Perceptron releases open Isaac 0.5 embodied foundation model
Perceptron released Isaac 0.5, a 36-billion-parameter sparse multimodal model for video understanding, embodied reasoning and robot control, with weights, code and a technical report.
Why it matters
The release exposes an unusually broad training-data and robot-control stack under open weights and code, making its reproducibility and cross-embodiment claims independently testable.
Evidence notes
- Inputs can include images, video, language, robot state and previous actions; outputs include grounded visual responses, task progress and robot actions.
- Perceptron reports training across more than 35 robot systems, 100,000 hours of robot experience, one million hours of general video and three trillion multimodal tokens.
- Weights, training and inference code, the technical report and LeRobot support were released publicly.
Company context
Perceptron develops vision and embodied-AI models for robots and other physical systems. Its model stack combines video understanding, spatial grounding, task-progress reasoning and robot control, with open Isaac releases and a commercial perceptive-language API.