From Clockwork Automata to ASIMO and HUBO

The Old Dream of a Walking Machine

With Tesla's Optimus walking a factory floor, Figure's robot stepping onto a BMW line, and a machine from China's Unitree dancing on a stage filling the news almost every week since 2025, the humanoid robot can easily look like a technology that surfaced only in the past few years. Yet the effort to make a machine in human form stand and walk on two legs reaches back half a century, and to understand where today's boom truly came from, it helps to trace that long lineage first.

The desire to build a machine in our own image is in fact older than electricity or the computer, for in eighteenth-century Europe clockmakers built dolls that wrote letters and struck keys using nothing but the force of a wound spring, and Jaquet-Droz's "Writer" went so far as to dip a quill in ink and compose a sentence one character at a time. The karakuri dolls that served tea and loosed arrows in Japan around the same period were products of the same impulse, and although they possessed nothing that could be called intelligence and amounted to no more than intricate mechanisms repeating a fixed routine, people nonetheless felt dread and fascination at once before them, moved by the single fact that a human shape was moving of its own accord.

The genuinely difficult problem, however, lay elsewhere, in the act of walking itself. Rolling on wheels is relatively simple, but balancing on two legs while placing one foot ahead of the other is a fundamentally different order of task, because with every step a person briefly supports the entire weight of the body on a single foot and continually catches the instability of near-falling with the next one. Translating this ordinary motion, so familiar that we never consciously register it, into the language of a machine took humanity several decades.

Tokyo, 1973: The First Humanoid

The humanoid in the modern sense first appeared in 1973 at Japan's Waseda University, as WABOT-1, completed in the laboratory of Professor Ichiro Kato. Standing close to human height, this machine walked on two legs, gripped objects with its hands and even exchanged simple conversation in Japanese through artificial ears and a mouth, though given that a single step took more than ten seconds, the word "walking" feels rather generous in hindsight. Even so, its standing as the first full-scale intelligent humanoid to integrate legs, hands, vision, hearing and language into a single body was beyond dispute.

Waseda's effort did not end there, for the WABOT-2 unveiled in 1984 read sheet music and played an electronic organ, pressing the keys with its fingers while working the pedals with its feet. The very notion of a machine imitating art was a fairly audacious one at the time, and the trajectory that would place Japan at the center of humanoid research for decades to come took shape around this period.

Honda's Secret Project, and ASIMO

In 1986, Honda quietly began a project that only a handful of people even inside the company knew about. A carmaker had set out to build a robot that walked like a person, and the engineers, refusing to chase any other capability, concentrated on the single problem of walking, working out the physics of balance one stage at a time across the leg-only test rigs that ran from E0 through E6. At the center of it lay a concept called ZMP, the zero moment point: keep the resultant of the forces the sole presses against the ground always within the area the foot covers, and the robot will not topple, though the real difficulty lay in computing that condition in real time while controlling dozens of joints at once.

After roughly a decade of silence, the P2 unveiled in December 1996 showed a 182-centimeter machine carrying its battery and computer entirely on its back, walking without an external tether, climbing stairs and recovering its posture even when shoved, and it delivered no small shock to a field still wrestling with small leg models. That a single company had effectively built a self-contained bipedal humanoid close to finished form on its own was extraordinary in itself, and a further refined P3 followed the next year.

The ASIMO that arrived in 2000 was the culmination of all that accumulation. Lowered to around 130 centimeters so as to meet people at eye level, it turned smoothly as it walked, could break into a light run, and performed with ease such actions as climbing stairs, carrying a tray and waving a hand. Growing beyond a formalized technology demonstration into a symbol of Honda as a company, ASIMO toured fairs and stages around the world and, for the first time, etched a clear image of what a humanoid was into the public mind.

ASIMO nonetheless carried its own essential limits, for while it excelled at performing pre-designed motions gracefully within a prepared environment, it was weak at improvising against an unpredictable and unruly reality, and so Honda ultimately wound down ASIMO's standalone development in 2018 and, after a final demonstration in 2022, let it withdraw from the stage. It was a comparatively graceful close to an era.

The American Path: From a Leg Lab to Atlas

Across the Pacific over the same years, an entirely different lineage was taking root, and its starting point was MIT's Leg Laboratory. Marc Raibert, who led the lab, chose to catch balance dynamically in the very act of hopping and springing rather than holding it statically, and in 1992 he founded Boston Dynamics and carried the work out of the university, so that as funding from the military and DARPA raised four-legged robots such as BigDog, the motion-control technology accumulated there gradually migrated into the domain of two legs.

What gave this current a concrete direction was the Fukushima nuclear accident of 2011. Out of the question of what might have been possible had there been a robot able to enter a radiation-filled building in a person's place, the DARPA Robotics Challenge (DRC) was launched, and Boston Dynamics supplied Atlas as the contest's standard platform. Unveiled in 2013, Atlas was a hydraulically driven humanoid of about 1.8 meters designed with disaster-site tasks in mind, such as shutting valves, opening doors and clearing debris, and in aiming for a worker that could go where people could not, rather than a human-like stage performer, it differed from ASIMO in its very purpose from the outset.

Korea's Leap: HUBO, and the Victory of 2015

It would be no exaggeration to say that Korea's humanoid grew out of one researcher's dogged persistence, and at the center of it stood Professor Jun-Ho Oh of KAIST. Having steadily accumulated bipedal-walking technology through the KHR series in the early 2000s, his team completed HUBO, Korea's first bipedal humanoid, in 2004, and the following year unveiled Albert HUBO, which combined an Einstein-like face built by America's Hanson Robotics, a pairing of an expressive head with a body that actually walked that was rarely seen at the time.

The decisive watershed came in June 2015, at the finals of the DARPA Robotics Challenge in Pomona, California. In a contest where robots had to resolve eight tasks without human intervention, among them getting out of a vehicle, opening a door, turning a valve, cutting through a wall and climbing stairs, the world's top teams competed and several fielded machines based on Boston Dynamics' Atlas, yet the final victory belonged to Professor Oh's Team KAIST, since DRC-HUBO+ completed all eight tasks and reached the top with the fastest time.

The significance of that victory ran well beyond a mere ranking, particularly in that a Korean university team had reached the top on the most demanding real-world stage, ahead of the United States and Japan, long regarded as the home ground of humanoid research. The shrewdness of the design was striking as well, for DRC-HUBO+ moved quickly on wheels mounted at its legs while kneeling and carried out precise work on two feet once upright, switching freely between walking and rolling as the situation required, a design that grew directly out of the pragmatic judgment that a robot need not move in exactly the same way as a human.

So Ends Act One

Across the half-century that ran from WABOT to ASIMO, from MIT's Leg Laboratory to Atlas, and on to HUBO's victory, the humanoid remained for the most part within laboratories, contest grounds and exhibition halls. It was precise but costly, impressive but slow, and above all weak in the "head" that perceives its surroundings and judges for itself, so that accurately reproducing a pre-designed motion remained, in practice, the limit.

What broke that solid limit was the 2020s, in which two changes interlocked with uncanny timing. Inexpensive yet powerful electric drivetrains and high-density batteries steadily displaced the heavy, control-intensive hydraulic approach, while at the same time artificial intelligence began, for the first time, to grant robots a practical level of sight and judgment, and the many names this series will examine in turn, Tesla and Figure and Unitree among them, rose precisely on these two currents. The next part will map in earnest the present of the humanoid as it steps onto factory floors and stages at this very moment, and the lay of the land among its giants.