Nomagic deploys vision-language-action model for warehouse robots

Nomagic announced deployment of its first vision-language-action model to paying customers. The model allows robots to perceive objects, interpret text instructions, and act in warehouse settings. The company said the system cut robot errors requiring human intervention by half at logistics customers.

It attributed the gains to targeting common edge cases where robots previously needed assistance.

Background on the approach Nomagic operates with headquarters in Warsaw, Poland, and Sandy Springs, Georgia. Earlier this year it established an AI research lab to develop models that achieve high accuracy on specific tasks from the outset. The company trains its models on data from robots already deployed with customers, generating millions of successful package picks each month.

Officials said this real-world data provides an advantage over simulation or remote-control training methods.

Performance and limitations Nomagic stated its vision-language-action models are not yet at 99.9 percent success rates on their own. The company uses its existing robotics software as a safety harness to maintain reliability while the AI improves. Officials said stronger future models may reduce reliance on the harness over time.

The company recently received an industry award for a shoebox-picking device that handles two-piece boxes without lids falling off.