Perceptron

Summary

About

About Perceptron

Perceptron builds multimodal models that connect visual understanding with action in physical environments. Its Isaac family targets embodied reasoning and robot control, while Perceptron Mk1 and the company API support grounded detection, localisation, OCR and visual question answering across video and images. The software is aimed at robotics, manufacturing, logistics, security and other systems that need structured perception outputs.

Isaac 0.5 is a 36-billion-parameter sparse model trained across more than 35 robot systems, 100,000 hours of robot experience and one million hours of general video. Perceptron released model weights, training and inference code, and a technical report in August 2026. The model accepts images, video, language, robot state and prior actions, and can produce grounded visual outputs or robot actions.

PRICE
Range 1Y
Return
High
Low
Updated
DATA: KORTHOS

Team

  • Armen Aghajanyan - Co-founder and Chief Executive Officer - Leads Perceptron’s physical-AI model research and company strategy.
  • Akshat Shrivastava - Co-founder and Chief Technology Officer - Leads technical development of Perceptron’s multimodal and embodied models.

Research

  • Perceptron releases open Isaac 0.5 embodied foundation model

    2026-08-26 - Research Publication - 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.

    Source

Timeline

Events

  • 2026-08-26 - Research Publication - Perceptron releases open Isaac 0.5 embodied foundation model

    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.

    Source