Open embodied AI moves toward actionable robotics releases

Perceptron AI's Isaac 0.5 reportedly offers a 36B open-weight multimodal robotics model with fine-tuning support, Hugging Face weights, GitHub tooling, a 97.2% LIBERO result, and substantially reduced teleoperation-data requirements. The release is actionable for experimentation, but company-reported benchmarks, hardware needs, and licensing require independent verification.