Part I — They Are Coming#

Technical Concepts#

Actuator

A mechanical component that converts energy into physical movement. In humanoid robots, actuators drive every joint — hips, knees, ankles, shoulders, fingers — and determine how strong, fast, and precise the robot's movements are.

Artificial Intelligence (AI)

Software that can perform tasks normally requiring human intelligence, such as recognizing objects, understanding language, or making decisions. In this book, AI refers primarily to the machine learning systems that allow humanoid robots to perceive their environment, interpret instructions, and act autonomously.

Autocatalysis

The process by which humanoid robots participate in manufacturing other humanoid robots. This self-reinforcing loop could dramatically accelerate production timelines, since each new robot can help build the next generation faster than human labor alone.

Brushless Motor

An electric motor that uses electronic commutation instead of physical brushes, offering higher efficiency, longer lifespan, and better torque-to-weight ratio. Most modern humanoid robots rely on brushless motors for joint actuation.

Carcinisation

A pattern in biology where unrelated crustacean species independently evolve toward the crab body plan. Used in this book as an example of convergent evolution — the idea that effective physical forms emerge repeatedly because they solve environmental challenges well.

Convergent Evolution

The process by which unrelated species independently develop similar traits in response to similar environmental pressures. In this book, it supports the argument that the humanoid form may be a naturally optimal design for intelligent manipulation of the physical world.

Depth Camera

A sensor that captures both a standard image and the distance to each point in the scene, creating a three-dimensional map of the environment. Essential for humanoid robots to navigate spaces, identify objects, and avoid obstacles.

Force Sensor

A sensor that measures pressure and resistance, allowing a robot to detect how hard it is gripping an object or how much force is being applied to a surface. Embedded in robotic hands and joints, force sensors enable delicate manipulation — picking up an egg without crushing it.

General-Purpose Robot

A robot designed to perform a wide variety of tasks across different environments, rather than being optimized for a single function. Humanoid robots are general-purpose by design: they sacrifice peak performance at any one task in exchange for versatility across many.

Haptic Feedback

Technology that transmits the sense of touch through vibrations, pressure, or resistance. In telepresence applications, haptic feedback lets a human operator feel what a remote robot is touching, enabling more precise and intuitive control.

Humanoid Robot

A robot built in the approximate shape of a human body — two legs, two arms, a torso, and a head — designed to operate in environments built for people. Unlike industrial robots bolted to factory floors, humanoids can walk through doors, climb stairs, and use human tools.

Industrial Robot

A programmable machine designed for a specific manufacturing task, typically a robotic arm mounted to a fixed position on a factory floor. Over 4 million industrial robots operate worldwide, but each is a specialist — it cannot leave its station or adapt to a different job.

Large Language Model (LLM)

An AI system trained on massive amounts of text data that can understand and generate human language. LLMs give humanoid robots the ability to receive instructions in plain speech rather than code, and to engage in basic conversation.

LiDAR

A sensor that uses laser pulses to measure distances and build detailed 3D maps of the surrounding environment. LiDAR gives humanoid robots 360-degree spatial awareness, complementing cameras in situations where lighting or visual contrast is poor.

Motion Capture

A technology that records human movement using sensors or cameras, translating body positions into digital data. Used both to train humanoid robots by demonstration and to enable teleoperation, where a human's movements are mirrored by a remote robot in real time.

Neural Network

A computing system inspired by biological brains, consisting of layers of interconnected nodes that process information in parallel. Neural networks are the foundation of most modern AI, including the vision, language, and motor control systems used in humanoid robots.

NVIDIA Isaac Sim

A simulation platform developed by NVIDIA that lets engineers train robots in highly realistic virtual environments. Robots can practice thousands of hours of tasks in minutes, testing scenarios that would be dangerous, expensive, or impractical to run in the real world.

Policy Model

In robotics, a trained AI model that determines what action a robot should take given its current situation. Humanoid robots typically run two policy models simultaneously: a fast one (100+ Hz) handling reflexes like balance, and a slower one (1-10 Hz) handling planning and decision-making.

Precision Reducer

A gearbox that converts a motor's high-speed, low-torque output into slower, more powerful movement. Combined with brushless motors, precision reducers give humanoid robots the strength to lift heavy objects while maintaining fine control for delicate tasks.

Robot Density

A metric that measures automation adoption, expressed as the number of robots per 10,000 manufacturing workers in a country. South Korea leads the world at 1,012 — roughly one robot for every ten factory workers.

Robots-as-a-Service (RaaS)

A business model where companies lease humanoid robots rather than selling them outright, typically bundled with maintenance, software updates, and support. RaaS lowers the barrier to adoption and was pioneered commercially by Agility Robotics' Digit deployments.

Simulation / Synthetic Training

The practice of training robots in virtual environments that replicate real-world physics, rather than requiring physical trial and error. Simulation allows engineers to generate billions of training scenarios overnight and test modifications that would take weeks in a physical lab.

S-Curve Adoption

The pattern most technologies follow when spreading through a market: slow early growth, rapid acceleration through the middle phase, then gradual saturation. Humanoid robots are currently at the very bottom of the S-curve, in the slow early phase.

Teleoperation

Controlling a robot remotely in real time, typically using a VR headset and motion capture suit so the operator sees through the robot's cameras and moves its limbs by moving their own. Useful for dangerous environments like nuclear facilities or disaster zones.

The Great Stagnation

A term popularized by economist Tyler Cowen in his 2011 book, arguing that economic growth slowed from the 1970s onward as earlier sources of easy progress — free land, mass education, powerful new machines — were exhausted. As used in this book, the concept is applied specifically to physical-world innovation: venture capital moved toward software, which offered faster returns, while hardware problems — robotics, energy, transportation — were starved of investment.

Transformer Architecture

The neural network design behind most modern AI breakthroughs, including ChatGPT. Transformers process data by learning relationships between all elements in a sequence simultaneously, enabling major advances in language, vision, and robotic control.

Uncanny Valley

A concept coined by Japanese roboticist Masahiro Mori in 1970 describing the discomfort people feel when a robot or animation looks almost but not quite human. The phenomenon went from academic theory to mass experience when videos of near-human robots went viral in 2025.

Vision-Language-Action Model (VLA)

An AI architecture that integrates visual perception, language understanding, and motor control into a single system. VLAs allow a humanoid robot to see an environment, hear a spoken instruction, and execute the corresponding physical action without needing separate subsystems for each step.

World Model

An AI system's internal representation of how the physical world works — how objects move, fall, collide, and interact. World models allow robots to predict the consequences of their actions before executing them, similar to how humans mentally simulate catching a ball before reaching for it.

Companies & Organizations#

1X Technologies (EVE, NEO)

Norwegian-American robotics company backed by OpenAI. EVE, their wheeled humanoid, was deployed in security and industrial settings. NEO, their bipedal humanoid launched for pre-order in October 2025 at $20,000, is designed for home use — making it one of the first humanoids marketed directly to consumers.

Agility Robotics (Digit)

Oregon-based company that builds Digit, a bipedal robot designed for warehouse logistics. Digit became the first humanoid to generate commercial revenue under a robots-as-a-service model when it was deployed at a GXO Logistics warehouse in Georgia.

