30 Years Ago, Robots Learned to Walk Without Falling

It walked on December 20, 1996, and the engineers in the room knew exactly how long it would last. The Honda P2 — 183 centimeters tall, 210 kilograms, thirty degrees of freedom — took its first steps at the company's research facility in Japan, and the Nickel-Zinc battery pack strapped to its frame gave it roughly fifteen minutes of charge. Fifteen minutes of walking. Then silence.

That's the number that survives every celebration of P2's achievement, and it's the number that tells you why the IEEE is set to name the machine an official Milestone in engineering history. The dedication ceremony is scheduled for April 28 at the Honda Collection Hall in Motegi, Japan — a formal recognition of an achievement that took the engineering world three decades to fully appreciate. The Milestone designation — IEEE's way of saying a piece of technology changed the world — doesn't come easy. P2 is in line for it not because robots could finally walk, but because walking, in the real world, requires more than a good algorithm. It requires power that doesn't run out in the time it takes to cross a room.

P2's development was a decade in the making. In 1986, a team of Honda engineers — Kazuo Hirai, Masato Hirose, Yuji Haikawa, and Toru Takenaka — set out to build what they called a domestic robot. The first versions were crude: the E0 could barely shuffle forward on two legs. By the early 1990s, iterations E1 through E6 brought progressive improvements in balance and gait. The P1 in 1993 was the first truly humanoid proportioned machine. Then came P2, the breakthrough.

The machine was, by the standards of the day, extraordinary. Four MicroSparc II processors ran its real-time posture control. Its thirty degrees of freedom let it climb stairs, recover from a push, and walk at two kilometers per hour — slow, but steady. Four cameras in its head handled vision processing. For the first time, a robot was truly autonomous: self-contained, untethered, capable of deciding for itself where to put each foot.

The battery didn't care about any of that.

Twenty kilograms of Nickel-Zinc cells. Fifteen minutes of life. The engineers had solved one of the hardest problems in robotics — getting a machine to walk without falling — and run straight into a wall that had nothing to do with algorithms. A robot in a controlled lab, doing a five-minute demonstration, is one thing. A robot in an actual home or factory, needing to work for more than a coffee break, was a different species of problem entirely.

Honda spent the next three decades trying to solve it. P3 followed P2 in September 1997, a refined version with improved actuators and a slightly better profile. ASIMO arrived in 2000, the machine Honda positioned as a genuine commercial platform — and by the end of its development, ASIMO could run for about an hour on a charge. One hour. Still nothing close to what a human can sustain, but progress.

The battery problem, of course, was never really solved. It was managed. Modern humanoid robots — Figure AI's Figure 01, Boston Dynamics' Atlas, the growing cohort of competitors from 1X Technologies, Apptronik, Agility Robotics — face the same wall their predecessors hit. Better motors, more efficient control software, smarter energy management. But the fundamental physics of energy density hasn't changed the way chip performance has. Endurance remains the limiting constraint: no humanoid robot has yet demonstrated reliable, multi-hour continuous operation in demanding real-world conditions at commercial scale.

The IEEE's proposed Milestone citation calls P2 "a self-contained autonomous bipedal humanoid robot capable of stable dynamic walking and stair climbing with real-time posture control, dynamic balance, gait generation, and multi-joint coordination." All of that is true. What the citation doesn't say is that the machine could do all of that for approximately as long as it takes to boil a kettle.

That tension — between what robots can do in the lab and what they can do in the world — is still the central story of humanoid robotics thirty years later. P2 showed engineers what was possible. The battery told them how far they still had to go. Both facts are the story.

Honda's Collection Hall in Motegi houses the machines that came before and after P2 — the E0 sketches, the ASIMO units, the whole long march toward a robot that could share space with humans and actually be useful. The Milestone ceremony on April 28 will celebrate an achievement that changed what engineers thought was possible. It will also be a reminder of the gap between a breakthrough demo and a machine that can function in the world it was meant to serve. That gap has narrowed. It hasn't closed.

P2 is in a museum now. The engineers who built it have mostly moved on. The battery it ran on is a historical artifact. The problem it represented — how do you make a robot that lasts? — is still waiting for someone to solve it.