News
Warsaw's Nomagic Claims Breakthrough in Warehouse Robot Autonomy
Poland's Nomagic says it has deployed vision-language-action models with paying customers, including Zalando, cutting the need for human intervention in half. The effort is led by a former Google DeepMind star researcher.
Contents
Warsaw-based robotics company Nomagic has announced that its vision-language-action models are now running in real, paying warehouse deployments, not just in the lab. It is one of the first deployments of its kind in the world outside a research setting, and the company says it has cut by half the number of situations in which a robot gets stuck and needs human help.
What was announced
Nomagic has spent years building robots for warehouse packing and order picking, but only this year did the company launch a dedicated AI lab to train vision-language-action models, systems that combine visual perception, command understanding and the physical operation of a robotic arm. Fortune was first to report that these models have already reached production for paying customers, not just internal testing.
The key figure concerns so-called edge cases, unusual situations where earlier warehouse robots would get confused and need an operator to step in, for example with an oddly packaged product or damaged packaging. Nomagic says its new models have cut the frequency of such interventions by half in live operations.
A mastery strategy, not generality
Most robotics labs worldwide are racing to build the most general model possible, one that can handle any physical task. Nomagic deliberately chose a different path: first perfect individual tasks to the edge of reliability, and only then try to build something more universal out of those skills.
Most of our community is racing to build the most general robot brain possible. We're betting that true mastery is harder, and that it has to be earned first in real deployments - Markus Wulfmeier, chief scientist at Nomagic
The company bases its training on what it calls the Library of Chaos, a constantly growing collection of millions of real-world edge cases gathered from robots working at customer sites. According to Nomagic, it's this day-to-day work of its device fleet in warehouses, rather than synthetic data or simulations, that gives it an edge over purely research-focused labs.
The math behind 99.9 percent
Nomagic co-founder and CEO Kacper Nowicki stresses that the threshold for acceptability in the physical world is far stricter than in software. A warehouse customer doesn't forgive mistakes, because every stuck robot means a shipping delay and a cost on their end.
The bar in the physical world is high: 99.9 percent reliability isn't a marketing gimmick, it's the price of admission to the building - Kacper Nowicki, co-founder and CEO of Nomagic
The company itself admits it hasn't reached that threshold yet. Its statement makes clear that no vision-language-action models deployed at customer sites, including competitors' models, currently reach 99.9 percent standalone success. Cutting interventions by half is a step in that direction, not the final goal.
Customers and industry recognition
Among the companies using Nomagic robots is Zalando, where device fleets generate two million individual product picks a month, and Swiss electronics distributor Brack.Alltron. Brack.Alltron founder Roland Brack praised the robots' ability to work unsupervised through night and Sunday shifts.
Today we're seeing robots that truly understand their surroundings. That intelligence lets us run autonomous shifts through the nights and on Sundays - Roland Brack, founder of Brack.Alltron
In June 2026, Nomagic's Shoebox Picker robot won the International Intralogistics and Forklift Truck of the Year award for solving a notoriously difficult problem: moving two-piece shoe boxes without losing the lid.
What it means for Poland's AI scene
Nomagic is one of the few Polish startups competing not in software but in physical AI-driven robotics, a field dominated by companies from the US and China. Hiring a former Google DeepMind researcher as chief scientist in April 2026, and now publicly announcing production results, shows the Warsaw team is genuinely trying to compete with labs that have far bigger budgets. For Poland's logistics and e-commerce market, it's a signal that warehouse automation built on physical AI models is moving past the experimental stage and becoming a commercially offered service with a local technology supplier. That could speed up the adoption of similar solutions among Polish logistics operators, for whom the cost of night and weekend labor is a significant part of the economic calculation.
Sources: Nomagic AI lab claims success with AI brain for warehouse robots (fortune.com), Nomagic Hires New Chief Scientist from Google DeepMind (nomagic.ai)
