A Chinese Honor-built humanoid robot ran 13.1 miles in 50:26, beating the human world record by nearly 7 minutes at a Beijing event.
At a Beijing half-marathon event, Honor's humanoid robot completed the course in 50 minutes 26 seconds, surpassing the human world record of 57:20 set by Jacob Kiplimo. Over 100 humanoid robots from 76 Chinese institutions participated alongside 12,000 human runners on separate courses. The winning robot featured near-meter-long legs, advanced balance systems, and liquid cooling — and ran fully autonomously using AI for real-time pace and terrain adaptation. Just one year prior, the fastest robot at the same event took over 2.5 hours to finish.
The performance jump — from 2.5 hours to sub-51 minutes in one year — is a direct signal that real-time balance, terrain adaptation, and thermal management algorithms are compounding fast. The architecture here (AI-driven pace control, autonomous stabilization, sensor fusion for terrain) maps directly to skills relevant in robotics SDKs, edge AI inference, and embodied AI pipelines. This isn't a lab demo — it ran 13 miles in the real world, unsupervised.
If you're building on any robotics middleware (ROS2, Isaac, Boston Dynamics SDK), benchmark your current locomotion stack against the published specs of this system — specifically real-time gait adjustment latency — to identify where your stack is behind.
Tags