Saturday, April 19, 2025, was a milestone for Humanoid Robotics engineering. A humanoid robot ran a half-marathon in Beijing. The robot ran 21.1 kilometers of real-world terrain, completed by a robot athlete.
Meet Tiangong 1.2 Max, developed by Beijing Humanoid Robotics, and now the first-place winner among all participating robots in the 2025 Humanoid Half Marathon. A real test of endurance and a showcase of what’s possible when robotic technology, motion capture, sensor technology and machine learning come together.
Held in Beijing’s Olympic Forest Park, the Humanoid Half Marathon brought together research teams, universities, and robotics companies from around the world to test the limits of humanoid robots in a real-world running scenario. In total, over 12 companies and institutions participated, including:
Beijing Humanoid Robotics
Tsinghua University Robotics Lab
DeepMotion AI
RoboGait Systems
Shenzhen Mechatech
Kyoto Tech Robotics
And others from across Asia, Europe, and North America
Each team entered with a humanoid robot designed to run autonomously or semi-autonomously over 21.1 kilometers. During the run, they faced changing terrain, inclines, and outdoor environmental conditions. This was a true test of stability, reliance, and control.
One of the secrets to Tiangong 1.2 Max’s success was how it learned to run.
Instead of relying solely on programmed movement patterns, the robot was trained using real human movement data, captured using an Xsens motion capture suit. The Xsens Link suit, combined with the Xsens Analyze software records biomechanics data used by the humanoid robot for machine learning.
This data can be used in a machine learning pipeline to teach Tiangong how to move like a human. Think about:
More natural gait
Improved real-time balance adjustments
Energy-efficient joint coordination
Smoother transitions
While the motion capture data powered the training, performance during the race relied heavily on real-time orientation data, this is where the Xsens MTi-630 came into play.
Mounted at the robot’s pelvis, the MTi-630 AHRS provided:
Real-time orientation
Acceleration and angular velocity data
Up to 400 Hz data update rate, enabling millisecond-level responsiveness
This sensor gave the robot the ability to react quickly to subtle shifts in balance, changes in terrain, and dynamic movement like unexpected wind.
This is a historic event for the future of robotics in human environments. Robotics is evolving, and we might live in a world where machines walk, run, and move alongside us in the near future.
Where Movella | Xsens contributes to creating this future:
Human motion capture for Robot training
Precision motion tracking for Robot stability
Apart from running a marathon, humanoid robots could support humans in urban mobility, delivery, emergency response, or even sports science, the technologies that powered Tiangong 1.2 Max are laying the foundation for a new generation of agile and reliable humanoid robots.
More information: