Perceptron visual AI is moving from the research lab to the warehouse, as the startup co-founded by two former Meta Fundamental AI Research scientists launches its latest model for industrial robots and reveals a broader product portfolio than its public debut suggested.
Isaac 0.5 and the Industrial Automation Play
The company, which was founded in November 2024, unveiled Isaac 0.5 this week: a vision model designed to help robots ‘perceive, reason and act’ in industrial settings such as warehouses and factory floors. The model is being released as an open-weight system, meaning its parameters and training materials are publicly inspectable.
What sets Isaac 0.5 apart from narrower industrial models, its creators say, is flexibility. Co-founder and CTO Akshat Shrivastava offered sorting packages as a deceptively simple illustration: a robot completing that task must read a label, conduct spatial analysis, decide which box to lift, and sequence each pick in the right order. Most existing software can handle parts of that chain. Few handle it as a unified, adaptable process.
‘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. Isaac 0.5 is Perceptron’s answer to that constraint.
The model was trained on a million hours of general video to develop scene and context recognition. Perceptron also used ego video, typically captured from a wearable camera at human eye-level, and UMI video, which records repetitive human actions to teach machines motor patterns. Shrivastava described the underlying data infrastructure as ‘internally built petabyte-scale data sets that span across modalities, whether it’s images, text, video, etc. all the way through robotic trajectories,’ though the company has not disclosed its data sources.
Perceptron Visual AI: A Two-Track Product Strategy
Isaac 0.5 is only part of the picture. According to Perceptron’s own About page, the startup is simultaneously developing a closed-source flagship model called Mk1, described as a vision-language model that is ‘frontier on embodied reasoning and video understanding.’ The open-weight Isaac 0.5 and the proprietary Mk1 together form a family of models, suggesting Perceptron is pursuing both community adoption and premium commercial deployment in parallel.
Co-founder Armen Aghajanyan and Shrivastava both led multimodal research initiatives at Meta’s Fundamental AI Research (FAIR) division before founding the company. Shrivastava’s tenure at Meta spanned approximately five years, during which his work covered large language models, multimodal systems, on-device AI, audio-based understanding, and semantic parsing.
‘Nothing like this really exists out there,’ Aghajanyan said. ‘We’re really excited about it.’
Backers and the Road Ahead
The startup’s $21 million founding round was led by Bessemer Venture Partners, with Foundation Capital, S32, and SmartGateVC also participating, according to TechCrunch, citing PitchBook. The company is reportedly in the process of closing an additional round.
Perceptron is positioning its technology as an intelligence layer that can be integrated across manufacturing, logistics and warehousing, security, mobility, and media and entertainment. Rather than selling robots directly, the company is marketing its software to equipment vendors, broadening its potential reach considerably.
The closing of that next funding round, and how quickly Mk1 finds paying industrial customers, will be the two numbers worth watching.