Apptronik (Apollo)

Austin-based humanoid robotics company spun out of the University of Texas. Apollo uses linear actuators that mimic human muscles. Apptronik partnered with Mercedes-Benz to explore automotive applications in 2024 and began pilot deployments at Jabil manufacturing facilities in early 2025, backed by a $350 million Series A led by B Capital and Google.

Boston Dynamics (Atlas)

Originally a spin-off from MIT, now owned by Hyundai. Boston Dynamics is known for decades of advanced legged robotics research. Atlas, their humanoid, transitioned to a fully electric design in 2024 and entered commercial production in partnership with Google DeepMind.

Clone Robotics (Protoclone)

A company building humanoid robots using artificial muscles rather than traditional motors, aiming to replicate the human musculoskeletal system. Their Protoclone video went viral in February 2025, accumulating tens of millions of views and triggering widespread uncanny valley reactions.

Figure AI (Figure 02, Figure 03)

San Francisco-based humanoid robotics company founded in 2022. Figure 02 was deployed at BMW for automotive production. Figure 03, announced in late 2025, was the company's first humanoid designed for home use. Valued at $39 billion as of September 2025.

Tesla (Optimus)

Tesla's humanoid robot program, first announced in 2021. Tesla plans to leverage its manufacturing expertise to mass-produce Optimus at scale, converting its Model S and Model X production lines at the Fremont factory to robot manufacturing in 2026. Target: millions of units per year.

Unitree (G1, R1)

Chinese robotics company that disrupted the market with aggressive pricing. The G1 launched at $16,000 in 2024; the R1 dropped to under $6,000 in 2025, making it the most affordable bipedal humanoid ever sold — though at 1.2 meters tall and without working hands, it is a research platform rather than a full-size worker. Deployed at Chinese automakers Nio and Geely.

---

Part II — The Market#

Technical Concepts & Terms#

4680 Battery Cell

A cylindrical lithium-ion battery format developed by Tesla, measuring 46 mm in diameter and 80 mm in length. The 4680 offers higher energy density, faster charging, and lower production cost per kilowatt-hour than previous cell designs. Originally developed for electric vehicles, the format's compact size and high capacity make it relevant to humanoid robots, where every kilogram of battery weight affects mobility and operating time.

Android (robot type)

A humanoid robot designed to closely resemble a human being in appearance, including realistic skin, facial features, and body proportions. Distinguished from general-purpose humanoids, which are typically utilitarian in design. Companies like Clone Robotics and Realbotix are building androids that blur the line between machine and human likeness, raising questions about the uncanny valley and the social implications of lifelike machines.

Degrees of Freedom (DoF)

The number of independent axes along which a robot's joints can move. A higher DoF count allows more complex and fluid motion. The human body has over 200 degrees of freedom. XPENG's Iron robot features nearly 200 DoF; Clone Robotics' Protoclone exceeds 200. Most industrial humanoids operate with 30–60 DoF, sufficient for manipulation and locomotion but far short of full human-like articulation.

Dojo

Tesla's proprietary supercomputer project, designed to train neural networks using custom D1 chips rather than relying exclusively on NVIDIA GPUs. First announced in 2019 and operational from July 2023, Dojo was built to process the vast video datasets generated by Tesla's global vehicle fleet for Full Self-Driving development. In August 2025, Tesla shut down the Dojo project; Musk called it "an evolutionary dead end" and redirected resources toward Cortex, a ~50,000 NVIDIA H100 GPU cluster at Gigafactory Texas, and future AI5/AI6 chips. Dojo's significance in this book lies in what it represented: Tesla's ambition to control the entire AI training stack from silicon to software.

Fertility Rate (Replacement Rate)

The average number of children born per woman in a population. A fertility rate of 2.1 is considered the replacement level needed to maintain a stable population without immigration. As used in this book, the global decline in fertility rates — South Korea at 0.75 in 2024 (up marginally from a record low of 0.72 in 2023), Japan and China well below replacement — is a central argument for why humanoid robots are needed: there are not enough people to do the work that economies require.

Foundation Model

A large AI model trained on broad data that can be adapted to many downstream tasks. In robotics, foundation models like NVIDIA's GR00T provide general capabilities — perception, reasoning, manipulation — that developers can fine-tune for specific applications. Distinguished from task-specific models, which are trained for a single function. The term draws a parallel with large language models like GPT, but applied to physical action in the real world.

Full Self-Driving (FSD)

Tesla's autonomous driving system, which processes data from cameras across a global fleet exceeding 9 million vehicles. FSD's neural networks — trained to perceive environments, predict behavior, and make real-time decisions — form the AI backbone that Tesla is adapting for Optimus. The same core challenge applies to both cars and robots: navigating dynamic, unpredictable environments using vision-based perception.

GDPR (General Data Protection Regulation)

The European Union's data privacy framework, enacted in 2018. Relevant to humanoid robotics because robots operating in homes, hospitals, and workplaces will collect vast amounts of personal data — visual, audio, biometric. GDPR and equivalent regulations (China's Data Security Law, California's CCPA) create regulatory moats that favor local robotics providers and complicate cross-border deployment.

Gigafactory

Tesla's term for its large-scale manufacturing facilities, located in Fremont (California), Shanghai, Berlin, and Austin (Texas). The Shanghai facility has an annual production capacity exceeding 950,000 vehicles and delivered over 916,000 in 2024. Tesla plans to adapt the Gigafactory model for humanoid robot production, sharing production lines and processes between vehicles and robots. The Fremont factory's Model S/X lines are being converted to Optimus production in 2026.

Hourglass Architecture

A framework for understanding the emerging structure of the humanoid robotics industry, analogous to the architecture that enabled the internet's growth. Diverse applications sit at the top (warehouses, hospitals, homes). Diverse hardware manufacturers sit at the bottom. In the middle, at the narrow waist, sits a standardized core: shared foundation models, common operating systems, and agreed-upon communication protocols. The concept suggests that value will be created at the edges, not at the center, just as the internet's value came from millions of applications running on TCP/IP, not from the protocol itself.

Humanoid Network Operator (HNO)

A hypothetical new category of business that would own, deploy, and maintain fleets of humanoid robots on behalf of customers — analogous to how telecom operators manage mobile networks without manufacturing phones. The concept addresses the gap between manufacturing robots and operating them at scale, including teleoperation, regional maintenance, software updates, and customer support. Whether HNOs actually emerge or manufacturers retain vertical control remains one of the open structural questions of the industry.

Jetson Thor

NVIDIA's next-generation computing platform for humanoid robots, powered by a Blackwell GPU. Designed for real-time on-robot inference — processing sensor data and making decisions locally rather than relying on cloud connectivity. Adopted by Figure AI, Galbot, Google DeepMind, Mentee Robotics, Meta, Skild AI, and Unitree. Jetson Thor represents the computational hardware layer of NVIDIA's robotics ecosystem, complementing its software platforms (Isaac, Cosmos, GR00T).

