Ex-Meta researchers unveil Isaac 0.5, a versatile visual AI for factories
Former Meta AI scientists have launched Isaac 0.5, an open-weight vision model designed to help robots navigate and act in industrial settings.
Perceptron, a startup created by former Meta FAIR researchers Armen Aghajanyan and Akshat Shrivastava, introduced its latest vision model, Isaac 0.5, this week. The open-weight AI system is built to enable robots to see, interpret and act within complex physical environments such as warehouses and factory floors. To train the model, the company fed it roughly one million hours of diverse video, including ego-view footage captured from a person’s perspective, to teach it to recognize settings, objects and motions.
Isaac 0.5 also extracts visual intelligence from robot-recorded video streams, offering flexible, general-purpose capabilities beyond narrow, task-specific solutions. The launch follows a $21 million financing round led by Bessemer Venture Partners, positioning Perceptron to market its technology to manufacturers, logistics providers, security firms, mobility services and media companies. The founders claim the model fills a gap between heavyweight foundation models and limited perception tools, promising broader automation potential.
Why it matters
The model could broaden robot automation in factories and warehouses, affecting manufacturing efficiency and job dynamics.
In this story
