directly affects resale value and may need an expensive replacement.
Runtime / operating timeemerging
How long the robot runs on a charge, driven by capacity vs the power that locomotion and tasks draw; dynamic bipedal motion is energy-intensive, so runtimes are often a few hours (commonly cited around 2-4).
Why it affects value A direct utility and deployment-economics factor: frequent charging means less useful work per shift, and since runtime depends on the aging battery, it declines over the asset's life.
Battery management system (BMS)
Electronics that keep the pack safe and long-lived - monitoring per-cell voltage, current, and temperature, estimating state of charge/health, balancing cells, and protecting against overcharge and thermal runaway.
Why it affects value Governs how gracefully the pack ages and reports state-of-health, strongly influencing a used robot's measurable remaining battery value and safety.
Hot-swappable batteries
Designs that let a depleted pack be exchanged for a charged one (manually or autonomously) without fully powering down, so the robot resumes work in minutes - a common path toward continuous duty.
Why it affects value Raises effective daily uptime and decouples availability from charge time, which can make a swap-capable platform more valuable for continuous operations.
e.g. UBTECH Walker S2
Charging
The port/dock and charge-control electronics that refill the cells with the BMS. Fast charging shortens downtime but adds heat, so charge rate is bounded by thermal design.
Why it affects value Faster, well-managed charging improves utilization; charging that stresses cells accelerates capacity loss, feeding back into battery health and resale value.
e.g. Figure 03
Power electronics / motor drivers
Circuitry that converts DC battery power into the controlled currents each joint motor needs; for brushless motors the driver acts as an inverter, switching DC into the waveforms that control torque and speed.
Why it affects value High-stress, heat-generating parts whose reliability and efficiency affect performance and repair cost, contributing to overall mechatronic value.
Thermal managementemerging
Systems that remove heat from actuators, power electronics, compute, and the battery via air or liquid cooling and heat-spreading structures - what allows sustained high power rather than brief peaks.
Why it affects value Bounds sustained performance and protects the battery and actuators from heat-driven degradation, supporting both utility and long-term value retention.
e.g. Figure 03
Hands, end-effectors & structure
The hands are where a humanoid does useful work - widely considered among the hardest, most expensive, and most differentiating parts to engineer - while the structure and materials set the robot's weight, durability, safety, and where it can be deployed.
Dexterous multi-fingered hands
Anthropomorphic end-effectors with multiple fingers and many DoF that approximate the human hand, enabling grasping, in-hand reorientation, and tool use rather than simple pick-and-place.
Why it affects value Among the hardest and most expensive parts to build and a major capability differentiator - rich manipulation determines how many real-world tasks a humanoid can actually do.
Tendon-driven vs motor-in-joint
Two ways to drive finger joints: tendon (cable) drive routes force from motors in the forearm/palm to keep fingers slim and compliant; motor-in-joint designs place actuators at each joint for stiffness and simpler control.
Why it affects value Drives a hand's cost, weight, robustness, and serviceability - a key tradeoff separating research-grade from mass-producible hands.
Grippers (simple end-effectors)
Non-anthropomorphic end-effectors - parallel-jaw clamps, suction/vacuum, magnetic - that trade dexterity for reliability on constrained tasks.
Why it affects value Far cheaper and more reliable than dexterous hands for fixed tasks, so the gripper-vs-hand choice is a central cost-versus-capability decision.
Structure / chassis (materials)
The load-bearing skeleton and housings that mount actuators - commonly aluminum alloys (6061, 7075), carbon-fiber-reinforced polymer, engineering plastics, and titanium at high-stress joints.
Why it affects value Material choice trades weight, stiffness, durability, cost, and repairability across the robot - a foundational cost-and-performance driver.
e.g. Aluminum alloy (6061 / 7075), Carbon-fiber-reinforced polymer
Weight & its tradeoffs
Total mass and its distribution, set largely by structure and battery choices and balanced against the need for strength and rigidity.
Why it affects value Lower weight improves payload ratio, runtime, and human-safety on impact, but cutting weight usually costs more (composites/titanium) or sacrifices stiffness.
Enclosures / ingress protection (IP rating)
Sealing of the body and joints, rated by the IEC 60529 IP code (first digit dust, second water) - e.g. IP65 is dust-tight plus water jets, IP67 adds brief immersion.
Why it affects value Determines where a robot can work (dusty/wet industrial lines vs clean indoor), directly gating addressable markets and ruggedization cost.
e.g. IEC 60529 IP code
Skin & expressive facesemerging
Non-structural covers and (for social robots) silicone faces with actuated features that protect internals, shape appearance, and convey expression.
Why it affects value For industrial robots, covers mainly add protection and finish; for social/service robots an expressive face is a primary value driver and cost center, since human acceptance depends on it.
From parts to price
These components are the cost drivers — and the things that wear, degrade, and get updated — so they're exactly what determines how fast a humanoid depreciates and what a used one is worth. That's what Valumech tracks across every model.
Frequently asked
What are the main parts of a humanoid robot?
A humanoid robot has five major subsystems: actuation (the motors and precision gearboxes that move each joint), sensing and perception (cameras, LiDAR, IMU, joint encoders, and force/tactile sensors), compute and AI (the onboard processor and the control and vision-language-action software), power and thermal (battery, battery-management system, and cooling), and the hands and structure (dexterous end-effectors plus the load-bearing frame and covers).
What is the most expensive part of a humanoid robot?
The actuators — the motors and precision reducers (harmonic, planetary, or cycloidal gears) in each joint — are typically the single largest share of a humanoid's bill of materials. Apptronik's CEO has said motor drivers and actuators are 60% or more of system cost.
Which parts most affect a used humanoid robot's value?
The battery (lithium-ion capacity fades with charge cycles and age) and the actuators' wear items (gears, screws, and bearings) are the biggest used-value factors, alongside the software and AI models, which can be improved over the air to extend a robot's useful life.
Component functions are general robotics engineering; representative makers and specific figures are linked to their sources inline (38 parts across 5subsystems). Where a claim couldn't be sourced, it was left out. How we value →