Perceptron, a startup co-founded by former Meta research scientists Armen Aghajanyan and Akshat Shrivastava, has launched its latest AI model, Isaac 0.5. The software is designed to enable machines to "perceive, reason and act" in industrial environments, assisting vision-guided robots in navigating complex settings such as warehouses and factory floors.
Isaac 0.5 is also being released as an open-weight model, allowing its parameters and training materials to be inspected. The creators state that the model is general-purpose and flexible, unlike existing narrow models built for specific, repetitive tasks.
The software learns operational skills by processing large amounts of video training data, including a million hours of general video and ego video, which is captured from a person's perspective while performing a physical task. Perceptron intends to market its software to various vendors across industries such as manufacturing, logistics, warehousing, security, mobility, and media and entertainment.