The History of the Robot Industry

01Origins: the first industrial arm (1954–1961)

The robot industry doesn't start with a humanoid — it starts with an arm. In 1954, American inventor George Devol filed a patent for a reprogrammable mechanical arm he called Unimate, short for "universal automation." Devol partnered with engineer Joseph Engelberger, often called the father of robotics, to found Unimation in 1956.

Their first customer was General Motors, which installed a Unimate arm at its die-casting plant in Trenton, New Jersey, in 1961. The roughly 1,900-kilogram machine lifted hot metal parts from a die-casting machine and stacked them — work too hot and repetitive for a human line worker. It wasn't glamorous, but it was the first time a programmable machine, rather than a fixed mechanism, did physical work on a factory floor. That distinction — programmable, general-purpose, reusable across tasks — is still roughly how the industry defines a robot today.

1954

Patent filed

1961

First production install

~1,900kg

Unimate's weight

GM

First customer

02Industrial robotics goes global (1968–1980s)

Unimation's technology spread quickly once its patents proved a factory could run one profitably. Kawasaki Heavy Industries licensed it in 1968 and built Japan's first industrial robot the following year. Sweden's ABB and Germany's KUKA moved just as fast with their own designs: ABB's IRB 6, launched in 1974, was the world's first industrial robot driven entirely by a microprocessor, and KUKA's FAMULUS (1973) was among the first robots built with six electromechanically driven axes — the arrangement still standard today. Within a decade, Japan, Northern Europe, and the US were each building serious industrial-robot industries in parallel, not one exporting to the others.

Japan's contribution was less about invention and more about scale: conglomerates — Kawasaki, Yaskawa, and Fanuc (spun out of Fujitsu's computer division in 1972) — turned the robot arm into a mass-manufactured product line, and by the 1980s Japan had installed more industrial robots than the rest of the world combined, giving rise to the domestic phrase robotto ōkoku, "robot kingdom." In 1978, Hiroshi Makino's team at Yamanashi University, working with Sankyo Seiki, added a genuine Japanese invention to the mix: the SCARA arm, a four-axis design optimized for fast, precise assembly that became a global standard in electronics manufacturing. Fanuc, Yaskawa, Kawasaki, ABB, and KUKA all remain among the world's largest robot-arm makers today, and Japanese suppliers Nabtesco and Harmonic Drive Systems still dominate the precision reduction gears that let any of these arms' joints move accurately, wherever they're built. Japan's volume lead itself didn't last, though: China overtook it in 2013 to become the world's largest market for industrial robots by annual installations, and has stayed there every year since — the manufacturing-scale story moved on well before humanoids ever entered the picture.

03The humanoid dream: from WABOT to ASIMO (1973–2000)

While arms were conquering factories, a harder and far less commercially useful project was underway elsewhere: making a robot look and move like a person. Humanoid research surfaced in labs across the US, Europe, and Japan through the 1970s, but nowhere was the effort as sustained or well-funded as in Japan. Waseda University's WABOT-1, unveiled in 1973, is generally credited as the first full-scale anthropomorphic robot — it could walk on two legs, grip objects with its hands, and "see" and "converse" with primitive vision and speech systems.

Progress after that was slow and expensive. Honda began a secretive humanoid program in 1986 and spent 14 years and several prototypes — E0 through E3, then P1 through P3 — refining a stable, self-contained bipedal walk before unveiling ASIMO in 2000. ASIMO became the public face of humanoid robotics for a decade, but it was built for demonstrations, not deployment: it couldn't run at speed, climb stairs quickly, or work autonomously outside a scripted routine, and Honda quietly ended the program in 2018. Other Japanese manufacturers — Sony's QRIO, Toyota's Partner Robot — ran similar humanoid programs through the 2000s with the same basic outcome: capable demos, no deployment. The lesson the industry took, in Japan and everywhere it later tried, was blunt — walking convincingly on two legs is a genuinely hard control problem, and solving it does not automatically produce a robot that's useful for anything in particular.

1. 1973WABOT-1First full-scale anthropomorphic robot (Waseda University)
2. 1986–2000Honda's E/P series → ASIMO14 years and 8 prototypes to a stable, self-contained bipedal walk
3. 2000–2010sASIMO eraPublic face of humanoid robotics; Sony's QRIO and Toyota's Partner Robot follow
4. 2018Program endsHonda quietly retires ASIMO — capable demos never became deployment

1. 1973

   WABOT-1

   First full-scale anthropomorphic robot (Waseda University)

2. 1986–2000

   Honda's E/P series → ASIMO

   14 years and 8 prototypes to a stable, self-contained bipedal walk

3. 2000–2010s

   ASIMO era

   Public face of humanoid robotics; Sony's QRIO and Toyota's Partner Robot follow

4. 2018

   Program ends

   Honda quietly retires ASIMO — capable demos never became deployment

04Boston Dynamics and the legged-robot underground (1992–2021)

A different lineage of legged-robot research was happening in parallel, and it would eventually matter more. Marc Raibert's Leg Laboratory — first at Carnegie Mellon, then MIT — spent the 1980s building single-legged hopping machines to study balance and dynamics rather than aesthetics. Raibert spun that research out into Boston Dynamics in 1992.