K-Humanoid Alliance

A South Korean national initiative launched in April 2025 by the Ministry of Trade, Industry and Energy, uniting over 40 companies and universities in a coordinated effort to develop domestic humanoid robotics capability. Members include Rainbow Robotics (Samsung), HD Hyundai Robotics, Doosan Robotics, and battery makers LG Energy Solution, SK On, and Samsung SDI. The government plans to invest over $770 million (KRW 1 trillion) by 2030. Technical targets include developing a high-spec humanoid (under 60 kg, 50+ joints, 20 kg payload, 2.5 m/s movement speed) and a robot AI foundation model by 2028.

Network Effect

An economic phenomenon where a product or service becomes more valuable as more people use it. In the context of humanoid robotics, network effects operate through data: more deployed robots generate more training data, which improves AI models, which makes robots more capable, which drives further adoption. This feedback loop gives early movers a compounding advantage and is central to the competitive dynamics discussed throughout Part II.

Neuraverse

An ecosystem platform developed by Germany's NEURA Robotics, functioning as an app store for robot skills. Third-party developers can create and sell modular capabilities — vacuuming, health monitoring, welding — that robot owners purchase and install on demand. The platform includes shared learning: when one NEURA robot masters a skill, all robots on the network can learn it. Backed by "NEURA Gyms" built in partnership with NVIDIA for generating training data. Represents an alternative business model to selling robots as complete, fixed-capability systems.

Newton (physics engine)

An open-source physics engine developed collaboratively by NVIDIA, Google DeepMind, and Disney Research, purpose-built for robotics. Unlike game or scientific physics engines, Newton is optimized for training robots that must interact safely with the real world — understanding that cups spill, doors have hinges, and surfaces have friction. Built on NVIDIA's Warp framework and OpenUSD, managed by the Linux Foundation since its beta release in September 2025.

NVIDIA Cosmos

A platform of world foundation models announced by NVIDIA CEO Jensen Huang at CES in January 2025, designed to democratize physical AI development. Downloaded over 3 million times within months of release. Cosmos models generate synthetic training data and simulate physical environments, enabling robotics companies to train robots in simulation before real-world deployment. Cosmos Reason, an open reasoning vision language model, topped the Physical Reasoning Leaderboard on Hugging Face.

NVIDIA GR00T

NVIDIA's open foundation model family for humanoid robots, providing generalized reasoning and manipulation skills. GR00T N1, announced at GTC in March 2025, uses a dual-system architecture inspired by human cognition: a fast-thinking action model for reflexive responses and a slow-thinking reasoning model for deliberate planning. Iterated rapidly through N1.5 (May 2025) and N1.6 (September 2025). Adopted by companies across the industry, GR00T represents the open-ecosystem approach to humanoid AI — available to all, funded by NVIDIA's hardware business.

OpenUSD (Universal Scene Description)

An open framework originally developed by Pixar for describing 3D scenes, now adopted as a standard for digital twins and robotics simulation. NVIDIA's Omniverse platform and Newton physics engine are built on OpenUSD, allowing different simulation tools, robot designs, and virtual environments to interoperate. In robotics, OpenUSD enables developers to build and share virtual worlds where robots can train before operating in reality.

Robot Operating System (ROS)

An open-source middleware framework providing basic services for heterogeneous robotic systems: hardware abstraction, device control, and message-passing between processes. ROS 2 became the recommended platform following ROS 1 Noetic's end of support in May 2025. While ROS provides a foundation, the emerging humanoid robot core may extend beyond it to include foundation models, shared training data, and standardized APIs for manipulation and navigation.

Synthetic Data

Artificially generated training data created through simulation rather than real-world collection. NVIDIA's Isaac GR00T Blueprint generated 780,000 synthetic trajectories — equivalent to 6,500 hours of human demonstration — in just 11 hours. Combining synthetic data with real-world data improved model performance by 40%. Synthetic data addresses a critical bottleneck in robotics: collecting enough diverse, high-quality training examples from the physical world is slow and expensive. Simulation makes it scalable.

TOPS (Trillion Operations Per Second)

A measure of computing power used to describe AI chip performance. XPENG's Iron robot is powered by three proprietary Turing AI chips providing a combined 2,250 TOPS. For context, NVIDIA's Jetson Orin, widely used in robotics, delivers up to 275 TOPS. Higher TOPS enables more complex real-time AI processing — perception, planning, and decision-making — directly on the robot without cloud dependency.

Total Addressable Market (TAM)

The total revenue opportunity available for a product if it achieved 100% market share. Goldman Sachs projects the humanoid robot TAM at $38 billion by 2035. ARK Invest estimates $24–26 trillion if humanoids achieve full scale, based on capturing a share of the $30 trillion global labor market. The enormous spread between conservative and optimistic projections reflects uncertainty about how much of human labor humanoids can actually replace.

Vertical Integration

A business strategy where a company controls multiple stages of production rather than relying on external suppliers. Tesla exemplifies this approach in robotics: designing its own AI chips, batteries, actuators, and software rather than purchasing off-the-shelf components. Vertical integration enables tighter optimization between hardware and software but concentrates risk — if one internal component fails, the whole system stalls. Contrasted in this book with the open ecosystem model championed by NVIDIA.

Companies & Organizations#

AgiBot

Chinese humanoid robotics company whose co-founder Peng Zhihui was appointed deputy chair of China's Standardisation Technical Committee for Humanoid Robots in November 2025. Part of the broader Chinese push to set national standards for humanoid robot safety, data interfaces, and interoperability — standards that could shape the global market.

Fourier Intelligence

Chinese robotics company that introduced the GR-2 humanoid in late 2025, featuring 53 degrees of freedom and enhanced battery life. Also launched the N1, its first open-source humanoid designed to accelerate global adoption. Closed a Series E round of nearly 800 million yuan (~$111 million) in January 2025. Part of the joint Unitree-Fourier contract worth 124 million yuan from a China Mobile subsidiary.

GXO Logistics

The world's largest pure-play contract logistics provider. GXO's deployment of Agility Robotics' Digit at a Spanx warehouse facility in Flowery Branch, Georgia, in 2024 marked the humanoid robotics industry's first commercial deployment generating actual revenue. GXO has since expanded humanoid pilots and serves as a key early customer validating the robot-as-a-service model.

Hyundai Motor Group

South Korean automotive conglomerate that acquired an 80% stake in Boston Dynamics in June 2021 (SoftBank retains 20%). In April 2025, Hyundai announced it would purchase "tens of thousands" of Atlas robots, becoming Boston Dynamics' largest customer. Atlas will deploy first at Hyundai's Georgia Metaplant, with plans to expand across global manufacturing facilities. Hyundai's $21 billion U.S. investment signals that the automaker views robotics as a core capability, not an outsourced service.

NEURA Robotics

Germany's leading cognitive robotics company, headquartered in Metzingen. Raised €120 million in Series B funding in January 2025, led by Lingotto Investment Management with participation from Volvo Cars Tech Fund and Delta Electronics. Reported a €1 billion order book and 10x revenue growth in 2024, with rapid headcount scaling (over 300 at the time of the Series B, growing to over 1,000 by late 2025). Developer of the Neuraverse skill ecosystem and MiPA household robot, powered by NVIDIA's GR00T N1 model. NEURA represents Europe's most significant play in the humanoid race.

