Runner-up · Robotics · Battery manufacturing
Spirit AI put dozens of humanoids onto production lines.
Spirit AI has deployed tens of Moz1 wheeled humanoid robots on production lines at battery maker CATL and JD.com. The company trains its models primarily on real-world motion data collected by about 1,000 contractors rather than relying mainly on simulation.
Spirit AI says its focus is shifting from isolated robot skills toward continuous workflows across larger spaces. Force control is intended to limit unsafe motion, while human demonstrations provide data for manipulation and navigation tasks that remain difficult to reproduce reliably in simulation.
The company has not disclosed exact fleet size, plant locations, assigned tasks, uptime, intervention rates or production output. Its reported 90% success rate applies to simple tasks in structured living-room environments, not the factory deployments, so it should not be read as line-performance evidence.
Read original story ↗01
What changed
Spirit AI had previously demonstrated Moz1 and released robot-model research. What changed in the September 18 reporting is evidence of named industrial use: tens of robots are now deployed on CATL and JD.com production lines, backed by a large real-world data-collection operation.
02
Why it matters
Industrial humanoids need more than dexterous demonstrations. They need task data, force limits, repeatable handoffs and fleet operations inside production environments where a failure can stop a line. Spirit AI’s approach links model training directly to physical work and puts dozens of robots with two named operators, moving the evidence beyond a lab. The missing operating metrics remain decisive: without exact tasks, uptime, interventions and units per hour, the deployment cannot yet be compared with conventional automation. For manufacturers, this is a credible early production foothold—and a reminder that data collection and safety engineering may scale before general-purpose autonomy does.
03
What to watch
Watch named tasks, exact fleet counts, autonomous operating hours, human interventions and throughput at CATL or JD.com. Expansion from tens to hundreds of units with repeat orders would separate a production learning program from a durable automation system.
Why it was a runner-up
The deployment is real and tied to named operators, but fleet counts, production tasks, uptime and throughput results remain undisclosed and the most ambitious capability claims are forward-looking.
Impact: 84/100 · Confidence: 84/100
Reuters observed Spirit AI’s data-training operation and directly interviewed its founder, while the company’s official site and model repository confirm its technology program. Confidence is moderated because CATL and JD.com have not published task-level operating results.
Read this edition’s Daily Signal →