Perceptron, a startup founded by two ex-Meta researchers, has launched Isaac 0.5 to bring visual AI to factory floors.
Founded in November 2024, the firm builds frontier vision models designed to help machines interact with physical environments. The new release allows vision-guided robots to navigate complex spaces like warehouses and extract visual intelligence from recorded footage.
Isaac 0.5 is available as an open-weight model, meaning its parameters and training materials are open for inspection.
The company raised $21 million in a round led by Bessemer Venture Partners. Co-founders Armen Aghajanyan and Akshat Shrivastava previously worked for Meta’s Fundamental AI Research division.
The pair describe their software as the future of industrial automated deployment.
“Physical AI today forces a false choice: generalist foundation models that need multiple dedicated cloud GPUs for every instance, or narrow models that handle perception or control, but never both,” the company states.
Aghajanyan and Shrivastava say their tool differs from existing models because it is general-purpose rather than built for one specific, repetitive task. The system adapts to the particular environment or situation it encounters.
In an interview, Shrivastava asked me to consider the steps involved in organizing boxes. “Imagine there’s a robot being deployed to sort packages right now. What are the tasks it would need to do?”
Even this simple task requires many steps. A robot must read the label on a package, perform spatial analysis to understand box locations, and decide which one to pick up. If handling a series of boxes, it must plan the order of collection.
Perceptron’s software helps robots manage each step. While industry software exists for most of these tasks, few programs offer this level of flexibility.
Where does the data for this algorithmic alchemy come from?
Models like Isaac 0.5 learn operational skills by ingesting vast amounts of video training data. The new model was fed a million hours of general video to teach the algorithm to identify settings, visuals, and scenarios.
The company relied heavily on ego video, typically captured through a GoPro or wearable camera from the perspective of a person completing a physical task. It also used UMI video, which records repetitive human actions to teach AI systems movements.
While Perceptron does not disclose the sources of its training data, Shrivastava said the firm has “internally built petabyte-scale datasets that span across modalities, whether it’s images, text, video, etc. all the way through robotic trajectories.”
Perceptron believes it is well-positioned to lead the wave of automation in warehouses. The startup plans to market its software to various vendors, potentially integrating its intelligence layer into a broad array of industries.
Those sectors include manufacturing, logistics and warehousing, security, mobility, as well as media and entertainment.
“Nothing like this really exists out there,” said Aghajanyan. “We’re really excited about it.”
What it means
Manufacturers and logistics managers no longer have to choose between expensive, generalist cloud models or narrow, single-purpose tools. Isaac 0.5 offers a flexible alternative that can adapt to different factory layouts and tasks without requiring a complete system overhaul.
