China’s Centaur Robot Was Built for the Jobs That Get People Killed

Steel mills hit 1,400°C. Methane fills mine shafts without warning. Radiation zones don’t negotiate. These are environments where sending a human has always been a grim cost-benefit calculation. Run Robotics, a Shanghai-based startup, just presented its answer at WAIC 2026: a “centaur” wheel-leg hybrid robot that pairs a four-wheel all-terrain chassis with a humanoid upper body capable of actual physical work. It already has commercial orders before a single production line exists, according to Global Times — worth noting that the “world’s first” label and all payload figures below reflect manufacturer-stated specs, not independent third-party validation.

The lower body rolls fast on flat ground, then shifts into leg-like articulation over stairs, ice, and rubble — think of it as the robotics equivalent of a rally car that can also rock-climb. Up top, dual arms with 30-plus free-moving joints and tactile-sensing hands manipulate valves, press buttons, and clear debris. The suspension auto-adjusts to terrain in real time.

Dynamic payload while moving: 100–120 kg (220–265 lb)

Maximum static load: 210 kg (463 lb)

Perception stack: lidar, binocular vision, and depth cameras

Explosion-proof chassis; 95%+ domestically produced components

Modular upper body with swappable arms and multi-axis force sensors

The onboard AI closes the loop from seeing to doing. It identifies pipelines, gauges, obstacles, and people, then autonomously plans routes and executes tasks — all using human-designed interfaces, no factory retrofits required. You don’t need to modify your plant. The robot adapts to your equipment, not the other way around.

China’s humanoid robot manufacturers have developed over 400 models — more than half the global total — with domestic quadruped robots accounting for nearly 70% of worldwide sales, according to China’s Ministry of Industry and Information Technology.#China has developed more than 400 complete #humanoid robot products, accounting for over h…