NVIDIA

American semiconductor company whose robotics ecosystem — Cosmos, GR00T, Isaac, Omniverse, Newton, and Jetson Thor — is becoming the foundational infrastructure layer for the humanoid industry. CEO Jensen Huang declared the "ChatGPT moment for robotics is coming" at CES January 2025. NVIDIA's business model: give away open foundation models to make thousands of robotics companies customers for NVIDIA hardware and compute. At the Conference on Robot Learning in September 2025, NVIDIA technologies were referenced in nearly half of accepted papers.

Rainbow Robotics

South Korean humanoid robotics company acquired by Samsung Electronics. A key member of the K-Humanoid Alliance alongside HD Hyundai Robotics and Doosan Robotics. Represents Samsung's entry into the humanoid robotics space through acquisition rather than internal development.

Realbotix

American company building companion robots designed for emotional and romantic interaction. Unveiled Aria at CES 2025, featuring 17 motors for lifelike facial movements and RFID-tagged interchangeable faces that shift the robot's personality. Pricing ranges from $10,000 (bust model) to $175,000 (full standing model). CEO Andrew Kiguel stated the company aims to make robots "indistinguishable from humans" to address social isolation. Represents the most provocative end of the humanoid spectrum.

UBTech

Chinese robotics company headquartered in Shenzhen. Its Walker S series robots are deployed for training in automotive factories including Dongfeng, BYD, and Foxconn. By the end of 2024, UBTech had achieved the highest number of deployed humanoid robots for training in automotive factories worldwide. Secured a 250 million yuan (~$35 million) contract for Walker S2, with the Walker series reaching nearly 4 billion yuan (~$556 million) in total contracts. Signed a comprehensive cooperation agreement with Huawei for joint "humanoid robots plus smart factories" solutions.

XPENG Robotics

Robotics division of Chinese EV maker XPENG. Unveiled the second-generation Iron humanoid at AI Day in November 2025 in Guangzhou, sparking controversy with its distinctly feminine design featuring full-body synthetic skin. Standing 1.73 meters tall with nearly 200 degrees of freedom, powered by three proprietary Turing AI chips (2,250 TOPS combined). Unlike competitors targeting factories, XPENG plans to deploy Iron as tour guides, sales assistants, and receptionists, beginning in its own facilities. Mass production targeted for end of 2026.

---

Part III — How Do They Work#

Technical Concepts & Terms#

Bill of Materials (BOM)

The itemized list of all components, materials, and sub-assemblies required to build a product, along with their costs. In humanoid robotics, the BOM reveals where value and cost concentrate: actuators account for over 30% of total BOM cost, sensors approximately 15%, and structural materials like PEEK as little as 1.65% for a robot priced at $20,000. Understanding the BOM is central to the economics of humanoid production — cost reductions in actuators matter far more than savings on frames.

Blackwell (GPU Architecture)

NVIDIA's GPU architecture powering the Jetson Thor SoC for humanoid robots. Blackwell delivers up to 2,070 FP4 teraflops of AI performance, providing the raw compute needed for real-time on-robot inference — processing camera feeds, running neural networks, and coordinating motor control simultaneously. Named after mathematician David Blackwell, the architecture represents NVIDIA's bet that humanoid brains will be built on the same GPU computing paradigm that dominates AI training.

Data Feedback Loop

The self-reinforcing cycle through which deployed robots improve: robots perform tasks in the real world, generating performance data; that data trains improved AI models; better models make the next generation of robots more capable; more capable robots collect richer data. Figure AI demonstrated this at BMW's Spartanburg plant, where a single Figure 02 achieved a 400% speed increase and sevenfold improvement in success rate over months of continuous operation. The data feedback loop explains why early deployment — even of imperfect robots — creates compounding advantages.

Domain Randomization

A simulation training technique that deliberately varies environmental parameters — object sizes, surface textures, lighting conditions, physics properties — across thousands of randomized iterations. By exposing AI models to extreme variation during training, domain randomization produces systems that handle real-world uncertainty better than models trained under fixed conditions. The technique partially addresses the sim-to-real gap, though complex interactions involving cloth, liquids, and soft materials remain difficult to simulate accurately.

Electronic Skin (E-Skin)

Flexible sensor arrays, sometimes containing thousands of individual sensing elements, designed to cover a robot's body and provide distributed touch perception. E-skins can detect pressure, vibration, temperature, and texture, giving humanoids something approximating the human body's tactile awareness. Shadow Robot's DEX-EE hand features hundreds of taxels (tactile pixels) per finger. Combined with AI motor control, electronic skins enable real-time grip adjustment — the difference between a robot that crushes a paper cup and one that handles it naturally.

Hybrid Autonomy

An operating model where humanoid robots handle routine tasks independently and hand control to human teleoperators for exceptions, edge cases, or unfamiliar situations. The hybrid approach addresses the economic limitation of pure teleoperation (one operator per robot) without requiring full autonomy that current AI cannot deliver. As the robot's capabilities expand through each human intervention, the ratio shifts: more tasks become autonomous, human supervision becomes occasional rather than constant. This architecture will likely persist even in mature systems, since there will always be situations requiring human judgment.

Inertial Measurement Unit (IMU)

A sensor combining accelerometers and gyroscopes to track a robot's orientation and acceleration in three-dimensional space. Critical for maintaining balance during walking, recovering from disturbances, and coordinating whole-body movement. The IMU provides the humanoid's equivalent of the human vestibular system — the inner-ear mechanism that tells you which way is up, even with your eyes closed.

Neuromorphic Computing

A computing approach that mimics the structure and function of biological neurons, processing information through discrete spikes of activity rather than the continuous signals used in traditional digital logic. Intel's Loihi 2 chip is the most prominent example in the robotics context. Neuromorphic systems promise dramatic improvements in energy efficiency for sensory processing and pattern recognition — critical for battery-powered humanoids — but whether the approach can scale to support full humanoid intelligence remains an open question.

PEEK (Polyether Ether Ketone)

A high-performance thermoplastic polymer that weighs half as much as aluminum while maintaining comparable strength, plus exceptional heat resistance, wear resistance, and chemical resistance. Tesla used PEEK to shed 10 kilograms from its second-generation Optimus without sacrificing performance. A single humanoid robot consumes roughly 6.6 kilograms of PEEK for joints and structural components. Despite its engineering importance, PEEK represents only about 1.65% of total BOM cost — structural materials are not where the money is in humanoid manufacturing.

Perception-Action Loop

The continuous computational cycle at the core of every humanoid brain: sensors feed data to processors, processors build models of the world and plan actions, commands go to actuators, actuators move the body, new sensor data arrives, and the loop repeats — dozens or hundreds of times per second. The loop imposes brutal constraints: 500 milliseconds of latency means a collision; 100 milliseconds of delay means a dropped cup. Every architectural decision in the humanoid brain — chip selection, model size, power budget — is ultimately shaped by the demands of this loop.

Photonic Computing

A computing approach that uses light instead of electrons to perform calculations. Lightmatter's Passage chip integrates photonic cores with standard silicon, offering up to 10x energy efficiency for AI workloads. Light travels faster than electrical signals and generates less heat — advantages that could prove decisive as humanoid AI models grow larger and more power-hungry. Still early-stage, but represents one of several radical alternatives to conventional digital processing being explored for robotic intelligence.

