Perceptron, a startup founded by two former Meta AI researchers, has released Isaac 0.5, an open-weight model designed to help robots perceive their surroundings, reason about tasks, and act in industrial environments. The model is aimed at applications such as warehouses and factory floors, where machines need to combine visual understanding with physical control.
Founded in November 2024 by former Meta Fundamental AI Research scientists Armen Aghajanyan and Akshat Shrivastava, Perceptron is developing general-purpose vision and robotics models rather than systems built for a single repetitive task.
Isaac 0.5 Combines Perception, Reasoning, and Control
According to Perceptron’s official Isaac 0.5 release, the model combines multimodal video understanding, embodied reasoning, spatial grounding, task-progress estimation, and robot control in one 36-billion-parameter sparse model. It can process images, video, language instructions, robot state, and previous actions before producing text, coordinates, task-state outputs, or robot actions.
Perceptron says the model is designed to address a gap between large general-purpose AI systems that require substantial cloud computing and narrower robotics models that handle only specific perception or control tasks.
Shrivastava described package sorting as an example of the type of workflow the model is intended to handle. A robot may need to read labels, understand where boxes are positioned, decide what to pick up, and plan the order in which multiple boxes should be moved.
The company says Isaac 0.5 can support those steps within a single system instead of relying on separate models for each stage.
Model Trained on Large Video and Robotics Datasets
Perceptron trained Isaac 0.5 on more than one million hours of general video, over 100,000 hours of robot experience, and data from more than 35 robot systems. Its training also included egocentric video, which records tasks from a person’s point of view, and UMI-style data used to capture human movements for robot learning.
Shrivastava said the company has internally assembled petabyte-scale datasets spanning images, text, video, and robotic trajectories, although Perceptron has not disclosed the individual sources of that data.
The release includes open model weights, training and inference code, evaluation tools, LeRobot integration, and documentation for reproducing model interfaces. Perceptron says developers can fine-tune Isaac 0.5 as a robot policy or use its visual outputs inside other planning and control systems.
Perceptron Targets Industrial Deployment
The startup plans to market the model across manufacturing, logistics, warehousing, security, mobility, media, and entertainment.
Perceptron raised $16 million in 2024 from Bessemer Venture Partners, The Explorer Fund, and SmartGateVC, according to PitchBook. The company is also in the process of closing another funding round.
Featured image credits: Magnific.com
For more stories like it, click the +Follow button at the top of this page to follow us.