Ex-Meta Scientists Launch Perceptron AI, Secure $21M for Industrial Visual AI Model Isaac 0.5

Former Meta AI researchers are taking their advanced vision expertise from the digital realm to the factory floor. Perceptron AI, a startup founded by ex-Meta Fundamental AI Research (FAIR) scientists, officially launched its Isaac 0.5 visual AI model the week of August 26, 2026, coinciding with the announcement of a $21 million funding round. The move signals a significant step in bringing sophisticated AI capabilities to industrial robotics, enabling machines to better understand and interact with their physical environments.

Perceptron, headquartered in Bellevue, Washington, and founded in November 2024, aims to address a critical gap in physical AI. According to a company statement, "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.” Isaac 0.5 seeks to offer a flexible, general-purpose solution.

What is Isaac 0.5 and what does it do?

Isaac 0.5 is described as an “open-weight vision model” or an “open-weight embodied foundation model” designed to empower industrial machines to "perceive, reason and act" in settings such as warehouses and factory floors. This 36-billion-parameter model combines video understanding, embodied reasoning, and robot control, effectively helping vision-guided robots navigate and extract visual intelligence from real-time footage, as reported by AiThority on August 31, 2026.

Perceptron previously launched its Mk1 model on May 12, 2026. Akshat Shrivastava, Co-founder and CTO of Perceptron AI, stated regarding Mk1 that "Robotics is the hardest test of real-world, physical AI. It demands perception, reasoning, and action in one closed loop, under real-world conditions. Solve that and every other physical AI problem becomes tractable. That's the bar we built Mk1 to clear.” Isaac 0.5 is the latest iteration of this ambitious vision.

Who is behind Perceptron AI?

Perceptron AI was co-founded by Armen Aghajanyan, who serves as CEO, and Akshat Shrivastava, the CTO. Both previously honed their expertise at Meta's Fundamental AI Research (FAIR) division, bringing deep knowledge of advanced AI models to the industrial sector. Their venture is part of a broader trend seeing AI innovation transition from digital applications to real-world physical systems through new startups.

How was Isaac 0.5 trained?

Isaac 0.5's capabilities stem from extensive training on diverse, large-scale datasets. The model was trained on 1 million hours of general video, supplemented by "ego video" (footage captured from a person's perspective) and UMI footage (repetitive human actions). AiThority specified that Isaac 0.5 utilized "three trillion multimodal tokens" and accumulated "100,000 hours of robotics-oriented experience across more than 35 robot systems." Additionally, the training included "internally built petabyte-scale data sets" encompassing images, text, video, and robotic trajectories. However, Perceptron has, according to reports from TechCrunch and Relve, "declined to disclose its data sources" for Isaac 0.5, a point that may draw scrutiny.

What does this mean for industrial AI?

The market for industrial AI is competitive, featuring established robotics software and well-funded foundation model rivals. Despite this, Perceptron claims Isaac 0.5 outperforms leading open robot models such as Physical Intelligence's π0.5 and NVIDIA's GR00T N1.7, as reported by AiThority. This performance claim positions Isaac 0.5 as a significant contender, offering a powerful, adaptable tool for various industries.

Armen Aghajanyan highlighted the practical benefit: “Companies need a model that performs at the frontier, learns a new task quickly and adapts to their hardware. Isaac gives them a strong, open starting point, and our team is working alongside our customers to bring it into real operations.” Perceptron plans to sell its software for applications in manufacturing, logistics and warehousing, security, mobility, and even media and entertainment, integrating its AI intelligence layer across these diverse sectors.

What's next for Perceptron?

Perceptron's strategy involves actively collaborating with customers to fine-tune Isaac 0.5 to their specific cameras, robot hardware, and operational workflows. The open-weight nature of Isaac 0.5 is designed to foster transparency and encourage partnerships, allowing customers to begin with the open model, adapt it with their own data, and then work with Perceptron for final deployment. Observers anticipate future announcements regarding warehouse pilot programs, named customer partnerships, and standardized benchmarks to validate the model's performance in navigation and manipulation tasks, as reported by Relve.