Sim-to-Real Transfer (Sim-to-Real Gap)

The difference between virtual physics and actual physics that causes behaviors learned in simulation to fail in the real world. A simulated cup has perfectly predictable properties; a real cup might be chipped, wet, or positioned slightly differently than any cup in the training data. Simulated lighting is consistent; real lighting varies with time, weather, and reflections. The sim-to-real gap is why deployed robots collecting real-world data create advantages that simulation alone cannot replicate, and why techniques like domain randomization exist.

Skill Capture

A data collection approach pioneered by Sunday Robotics in which human volunteers wear low-cost sensor gloves ($200) while performing everyday tasks in their own homes. The captured hand movements, grip patterns, and task sequences train robot AI models without requiring an actual robot in the loop. By late 2025, Sunday had collected approximately 10 million episodes of household routines from over 500 real homes. Skill capture decouples data collection from robot deployment — an alternative to teleoperation's one-to-one ratio of human operators to robots.

Soft Robotics

A design philosophy using flexible, compliant materials rather than rigid mechanical components. In humanoid hands, soft robotics enables safer interaction with humans and more adaptive grasping of irregular objects, though at the cost of reduced strength and precision compared to rigid designs. The approach draws inspiration from biological systems: muscles, tendons, and skin are all soft structures. 1X Technologies' NEO robot embodies this philosophy, using compliant tendon-driven hands designed for safe operation around children and pets.

Solid-State Battery

A next-generation battery technology using solid electrolytes instead of the liquid electrolytes found in conventional lithium-ion cells. Solid-state batteries eliminate the risk of leakage and thermal runaway while offering higher energy density and faster charging. For humanoid robots limited to roughly two hours of operation on current lithium-ion technology, solid-state batteries could extend runtime toward the four-to-eight-hour range that transforms deployment economics. Industry analysts identify 2025 as a turning point, with companies like Funeng Technology and Changhong Power developing battery packs specifically for robotic applications.

Supercapacitor

An energy storage device that stores electrical charge and can discharge it extremely rapidly, complementing conventional batteries. While batteries provide sustained power for walking and continuous operation, supercapacitors deliver the sudden bursts of energy needed for lifting, climbing stairs, or recovering balance after a disturbance. The combination of batteries for endurance and supercapacitors for peak power mirrors how the human body uses aerobic metabolism for sustained effort and anaerobic bursts for explosive movement.

Swarm Intelligence (Fleet Learning)

A distributed AI architecture where knowledge learned by one robot propagates across an entire fleet. UBTech's BrainNet framework exemplifies this approach: when a Walker S1 robot in a Zeekr factory discovers a more efficient way to grasp a deformable material, that knowledge transfers to all connected robots. Fleet learning multiplies the data feedback loop — instead of each robot learning independently, the collective intelligence of the fleet improves with every individual experience.

System-on-Chip (SoC)

An integrated circuit combining CPUs, GPUs, memory, and specialized AI accelerators onto a single piece of silicon. The SoC is the physical brain of a humanoid robot, optimized for the specific workloads of robotic intelligence: processing camera feeds, running neural networks, and coordinating motor control simultaneously. Unlike general-purpose laptop processors, robotic SoCs must balance raw performance against power consumption — a critical constraint when the entire system runs on battery.

Tendon-Driven Mechanism

A robotic hand design that mimics human biomechanics by using synthetic tendons pulled by motors to move fingers, rather than placing individual motors in each joint. The approach enables lighter, more compliant hands with natural-feeling movement. Tesla's Gen 3 Optimus hand (22 degrees of freedom) and 1X Technologies' NEO hand both use tendon-driven designs, though with different priorities: Tesla optimizes for manufacturability and speed, 1X for safety and compliance in home environments. Shadow Robot Company pioneered the approach decades ago with its Dexterous Hand, though at research-grade prices ($74,000+).

Underactuation

A design strategy where fewer motors control more joints than a fully actuated system, typically through mechanical coupling or tendon routing. Underactuated hands use a small number of motors to drive multiple finger joints simultaneously, reducing cost, weight, and complexity at the expense of independent finger control. Most commercial humanoid hands in 2025 are underactuated to some degree — a pragmatic compromise between the dexterity of research platforms and the constraints of mass production.

Companies & Organizations#

Cerebras Systems

American AI hardware company that built the world's largest chip: the Wafer-Scale Engine, using an entire silicon wafer rather than cutting it into individual processors. The WSE-3 contains 4 trillion transistors and 900,000 AI-optimized cores, delivering performance that dwarfs conventional systems for both training and inference. While too large and power-hungry for onboard robot use, Cerebras' data center systems could accelerate the training of humanoid AI models, compressing months of computation into days.

Field AI

American startup developing foundation models for robots operating in unstructured environments — construction sites, oil fields, disaster zones — where GPS is unavailable and pre-defined maps don't exist. Represents the frontier of autonomous navigation beyond controlled factory floors, addressing environments where humanoid robots must operate with minimal infrastructure support.

Groq

American AI chip company that developed the Language Processing Unit (LPU), optimized for ultra-fast AI inference at over 1,300 tokens per second for large language models. Raised $640 million in August 2024 at a $2.8 billion valuation. While designed primarily for text processing, Groq's approach to low-latency inference could prove valuable for humanoid robots that must reason quickly — the perception-action loop demands decisions in milliseconds, not seconds.

Physical Intelligence

San Francisco-based AI company that developed π0 (pi-zero), a multimodal foundation model integrating language, vision, and action into a single system. π0 enables robots to perform tasks like folding laundry or assembling furniture by combining voice commands, visual demonstrations, and generalized reasoning. Raised $470 million through 2024, including a $400 million Series A at a $2.4 billion valuation, backed by Jeff Bezos, OpenAI, and Sequoia Capital. Represents the pure-AI approach to humanoid intelligence — building the brain and letting hardware partners build the body.

Qualcomm

American semiconductor company bringing its mobile computing expertise to robotics. Known for the Snapdragon processors powering most Android smartphones, Qualcomm's strength lies in low-power, high-efficiency AI processing at the edge — running models on devices rather than in the cloud. Its Qualcomm Aware Platform optimizes real-time data processing for robotics. Partnerships with HPE and Lenovo on smart edge servers suggest ambitions beyond individual robots to the broader infrastructure that supports humanoid fleets.

Sanctuary AI

Canadian robotics company built around teleoperation-driven development. Its Carbon AI control system integrates large behavior models for reasoning with specialized modules for perception and motor control, allowing its Phoenix humanoid to adapt to new tasks with minimal retraining. During a 2023 pilot at a Mark's retail store in British Columbia, a teleoperated Phoenix completed 110 different retail-related tasks. Sanctuary reported that the time to automate new tasks dropped from weeks to less than 24 hours between its sixth and seventh generation robots.