For its first two decades the company worked mostly on DARPA-funded research prototypes — the quadruped BigDog (2005), built to carry military gear over rough terrain, and later Atlas — rather than a commercial product. Google's parent company acquired Boston Dynamics in 2013 as part of a short-lived robotics buying spree, sold it to SoftBank in 2017, and SoftBank in turn sold a majority stake to Hyundai Motor Group, with the deal completing in 2021. Only in the last few years, with viral videos of Atlas doing parkour and backflips, has Boston Dynamics' decades of balance-and-locomotion research become the reference point for the current generation of humanoid startups — several of which are staffed by Boston Dynamics or MIT Leg Lab alumni.

05Cobots and the warehouse automation wave (2003–2015)

Not every useful robot needs legs or a face. Two quieter developments in the 2000s did more to put robots into daily commercial use than any humanoid did: the collaborative robot arm and the warehouse mobile robot. Universal Robots, founded in Odense, Denmark in 2005, launched the UR5 in 2008 — a robot arm designed to work directly alongside people on a factory floor without a safety cage, at a price small manufacturers could afford. It created the "cobot" category that is now standard across light manufacturing.

Around the same time, Kiva Systems (founded 2003) built orange mobile robots that carried entire shelving pods to warehouse workers instead of sending workers to the shelves. Amazon bought Kiva for roughly $775 million in 2012 and rebuilt its own fulfillment centers around the idea, renaming the unit Amazon Robotics; by the mid-2020s Amazon alone had deployed over a million robots on this model. Between cobots and warehouse AMRs, robotics quietly became a mainstream logistics and manufacturing technology years before "humanoid" was a boardroom word.

DK 2008

Universal Robots' UR5

The first cobot — a robot arm safe enough to work alongside people without a safety cage.

US 2003→2012

Kiva Systems → Amazon Robotics

Mobile robots that bring the shelf to the worker instead of sending the worker to the shelf; Amazon scaled it past a million units.

06AI meets hardware: the foundation-model turn (2012–2024)

The next shift came from outside robotics entirely. The 2012 ImageNet result that kicked off the deep-learning boom had little to do with physical machines, but it changed what people building them assumed was possible. Through the 2010s, reinforcement learning let researchers train locomotion and grasping behavior in simulation rather than hand-coding it — visible in Boston Dynamics' own progress and in a wave of academic sim-to-real results.

The turn that mattered most for today's industry came from treating a robot's control policy the same way large language models treat text: as something a single big model, trained on enormous and varied data, could learn end-to-end. Google DeepMind's RT-2 (2023) showed a robot could follow instructions it had never explicitly been trained on by inheriting reasoning from a web-scale vision-language model. That kicked off a race to build "robot foundation models" — general-purpose control policies meant to transfer across many robot bodies and tasks, the way GPT transfers across writing tasks — with Physical Intelligence's π0 (2024) and Nvidia's Isaac GR00T (announced March 2024) among the most closely watched entrants. Nvidia's Jetson Thor and Isaac/Omniverse simulation stack turned this from a research bet into commodity infrastructure that any humanoid startup could buy rather than build.

07Three threads, one convergence

Zoom out and the last seventy years look less like one continuous humanoid story and more like three separate engineering lineages — arms, legs, and software — that only recently converged. Each grew for decades in relative isolation, in different countries, before today's humanoid boom pulled them together.

08The humanoid funding boom and the physical AI era (2021–2026)

Foundation models gave the industry a plausible answer to the question that had stalled ASIMO for twenty years — not just how to make a robot walk, but how to make it generally useful — and venture capital responded accordingly. Tesla's August 2021 AI Day announcement of the Optimus humanoid, more a statement of intent than a working product at the time, is usually cited as the moment "humanoid" became a mainstream investment thesis rather than a research curiosity.

Figure AI (founded 2022), 1X Technologies, Skild AI, and Physical Intelligence each raised nine- or ten-figure rounds within a few years, while Agility Robotics' Digit — one of the earliest humanoids in actual paid pilot deployment — proved the category could reach a real warehouse floor, not just a stage. China took a different path to the same moment: Unitree, better known for its quadruped "robot dogs," launched the G1 humanoid at roughly $16,000 in 2024 — a fraction of the cost of Western competitors — and became the world's top humanoid seller by units in 2025, the same year it recorded its first profit.

Europe took yet another path: Germany's Neura Robotics raised roughly $1.4 billion in a single 2026 round — one of the largest robotics funding rounds ever — betting on cognitive humanoids built for European manufacturing rather than US-style software-first scaling or Chinese-style low-cost volume. And the ownership of the industry's oldest research lineage tells its own global story: Boston Dynamics, born from an MIT lab, passed through Google, then SoftBank, and finally to Hyundai Motor Group, a South Korean company betting that manufacturing know-how plus AI software is the winning combination. By 2026, the industry looks less like a handful of research labs and more like a multi-region, capital-intensive manufacturing race, with pilot deployments running at BMW, Mercedes-Benz, Amazon, and Ford, and the open question no longer whether a humanoid can walk, but whether any of it can be built, and sold, profitably at scale.

09Timeline at a glance

10FAQ

11Where this goes next

Seventy years separate George Devol's patent filing from today's humanoid funding rounds, and for most of that span, "robot" meant a fixed arm bolted to a factory floor, not a walking, reasoning machine. What changed the trajectory wasn't a single breakthrough in one country — it was arms (US, Japan, Europe), legs (Japan, US), and software (US, Canada, UK) developed on separate continents for decades, then pulled together by borrowing the scaling logic of large AI models and pointing it at physical control. Whether that turns into a durable industry, the way industrial robot arms did in the 1970s and 80s, or into another cycle of the hype that stalled after ASIMO, is the open question this series will keep tracking.