Perceptron launches Isaac 0.5 vision model for industrial robots
Perceptron, a startup founded in November 2024 by former Meta Fundamental AI Research scientists Armen Aghajanyan and Akshat Shrivastava, has launched Isaac 0.5, a vision model for industrial robots. The company says the open-weight release is designed to let machines “perceive, reason and act” in settings including warehouses and factory floors.
Isaac 0.5 is intended for vision-guided robots navigating complex physical environments and for extracting visual intelligence from footage recorded by those machines. Perceptron recently raised $21 million in a round led by Bessemer Venture Partners.
A general-purpose layer for physical tasks
Perceptron positions Isaac 0.5 as a general-purpose model rather than software built around one repetitive task. Its founders argue that physical AI deployments often require a choice between generalist foundation models that need multiple dedicated cloud GPUs for each instance and narrow models that perform either perception or control, but not both.
The company says its model is meant to adapt to a particular environment or situation. In package sorting, for example, a robot must read labels, identify where boxes sit in space, decide what to pick up, and determine the sequence for handling multiple items. Perceptron says Isaac 0.5 is designed to support each stage of that process.
Software already exists for many individual robotic functions. Perceptron’s claim is that a flexible model can combine those capabilities for industrial deployment without being limited to a single predefined workflow.
Video data underpins the model
Perceptron says Isaac 0.5 was trained on one million hours of general video to help it recognise settings, visual elements and scenarios. The company also used ego video, typically captured through a GoPro or wearable camera from the perspective of a person performing a physical task, and UMI video that records repetitive human movements for AI training.
The startup has not disclosed the sources of its training data. Shrivastava said Perceptron had internally built petabyte-scale datasets spanning images, text, video and robotic trajectories. Isaac 0.5 is released with inspectable parameters and training materials, allowing others to examine the model and its development resources.
Implications for industrial adopters
Perceptron plans to offer its intelligence layer to vendors serving manufacturing, logistics and warehousing, security, mobility, and media and entertainment. The immediate business implication is that organisations evaluating robotics should distinguish between specialised point solutions and models intended to support several connected perception and action steps in a changing workplace.