A Polish warehouse robotics startup says its new AI model cut human intervention rates by roughly 50% in live customer deployments.
Nomagic, based in Warsaw, integrated a vision-language-action model directly into its operational robots — not a lab demo, but real customer warehouses. The company also stood up a new AI lab to drive the work, led by a researcher who previously worked at Google DeepMind. The core bet: get very good at a narrow task before chasing general-purpose capability.
That "mastery before generality" stance is worth noting. Most robotics firms chase the broadest possible demo, optimizing for press coverage over reliability. Nomagic is instead treating a 50% reduction in stall-and-ask-a-human moments as the metric that matters — which, for warehouse operators running on thin margins, it is.
Vision-language-action models are still an emerging class of technology, and a single company's self-reported intervention rate is not an independent audit. But if the numbers hold in third-party settings, Nomagic would be a rare case of a European robotics firm punching above its weight in a field dominated by Boston Dynamics, Figure, and well-funded US startups.