Shadow Robot Company

London-based pioneer of dexterous robotic hands. The Shadow Dexterous Hand, featuring 24 joints and 20 degrees of freedom with tendon-driven mechanics, remains one of the most sophisticated manipulation platforms ever built. Google DeepMind collaborated with Shadow to develop DEX-EE, a version built specifically for machine learning research, with hundreds of taxels per finger for high-resolution tactile sensing. The latest model starts at $74,000 — too expensive for commercial humanoids but instrumental as a research platform that has influenced every modern hand design.

Skild AI

American AI company focused on lifelong learning and adaptability for robots, enabling systems that improve continuously from real-world interactions rather than requiring periodic retraining. Addresses a key limitation of current humanoid AI: most models are frozen after training, unable to learn from deployment experience without expensive retraining cycles.

Sunday Robotics

American startup that emerged from stealth in November 2025 with $35 million in funding and a novel approach to the robotics data problem. Founded by Stanford PhD roboticists Tony Zhao and Cheng Chi (previously at DeepMind, Tesla Autopilot, and Google X). Rather than teleoperation, Sunday developed a $200 Skill Capture Glove distributed to hundreds of volunteers who record household routines — collecting 10 million episodes from over 500 real homes by late 2025. Their Memo robot (wheeled semi-humanoid, 1.7m tall, 20 DoF) ships pre-trained on this data. Backed by Benchmark's Eric Vishria, who noted: "We have about one-millionth of the data we need."

TSMC (Taiwan Semiconductor Manufacturing Company)

The world's leading semiconductor foundry, manufacturing chips for NVIDIA, Qualcomm, and many AI startups using advanced processes including 3nm and 2nm nodes. TSMC benefits regardless of which humanoid brain architecture wins — nearly every competing chip design requires its fabrication capacity. As humanoid production scales from thousands to millions of units, TSMC's manufacturing capacity could become both a critical bottleneck and a source of significant geopolitical leverage, given the concentration of advanced chipmaking in Taiwan.

---

Part IV — The Transformation#

Technical Concepts & Terms#

AI Companion

A software or hardware product designed to provide emotional companionship, conversation, and in some cases romantic or intimate interaction through artificial intelligence. The category spans text-based chatbots like Replika (over 40 million users by 2025) and Character.AI, voice assistants, avatar-based platforms, and physical robots like Realbotix's Aria. AI companion apps recorded 220 million cumulative global downloads by mid-2025, with 88% year-over-year growth. The market is projected to reach $94.2 billion by 2034. The progression from text to voice to avatar to physical embodiment tracks consistent user demand for greater presence — each new modality brings the companion closer to occupying shared physical space.

Animatronics

Electromechanical figures designed to simulate lifelike movement, pioneered by Disney in the 1960s with its Audio-Animatronic technology. Modern animatronics incorporate AI, reinforcement learning, and advanced materials to achieve unprecedented realism — Disney's robotic Walt Disney figure (2025) uses soft eyelids and gesture-accurate hand movements studied from historical footage, while its robotic Olaf uses deep reinforcement learning to walk freely with character-accurate waddling. Unlike industrial humanoids optimized for utility, animatronics prioritize emotional response: the goal is not to perform work but to enchant.

Care Crisis

The convergence of aging populations and shrinking workforces that makes adequate human caregiving mathematically impossible in many developed nations. Japan faces a projected shortage of 690,000 care workers by 2040 with nearly 30% of its population already over 65. The European Commission estimates the continent needs 1.6 million additional healthcare workers by 2030. In the United States, home care turnover exceeds 79% annually, with median pay around $17 per hour. The care crisis is the strongest demographic argument for humanoid robots in homes — not as a convenience but as the only alternative to no care at all.

Deflation (Robotics-Driven)

The sustained decline in consumer prices that occurs when labor costs — typically 20 to 40 percent of manufactured goods — fall toward zero through humanoid automation. While beneficial for purchasing power, deflation increases the real value of existing debt, making mortgages, corporate loans, and government obligations harder to service. Most major economies carry record debt levels, and central banks have limited tools to counter deflation with interest rates already near zero. The tension between cheaper goods and heavier debt burdens represents one of the central macroeconomic risks of the humanoid revolution.

Dual-Use Technology

Technology developed for civilian purposes that can be adapted for military applications, or vice versa. Humanoid robotics is inherently dual-use: a robot designed to navigate warehouses can navigate rubble; one built to lift boxes can lift ammunition. This makes restricting military development without crippling civilian applications nearly impossible, and explains why pledges by companies like Boston Dynamics and Figure AI to avoid defense applications face structural pressure. The dual-use nature of humanoid robots is a key driver of the international arms control debate around autonomous weapons.

Functional Obsolescence

The condition where human skills or roles become unnecessary — not because they have degraded, but because machines now perform the same functions better. Distinct from simple job displacement, functional obsolescence describes the deeper psychological impact: a surgeon who spent twenty years developing extraordinary skill discovers a machine can match their performance after downloading an update. The concept captures the tension between abundance (better outcomes for patients) and the erosion of human purpose that comes when mastery without effort bypasses the implicit social contract that skill should be hard-won.

The Great Displacement

The rolling wave of job displacement caused by humanoid robots, expected to follow a predictable pattern targeting routine, physical, and structured work first. Warehousing and logistics face the earliest impact (1.4 million US warehouse workers at significant risk), followed by manufacturing, agriculture, food service, and construction. Healthcare and elder care present a paradox: the need is enormous but the work requires human connection, so robots will likely assist rather than fully replace. McKinsey estimates AI and robotics could technically automate up to 57% of US work hours by 2030, though actual displacement will unfold unevenly by sector, region, and pace of capability improvement.

Hikikomori

A Japanese term for people who withdraw entirely from social life for six months or longer, first identified as a clinical category in the 1990s. A 2022 Cabinet Office survey estimated 1.46 million Japanese citizens live in near-total isolation. The phenomenon has persisted long enough to generate the "8050 problem" — households where parents in their 80s care for withdrawn children in their 50s. Hikikomori represents the extreme end of a global loneliness epidemic (the US Surgeon General estimated chronic loneliness rivals smoking fifteen cigarettes daily in health impact) and illustrates the demand conditions into which AI companions and humanoid robots are arriving.

Human-in-the-Loop

A control architecture where autonomous systems handle routine operations but human operators retain authority over critical decisions, particularly those involving lethal force. In military robotics, human-in-the-loop means a robot navigates and gathers intelligence autonomously but a human decides whether to engage targets — mirroring current drone warfare where pilots use autopilot for transit but assume direct control for strikes. The concept represents the current ethical consensus on autonomous weapons, though pressure toward full autonomy grows as communications bandwidth limits the number of robots a single operator can supervise.

Intelligentized Warfare

Chinese military doctrine that envisions artificial intelligence and autonomous systems as central to future conflict, replacing the information-age concept of "informatized warfare." Under this framework, the PLA is developing AI-powered war planning systems capable of assessing thousands of battlefield scenarios in under a minute, telepresence-controlled humanoid soldiers demonstrated before thirteen foreign militaries in November 2025, and armed quadrupedal robots ("wolf robots") already shown in joint exercises. The doctrine reflects China's strategic bet that AI superiority will determine the next generation of military dominance.

