NVIDIA Lays Out Full-Stack Halos Safety System for Scaling Physical AI

NVIDIA Blog · rss · 2026-09-22

NVIDIA argues that scaling physical AI—49 million L3-5 AVs by 2035 and 60 million industrial robots by 2035, per ABI Research and Omdia—demands safety across hardware, software, AI behavior, and the operating environment, not a one-time pre-deployment check.

Four shifts define the new safety model

NVIDIA's Halos stack

Ecosystem & certification: Geely, Isuzu, Nissan (Wayve software), and Einride build L4-ready vehicles on Hyperion; Uber, Grab, and Lyft use it for robotaxi scaling; Agility integrates IGX Thor into Digit 5's safety system. TÜV SÜD certified DriveOS 6.0 to ISO 26262 ASIL D, and ANAB accredited the Halos AI Systems Inspection Lab as an ISO/IEC 17020 inspection body.

Original post →

More from Embodied

Embodied channel →