Last year at the Humanoid Robot Half-Marathon in Beijing the winning robot stumbled across the finish line in 2 hours and 40 minutes; much slower than a human. This past week, a Chinese smartphone maker’s humanoid robot ran the entire 21-kilometer, 13-mile course in just 50 minutes and 26 seconds, soundly beating the human world record of 57 minutes and 20 seconds. This is a big blinking sign: humanoid robots are here and they’re starting to get really good. Industry, governments and all those who care about the future of work should start paying attention.
The big standout news: humanoid robots’ improvement rate.
In a year, humanoid robots went from 2.4X slower than a human to significantly faster. 20 teams from 2025 became 100 teams fielding 300 robots in 2026. And where last year none of the robots ran autonomously – they were remote controlled by humans – this year 40% of the robots ran their own race.
It’s an impressive improvement in just a year.
The big question is: what does it mean for humanoid robots? And, what does it mean for us … Humans 1.0?
First off, while there were still plenty of mistakes – one robot fell and smashed itself right at the start line, others ran into vehicles or barriers – overall, the trajectory is clear. Humanoid robots are getting better, faster, more durable and they’re increasingly able to guide themselves and withstand a long, grueling race. In fact, they can do it faster than the fastest human.
What does this mean?
- Robot joints and motors are getting stronger
That might sound like something obvious, because clearly robots are built largely for repetitive work. But long hard workloads on motors and joints build up heat over time that wear down robots. The winning robot’s liquid cooling kept it going: and that’s exactly what most humanoid robot manufacturers are working on right now. - Actuator and component endurance is increasing
Long workloads aren’t just about the use. They’re also about the wear and tear of repetitive jolts. Running is the one of the best possible test of joint and actuator durability under repeated identical stress, because every stride is essentially the same motion, thousands of times, with ground reaction forces hitting knees and hips and transmitting through the entire body. - Autonomous navigation is betting better
Last year, all the robots were remote controlled. This half-marathon has sharp bends, slopes and crowded conditions outside barriers, and many of the robots guided themselves around the course. That means SLAM (simultaneous localization and mapping) and real-time obstacle avoidance under dynamic, unpredictable conditions is getting better.
Summing up: humanoid robots are approaching utility. Just last week, we had a world-first humanoid robot on an industrial-scale electronics production line and as they get more and more capable, increasing numbers of jobs and roles will open up for humanoid robots.
That said, there’s still plenty of room to go.
Humanoid robots need to continue to get much better at dexterous manipulation tasks like threading a bolt, handling fragile objects, operating tools designed for human hands and more. The fine motor control problem remains largely unsolved at commercial scale, although we’re seeing multiple examples of great robot hands coming out lately.
Also, in spite of the impressive navigation of many of the humanoid robots in the half-marathon, robots still need better physical AI and semantic understanding of environment. They need to know what to do in new situations, not just how to move through space. That’s clear based on a recent report that showed humanoids failing to complete 88% of tasks in a home setting. Generalization is important: a robot trained to run a specific course is not necessarily a robot that can handle a new factory floor layout on day one. Sim-to-real transfer remains a big physical AI research problem.
And, of course, cost is an issue.
There are multiple relatively cheap humanoids coming out of China right now, including the Unitree R1 for less than $5,000, but let’s be honest: they’re not going to work on the factory floor. Until humanoids reach price parity with industrial cobots (perhaps in the $50,000 to $80,000 range), the ROI math is harder to justify.
Completing the half-marathon proves some level of endurance, speed and navigation. It does not prove a humanoid robot can replace a human in a factory.
Let’s be honest, though: it’s a step in that direction. And it’s much better than showing dancing robots, which, despite being interesting, reveals very little of their actual utility.
Most likely, we’re three to five years away from broad deployment of humanoid robots in our physical labor workforce. Of course, that’s a blink of an eye in terms of regulation, taxation, training workforces, and figuring out what to do with jobs and salaries as AI attacks white collar work and robots take on blue collar work.
Which means us Humans 1.0 better start figuring those questions out. Because the labor disruption is real, and it is coming.
