Ex-Meta scientists launch open-weight visual AI model Isaac 0.5 for industrial robots
Perceptron, founded by former Meta researchers, launched Isaac 0.5, an open-weight vision model to help robots perceive, reason and act in warehouses and factories.
The model is designed to allow vision-guided robots to “perceive, reason and act,” according to the company. It can help robots read labels, analyze spatial layouts, and plan the order for picking up packages. Perceptron says the model is general-purpose and flexible rather than built for one specific repetitive task. Isaac 0.5 is also being released as an open-weight model, meaning its parameters and training materials can be inspected by anyone.
The startup recently raised $21 million in a funding round led by Bessemer Venture Partners. It was co-founded by Armen Aghajanyan and Akshat Shrivastava, who previously worked at Meta’s Fundamental AI Research division.
Shrivastava illustrated the complexity behind a seemingly simple task such as organizing boxes, noting that a robot must first read package labels, perform spatial analysis to understand where the boxes are, and decide which box to pick up and in what order. Perceptron’s software is designed to guide robots through each of these steps. While existing software can handle many of those subtasks, the company says few programs are designed to do so flexibly.
Isaac 0.5 was trained on a million hours of general video, as well as ego video and UMI video that record repetitive human actions from a first-person perspective. Perceptron said it “internally built petabyte-scale data sets that span across modalities” but did not disclose the sources of its training data.
The company aims to market the software to vendors across industries including manufacturing, logistics and warehousing, security, mobility, and media and entertainment. Co-founder Aghajanyan said, “Nothing like this really exists out there. We’re really excited about it.”