Perceptron open-sources Isaac 0.5, a 36B-parameter embodied foundation model spanning 35+ robots

AkshatS07 · x · 2026-09-10

Perceptron released Isaac 0.5, an open-source 36B-parameter sparse embodied foundation model claiming frontier multimodal video understanding, embodied reasoning and robot control. Trained on 35+ embodiments, 100K hours of robot data, 1M hours of video and 3T tokens, it introduces a data-mix scaling law: scaling general video from 1K to 1M hours cut required teleoperation data from 5,900 to 28 hours. Repetitive tasks like box packing fine-tune reliably with 30 episodes. Weights and paper are public.

Related event: Perceptron Open-Sources 36B Embodied Foundation Model Isaac 0.5(2 posts)→

Original post →

More from Embodied

Embodied channel →