Perceptron, a startup founded in November 2024 by two former Meta FAIR researchers, Armen Aghajanyan and Akshat Shrivastava, is bringing frontier vision models to industrial robotics.

The company recently launched Isaac 0.5, an open-weight model designed to let machines “perceive, reason and act” in physical settings like warehouses and factory floors.

The model helps vision-guided robots navigate complex environments and extract visual intelligence from video captured by those robots.

The founders argue that existing approaches force a false choice between generalist foundation models that require multiple dedicated cloud GPUs and narrow models that handle either perception or control but not both.

Isaac 0.

5 aims to be general-purpose yet flexible enough for specific deployment contexts, handling multi-step tasks such as reading package labels, analyzing spatial layouts, and planning picking sequences.

Training data includes a million hours of general video plus ego video (typically from GoPro or wearable cameras) and UMI video (recorded human manipulations), all assembled into internally built petabyte-scale datasets spanning images, text, video, and robotic trajectories.

Perceptron has raised $21 million in a funding round led by Bessemer Venture Partners.

The startup plans to market its software to vendors across manufacturing, logistics, warehousing, security, mobility, and media/entertainment, with Aghajanyan stating that “nothing like this really exists out there.

”