A humanoid robot built by Chinese firm Unitree Robotics has reportedly reached a peak running speed of 10 meters per second, roughly 22.4 miles per hour, during controlled testing. If accurate, that puts a two-legged machine within arm’s reach of the average speed Usain Bolt maintained during his world-record 100-meter dash. The milestone arrives just days before the 2026 Humanoid Robot Half-Marathon in Beijing, where a growing field of bipedal machines will test whether raw sprint ability can translate into sustained performance over 21 kilometers.
The speed claim and what’s behind it
The 10 m/s figure comes directly from Unitree, the Hangzhou-based company best known for its quadruped robots, and was published through a Hangzhou municipal report republishing China Daily coverage in April 2026. According to that source, engineers achieved the speed by stripping the H1 of its head and hand components to cut drag and weight. Even in that reduced configuration, the robot still weighs 62 kg (about 137 pounds) and runs on legs measuring 0.8 meters long, proportions that directly affect stride frequency and ground contact time for any bipedal runner, biological or built.
For perspective, Bolt’s 9.58-second 100-meter world record works out to an average of 10.44 m/s across the full distance. His peak speed during that 2009 race in Berlin was estimated above 12 m/s. So the H1’s burst is closer to Bolt’s sustained average than to his top gear. Still, no other humanoid robot has publicly demonstrated bipedal running at this pace, which makes the number noteworthy regardless of the caveats attached to it.
And there are caveats. No independent lab or third-party measurement body has confirmed the figure. No video with embedded telemetry has surfaced. The data trail runs from Unitree through state-affiliated media and stops there. That does not make the claim false, but it does mean it carries the weight of a manufacturer’s announcement rather than a verified benchmark.
Sprinting vs. surviving a half-marathon
The gap between a short burst and a sustained race effort is where the real engineering challenge lives. The Beijing events page lists the 2026 Humanoid Robot Half-Marathon for April 19, 2026, with a 7:30 a.m. start and a course routed through Beijing E-Town. The race follows last year’s inaugural edition, where the winning robot, known as Tiangong Ultra (also called Sky Project Ultra), crossed the finish line in 2 hours, 40 minutes, and 42 seconds. Race rules allowed battery swap pit stops, functioning like aid stations for human runners but designed to keep machines powered through the full course.
That 2:40:42 finish translates to an average pace well below 5 mph. For comparison, the human half-marathon world record stands at 57 minutes and 31 seconds, set by Jacob Kiplimo in 2021. Current robots are more than two and a half times slower than the fastest humans over the same distance, even though mechanical joints do not fatigue the way muscles do. The bottleneck is not leg speed. It is energy storage, thermal management, and the control algorithms needed to keep a tall, narrow machine balanced stride after stride over uneven terrain.
The H1’s 10 m/s test was a short burst, not a sustained effort. How long the robot can hold anything close to that pace, how quickly its battery drains at full output, and what thermal limits its motors face under load are all unanswered questions. Until Unitree or another manufacturer publishes endurance data over meaningful distances, the sprint number and the marathon number exist in separate worlds.
The stripped-down problem
Removing the H1’s head and hands also complicates the comparison. A lighter, more aerodynamic frame naturally runs faster, but it is also less “humanoid” in any functional sense. Arms and hands contribute to balance during bipedal locomotion. A head typically houses sensors critical for perception and navigation. Whether race organizers would allow a stripped-down configuration in competition is unclear. The 2026 event listing provides logistics but does not specify technical regulations for participating robots.
Neither Unitree nor the Beijing organizers have publicly confirmed an H1 entry for the April 19 race. If the robot does compete, it would face conditions far removed from a controlled lab: outdoor surfaces, variable weather, turns, and the need to run alongside other machines for 21 kilometers. Any projection from lab to street remains speculative until a robot like the H1 is seen performing at speed in those conditions.
What the April 19 race will actually tell us
The Beijing half-marathon functions as the closest thing the humanoid robotics field has to a public, timed benchmark. Last year’s race established a baseline. If a new generation of robots can significantly cut into the 2:40 mark this month, it will signal that manufacturers are beginning to solve the energy and control problems separating short sprints from long efforts. If finishing times stay roughly the same, the H1’s 10 m/s burst will look more like a specialized lab feat than a sign of closing parity with human runners.
The broader competitive landscape adds pressure. Companies including UBTECH, Figure, and others are developing humanoid platforms with locomotion ambitions of their own, though none have published comparable sprint figures. Tesla’s Optimus program has focused more on manipulation tasks than running speed. For now, Unitree’s claim puts it at the front of a narrow race, one where the finish line keeps moving.
What is clear is that the most meaningful data will not come from press releases. It will come from stopwatches and finish lines. The H1’s reported 10 m/s peak shows that humanoid robots can, under optimized conditions, reach speeds that were out of reach just a few years ago. Whether that translates to anything beyond a headline depends on what happens when the machines line up in Beijing E-Town on April 19 and the clock starts running.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