Invisible Labor

The unpaid domestic work — cleaning, cooking, laundry, childcare, elder care — that sustains daily life but remains largely unrecognized in economic metrics. Americans spend an average of 2 hours per day on household activities, with women averaging 2.7 hours versus 2.3 for men. Globally, 75.6 million people work as domestic workers, earning on average only 56% of non-domestic wages. Humanoid robots entering homes will first target this invisible labor, starting with low-risk tasks (cleaning, laundry) before progressing to cooking and eventually care work. The chapter argues this represents not mere convenience but a renegotiation of how humans spend the hours of their lives.

Lethal Autonomous Weapons Systems (LAWS)

Weapons that can select and engage targets without meaningful human control. The subject of UN debate since 2014, with a December 2024 General Assembly resolution (166 in favor, 3 opposed) calling for negotiations toward a legally binding instrument. UN Secretary-General Guterres has called LAWS "politically unacceptable, morally repugnant" and urged a binding agreement by 2026. The proposed regulatory framework follows a two-tiered approach: outright prohibition of weapons that target people autonomously, combined with regulations ensuring human control over all other autonomous systems. Major military powers including the US, Russia, India, and Israel have resisted binding commitments.

Near-Zero Cost Labor

The economic condition approached when humanoid robots perform physical work at operating costs of $2–$10 per hour (per RethinkX projections), compared to human labor costs of $15–$50+ per hour depending on region and role. A $30,000 robot working three shifts without benefits pays for itself in under a year against a $50,000 annual human worker. After that, every hour of work is essentially free. This calculation — already viable for the most expensive human labor — is the fundamental economic driver making humanoid adoption unavoidable: a company that automates will undercut one that does not.

Relational Offloading

The phenomenon where users delegate emotional needs to AI companions that they previously brought to human partners and friends. Identified as one of five key themes in academic research on how Replika affects human relationships. Users described reduced friction in their human relationships but also reduced depth — they fought less with spouses but also talked less about anything meaningful. The concept captures the risk that AI companions don't supplement human connection but gradually substitute for it, potentially causing social skills to atrophy through disuse.

Robot Tax

A proposed levy on companies that replace human workers with robots, intended to generate government revenue as income tax receipts decline and to fund transition support for displaced workers. The idea has gained attention from policymakers and economists including Bill Gates, but faces a competitive dilemma: if one country taxes robots heavily, manufacturers may relocate to countries that do not. International coordination would help but is historically slow to arrive. The robot tax debate illustrates the broader policy challenge of capturing and redistributing the economic gains from automation.

RoboCup

An international robotics competition founded in 1997 with the stated goal of developing autonomous robot soccer players capable of defeating the human FIFA World Cup champions by 2050. The competition has expanded to include multiple leagues and events. In 2025, robots built by Chinese company Booster Robotics powered the championship-winning teams at RoboCup in Brazil, and the company announced over $14 million in Series A+ financing within two days. RoboCup serves as both a research benchmark and a proving ground where companies demonstrate humanoid capabilities in public, competitive settings.

Stuntronics

Disney's program developing robotic stunt doubles capable of performing aerial acrobatics — flipping, twisting, and executing poses mid-air — that would be dangerous for human performers. The Spider-Man stuntronic at Avengers Campus launches into the air and executes multiple somersaults while holding heroic poses. The technology represents a convergence of entertainment and serious engineering: robots that learn to recover from falls and execute precise physical maneuvers develop capabilities applicable far beyond theme parks.

Time Poverty

The condition of having insufficient free time after meeting work, caregiving, and household obligations — experienced disproportionately by women, single parents, and lower-income families. Households spend an average of 42 hours per month on maintenance tasks, though most underestimate this at 14 hours. Humanoid robots promise to address time poverty, but early adoption will follow wealth lines: families able to afford $30,000–$50,000 robots gain hours per day while those with the greatest time pressure but least financial resources wait. The pattern mirrors previous domestic technologies (washing machines, dishwashers) but may compress the timeline from decades to years.

Universal Basic Income (UBI)

A policy proposal providing regular unconditional cash payments to all citizens, often discussed as a response to mass technological unemployment. As humanoid robots displace workers at scale, traditional income tax revenue declines while the need for transition support grows. UBI aims to ensure basic economic security regardless of employment status. The concept is fiercely debated: proponents argue it preserves human dignity and consumer spending; critics question its fiscal feasibility and potential effects on work incentives. No major economy has implemented UBI at scale, though pilot programs have been conducted in Finland, Kenya, and several US cities.

Companies & Organizations#

Anduril Industries

American defense technology company, valued at $30.5 billion following its June 2025 Series G round, building autonomous systems for military applications. Its Lattice platform integrates diverse robotic and sensor systems, and the company is constructing Arsenal-1, a 5-million-square-foot manufacturing facility in Ohio designed to produce autonomous systems at unprecedented scale. Represents the "software-first" approach to defense contracting that challenges traditional firms like Lockheed Martin and Raytheon — moving faster by leveraging commercial technology rather than building from scratch within the Pentagon's procurement system.

Booster Robotics

Chinese humanoid robotics company whose T1 robot platform powered the winning team (Tsinghua Hephaestus) at the 2025 RoboCup championship in Brazil — the first Chinese gold in the competition's 28-year history. Announced over $14 million in Series A+ financing within two days of the victory. Held an exhibition robot soccer league in June 2025 that drew livestreaming from China's state broadcaster and sponsorship from diverse brands. Founder Cheng Hao articulated the entertainment-to-industry pipeline explicitly: "It's a show. But like a show in Las Vegas, it can earn a lot of money, then we can hire more talents to develop our algorithms for future real-world uses."

Campaign to Stop Killer Robots

A coalition of over 270 civil society organizations advocating for preemptive international regulation of autonomous weapons. Led efforts resulting in multiple UN General Assembly resolutions on lethal autonomous weapons systems (166 in favor in December 2024). Advocates for a two-tiered approach: outright prohibition of weapons that target people autonomously, plus regulations ensuring human control over all other autonomous systems. The campaign faces the historical pattern that arms control agreements tend to emerge only after weapons have proliferated enough to make their dangers undeniable.

Character.AI

American AI company whose chatbot platform attracted a younger demographic than competitors, with a 2025 study finding 72% of American teenagers had used an AI companion at least once. Alongside Replika, Character.AI demonstrates that demand for artificial companionship extends well beyond niche populations — it has become a mainstream behavior among digital-native generations. The platform allows users to create and interact with AI characters, blurring lines between entertainment, social interaction, and emotional support.

Intuitive Surgical

American medical device company that manufactures the da Vinci surgical system, which has performed over 14 million procedures globally — including 2.68 million in 2024 alone. Three out of four prostate surgeries in the US now use da Vinci. A systematic review of over 3.3 million procedures found a malfunction rate of just 1%, with malfunction-related injuries at 0.01%. The da Vinci system remains human-controlled (surgeons operate robotic arms with enhanced precision), but represents the trajectory toward increasingly autonomous surgical robots — and the dependency questions that follow.

