Perceptron, a startup founded by two former Meta AI researchers, has released Isaac 0.5, an open-weight vision model intended for robots in factories and warehouses. The company says the model can interpret scenes, plan actions and help machines navigate rather than handling only one narrow perception task.
A package-sorting robot illustrates the intended workflow. It must read a label, locate boxes, choose an order for moving them and execute each pickup. Perceptron wants one general system to connect those steps and also extract useful information from video recorded by the robot.
The company says Isaac 0.5 was trained on one million hours of general video, supplemented by first-person recordings and demonstrations of human movement. Perceptron does not disclose the sources of that training material, which limits outside scrutiny of its coverage and rights.
Open weights allow developers to inspect and adapt the model parameters, but they do not independently verify the startup’s performance claims. Perceptron is targeting manufacturing, logistics, security, mobility and media customers. The key test will be whether the model remains reliable across changing physical environments, where errors carry consequences beyond a wrong answer on a screen.