China debuts Tiangong, an electric humanoid that runs like a human

Getting a heavy metal robot to run like a human—feet completely leaving the ground—has long been one of the toughest engineering hurdles in robotics, usually reserved for expensive, clunky hydraulic systems.

Beijing Humanoid Robot Innovation Center just unveiled Tiangong, a full-sized humanoid robot powered entirely by electric drives that can run naturally at a stable speed of 6 km/h.

Ditching hydraulics for pure electric drive

During its live demo, the 1.63-meter-tall, 43-kg robot jogged smoothly onto the stage with rhythmic arm swings, turned around at the edge of the audience, and trotted back to the center.

Each of its legs features 6 degrees of freedom, while each arm has 3. That light build helps manage the immense mechanical stress of running.

According to general manager Xiong Youjun, running requires a massive burst of energy at takeoff, generating joint torque three to four times higher than walking. The robot also has to absorb heavy impact upon landing without losing balance.

Tiangong pulls this off using on-board compute capable of 550 trillion operations per second, paired with 3D vision sensors, inertial measurement units, and high-precision six-axis force sensors.

Even in blind conditions without camera input, the center claims Tiangong can navigate slopes, climb stairs, and adjust its footing instantly if it stumbles or steps into empty space.

Learning to run by copying humans

Instead of relying solely on traditional model predictive control or pure reinforcement learning, Tiangong uses a new predictive reinforcement imitation learning method with state memory.

The team fed the robot extensive motion capture and video data of humans running. Tiangong extracts joint positions and velocities from human movement and essentially learns to imitate human gait dynamics, making its running look natural while staying resilient on rough terrain.

The developers also pitched Tiangong as an open base platform, offering open communication interfaces and modular expansion so other robotics firms can build on top of it without reinventing the wheel.

Why it matters

We think Tiangong’s real achievement is pulling off dynamic running on a purely electric drivetrain. Hydraulics are notoriously expensive, messy, and hard to maintain, so proving that electric motors can handle the peak torques of running is a huge nod toward scalable commercial production.

Positioning it as an open hardware and software base platform is a smart play. It allows smaller startups and university labs to bypass the expensive hardware grind and focus purely on high-level AI and task planning.

That said, Tiangong is clearly still a locomotion prototype. A jog of 6 km/h is essentially an easy human warm-up pace, and with only three joints per arm, it currently lacks the manipulation capability needed for real-world labor. The real test will be whether this open platform can stay stable once it has to carry payloads and use hands.