Ex-Meta scientists raise $21M to bring vision AI to the factory floor
What's the deal? PerceptronDealroom has a profile for this one. Try Dealroom →, a startup founded by two former Meta research scientists, has raised $21 million in a round led by Bessemer Venture Partners. The company builds frontier vision models that help machines interact with their physical surroundings.
Who's behind it? Co-founders Armen AghajanyanDealroom has a profile for this one. Try Dealroom → and Akshat ShrivastavaDealroom has a profile for this one. Try Dealroom → both worked at Meta's Fundamental AI Research division. They founded Perceptron in November 2024.
The product: This week the startup launched Isaac 0.5, a model it says lets machines "perceive, reason and act" in industrial settings. It helps vision-guided robots navigate warehouses and factory floors, and extracts visual intelligence from footage those robots record.
Isaac 0.5 is being released as an open-weight model, meaning its parameters and training materials can be inspected by anyone.
Why it's different: The founders say the model is general-purpose, built to adapt to its environment rather than handle one repetitive task. "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 says.
Consider sorting packages, Shrivastava said: a robot must read a label, analyse where the boxes are, decide which to pick up, and plan the order. Existing software can do most of those steps, but few programs do them flexibly.
The data: Isaac 0.5 was trained on a million hours of general video, plus "ego video" captured from a person's perspective and UMI footage that records repetitive human actions. Perceptron declined to disclose its data sources, but Shrivastava said it had "internally built petabyte-scale data sets" spanning images, text, video, and robotic trajectories.
What's the endgame? Perceptron plans to sell its software to a range of vendors, embedding its intelligence layer across manufacturing, logistics and warehousing, security, mobility, and media and entertainment.
What could go wrong? The market it targets is crowded, and Perceptron faces both established robotics software and better-funded foundation model rivals. Its refusal to name training data sources may also draw scrutiny as regulators sharpen focus on AI provenance.
The signal: AI has largely stayed in the digital realm, but a wave of startups is now pushing it into the physical world. Perceptron's raise reflects investor appetite for the intelligence layer that could automate warehouses and factory floors.
Read more: TechCrunch
Image credit: Generated with Gemini
