Perceptron, a startup founded by former Meta AI researchers, has launched Isaac 0.5, a new vision model designed to help robots "perceive, reason and act" in industrial settings such as warehouses and factories. The move reflects a broader push by AI companies to extend machine intelligence beyond digital applications and into physical, real-world tasks, an area often described as "physical AI".

Founded in November 2024 by Armen Aghajanyan and Akshat Shrivastava, both formerly of Meta's Fundamental AI Research division, Perceptron recently raised $21 million in funding led by Bessemer Venture Partners. Isaac 0.5 is released as an open-weight model and is trained on around a million hours of video, including footage captured from human perspectives, to help vision-guided robots navigate complex environments flexibly rather than performing only narrow, repetitive tasks.