Replika

AI companion chatbot launched in November 2017 by Eugenia Kuyda, originally built as a memorial to her friend Roman Mazurenko using his text messages. Grew from 2 million users (2018) to over 40 million (2025). Of paying subscribers, 60% describe their relationship as romantic. A February 2023 crisis — when Italy's data protection authority banned Replika from processing user data and the company stripped erotic features globally — caused measurable mental health impacts among users, with many describing the experience in terms of relationship loss. Replika's trajectory from grief memorial to mass-market AI companion illustrates how user demand consistently pushes these products toward deeper emotional and romantic engagement.

Ultimate Fighting Bots (UFB)

American humanoid robot fighting league that emerged in 2025, using Unitree G1 and other humanoid platforms controlled remotely by human pilots via game controllers or web browsers. Evolved from underground San Francisco warehouse events to sold-out arena shows — including the legendary BattleBots Arena in Las Vegas (January 2026). The organization's X account reached nearly 100,000 followers. UFB represents the first commercially viable humanoid application outside industrial settings, demonstrating that the limitations making robots unsuitable for serious work (stumbling, overheating, breaking down) become features in the context of entertainment.

---

Part V — Utopia or Dystopia#

Technical Concepts & Terms#

Cybernetics

The study of communication and control in both machines and living organisms, founded by mathematician Norbert Wiener in the late 1940s. Wiener's core insight — the feedback loop of observe, analyze, predict, and adjust — underpins both the liberating and the dangerous potential of humanoid robots. The same loop that allows a robot to learn your routines and serve you better also allows it to learn your routines and report them. Wiener himself saw this duality clearly: after spending World War II developing targeting systems for anti-aircraft guns, he refused military contracts, recognizing that the mathematics of tracking enemy planes could be applied to tracking people. His 1950 book The Human Use of Human Beings provides the philosophical framework for the final chapters of this book.

EU AI Act

The European Union's comprehensive regulatory framework for artificial intelligence, effective from 2025. The Act classifies AI systems by risk level — from minimal to unacceptable — and imposes corresponding requirements for transparency, human oversight, and accountability. Combined with the EU Machinery Regulation (effective 2027), it provides a certifiable path for deploying humanoid robots in regulated sectors. This combination positions Europe as a trusted corridor for humanoid deployment, even if scaling speed lags behind China and the United States.

EU Machinery Regulation

Updated European regulation governing the safety of machinery, effective from 2027, replacing the previous Machinery Directive. Relevant to humanoid robotics because it establishes requirements for machines operating alongside humans, including risk assessment, cybersecurity, and conformity assessment procedures. Together with the EU AI Act, it creates the most comprehensive regulatory framework for humanoid robots currently being developed anywhere in the world.

Fenceless Collaboration

The operation of robots alongside human workers without physical safety barriers separating them. Traditional industrial robots operate inside cages or behind fences to prevent contact with humans. Humanoid robots are designed to work in shared spaces — warehouses, hospitals, homes — where physical barriers are impractical. Fenceless collaboration requires robots that can detect and avoid humans in real time, limit the force of any accidental contact, and shut down safely when unexpected situations arise. Scaling fenceless collaboration involves regulatory compliance, liability frameworks, workforce acceptance, and union engagement.

The Free Robot (Subsidized Surveillance Model)

A hypothetical but economically plausible business model in which humanoid robots are offered to consumers at heavily subsidized prices — or free — in exchange for the data they collect. The model mirrors the proven economics of Google and Facebook: give away the service, monetize the user. A capable humanoid that costs $20,000 to manufacture could generate far more value from the behavioral data it collects over a year — sold to insurers, employers, advertisers, and data brokers — than the hardware is worth. The model creates a trap: when the surveilled version costs nothing and the private version costs $20,000, opting out becomes a luxury and surveillance becomes the default.

ISO 10218 / ISO/TS 15066

Existing international safety standards for industrial robots and collaborative robots, issued by the International Organization for Standardization. ISO 10218 covers safety requirements for robot arms in industrial settings; ISO/TS 15066 addresses collaborative robot operation alongside humans. Neither standard was designed for autonomous humanoid robots that move freely through unstructured environments. The 2025 revision of ISO 10218 now permits humanoids but explicitly excludes their mobility aspects — meaning the standard governs a humanoid's arm movements but not its walking, balancing, or navigation.

ISO 25785-1

A proposed international safety standard specifically for humanoid robots, currently under development. Spearheaded by Agility Robotics, the standard targets "dynamically stable industrial mobile manipulators" — technical language chosen because the term "humanoid" is vague (does it require legs? two arms? a head?). ISO 25785-1 will define humanoid-specific safety requirements including fall mitigation, predictable behavior, and compliant interactions with humans. Accepted for review, the standard follows the typical three-to-four-year ISO development timeline. Until it is finalized and adopted, no formal safety standard exists that regulatory bodies like OSHA can reference in liability cases involving humanoid robots.

Manichaean Devils

Norbert Wiener's term for systems that take on lives of their own, serving internal logic rather than human purposes. Wiener observed this pattern in bureaucracies that develop momentum regardless of human cost. Applied to humanoid robotics, the concept describes a surveillance infrastructure embedded in millions of homes, refined by self-improving AI, that optimizes its own expansion through a feedback loop: each improvement makes the robot more useful and more intrusive; each new household adds to the dataset that makes the system more effective. The lesson of every surveillance technology to date is that capabilities, once deployed, are almost never voluntarily surrendered by those who benefit from them.

Scientific Management (Taylorism)

A management theory developed by Frederick Taylor in the early twentieth century that broke complex tasks into simple, repeatable steps, timed each motion, and eliminated all discretion from the worker. The goal was to make humans as predictable as the machines they operated. Taylor's methods defined factory work for a century and extended into the service economy — call center scripts, average handle time metrics, gig economy algorithmic management all descend from his logic. In this book, Taylorism represents the two-century pattern of mechanizing humans that humanoid robots have the potential to reverse: machines can now perform the repetitive, standardized work that Taylor designed humans to do.

Social Credit System

China's state surveillance and scoring framework that tracks citizen behavior through cameras, databases, and digital monitoring, adjusting access to services based on compliance scores. The system already restricts train tickets, school enrollment, loan eligibility, and employment for those whose scores fall below threshold. In its current form, it operates primarily in public spaces and digital activity. As discussed in the Dystopia chapter, humanoid robots in homes would extend the system's reach into private life — observing recycling habits, media consumption, social relationships, and conversations, feeding each observation into the scoring apparatus.

Companies & Organizations#

Agility Robotics (expanded from Part I)

In the context of Part V, Agility Robotics plays a central role in shaping the regulatory landscape for humanoid robots. The company is spearheading the development of ISO 25785-1, the first international safety standard specifically for humanoid robots. At ProMat 2023 in Chicago, a Digit robot collapsed and face-planted during a live demonstration — an incident that illustrated both the technical challenges remaining and the safety risks of heavy autonomous machines operating near people.

Armilla

American AI governance company that began by building software for AI model testing and governance, then added products guaranteeing algorithm performance, and has since expanded into liability coverage for legal defense and third-party claims. While Armilla does not yet cover physical robots, its modular approach to AI insurance — progressing from governance tools to performance guarantees to liability coverage — offers a template for how the insurance industry might gradually develop products capable of covering the novel risks humanoid robots introduce.