Unitree H1: The Humanoid Robot That Runs Faster Than Most People
Chinese robotics company Unitree's H1 humanoid robot reached a running speed of 3.3 meters per second — faster than the average human jog — using reinforcement learning trained entirely in simulation.
By Riley Cross · May 20, 2024 · 4 min read · robot-athletes
The average human jogging pace is somewhere between 2.7 and 3.5 meters per second. In early 2024, Unitree Robotics' H1 humanoid robot was clocked running at 3.3 meters per second — placing it squarely within human recreational running range and earning it recognition as the fastest full-sized bipedal robot in the world at the time of demonstration.
This is not a trivial achievement. Making a bipedal machine run at speed requires solving a set of interlocking problems in real time: balance, gait generation, terrain adaptation, and recovery from the constant minor perturbations that running introduces. The faster you go, the less time the system has to correct errors before a fall becomes unavoidable.
Unitree's Approach: Simulation-First Training
The H1 uses a reinforcement learning approach developed heavily in simulation before being transferred to the physical robot — a method called sim-to-real transfer. The robot's neural network control policy is trained in a physics simulation running millions of steps before the robot ever takes a physical step.
This approach has several advantages. You can run training at massively accelerated speeds in simulation, crash the robot thousands of times without destroying hardware, and explore a far larger range of conditions than you could manually program for. The result is a controller that has been exposed to enough variation that it handles real-world disturbances gracefully.
The H1's running demonstration shows exactly this: the robot's gait has the slightly mechanical but genuinely efficient quality of something that has been optimized rather than hand-crafted. It does not move like a human, but it covers ground.
What the H1 Is
The Unitree H1 is a 47-kilogram full-size humanoid standing approximately 1.8 meters tall. It has 19 degrees of freedom, a maximum payload capacity of around 30 kg, and a runtime of approximately 90 minutes on a single charge.
Unitree prices the H1 at around $90,000 — significantly below what Boston Dynamics, Figure AI, or Agility Robotics have indicated for their humanoid platforms. The company has positioned the H1 as a research and development platform rather than an industrial product, targeting universities, research labs, and developers building applications.
The Running Capability in Context
3.3 meters per second is impressive for a biped robot but needs context. The world record for the 100 meter sprint is just under 10.5 meters per second. Elite distance runners average around 6 meters per second over a marathon. The H1 running demonstration lasted a controlled short distance in favorable conditions.
What the demonstration proves is not that robots can outrun humans — they cannot yet — but that bipedal locomotion at human speeds is now an achievable engineering problem, and that the tools to solve it (simulation training, reinforcement learning, high-torque compact actuators) are accessible to a mid-sized company, not just Boston Dynamics-level operations.
Unitree's Broader Platform
Unitree also makes the Go2 and B2 quadruped robots, which have found commercial traction in security, inspection, and research roles. The Go2 in particular has become something of a reference platform in the quadruped research community — it is capable, reasonably priced, and well-documented enough that third-party developers have built significant toolchains around it.
The H1 represents Unitree's bet on the eventual dominance of humanoid form factors for general-purpose robotics. Whether that bet pays off depends largely on how quickly the dexterous manipulation problem — the other half of what makes humanoids useful — gets solved.