NVIDIA, a leader in AI computing technology, has announced NVIDIA Halos, a comprehensive safety system for autonomous vehicles (AV), taking a step closer to the development of future autonomous driving technology.

Nvidia unveiled Halo at GTC held in San Jose, USA on the 21st.

Halos is a system that combines NVIDIA's unique hardware and software safety solutions and is expected to significantly enhance the safety of autonomous vehicles.

Halos is an AI-based end-to-end stack that integrates chips, software, tools, and services to support the development of safe autonomous vehicles from the cloud to the vehicle body.

It encompasses safety across three dimensions: technology, development, and computing, and builds a comprehensive framework that encompasses AI training, simulation, and deployment using NVIDIA DGX and Omniverse Cosmos based on OVX, Drive AGX, etc.

In particular, Halos proactively secures safety through powerful algorithms and datasets.

Inappropriate data is filtered out before training, and is trained using NVIDIA Omniverse Blueprint and Cosmos World Foundation Model (WFM). Further enhances safety during training, simulation, and validation.

NVIDIA operates the NVIDIA AI Systems Inspection Lab, along with Halo, to verify the safety of autonomous vehicle technology. The lab verifies functional safety, cybersecurity, and AI safety in a unified framework, and major global companies are participating.

A Continental official who participated in the AI System Inspection Lab expressed his expectations, saying, “We will strengthen the safety standards for autonomous vehicles through cooperation with NVIDIA.”

Halo’s core is three elements: platform safety, algorithm safety, and ecosystem safety. Platform safety provides a system-on-chip (SoC) with hundreds of safety mechanisms and a safety-certified operating system, while algorithm safety supports debiasing in data generation and curation processes and training advanced AI models.

Ecosystem Safety promotes standardization of AV technology through continuous data improvement and automated safety assessment.

NVIDIA has put in over 15,000 engineering hours, over 1,000 patents, and 240 published research papers to build Halo.

It has met global safety standards by obtaining major international safety certifications such as ISO 26262 ASIL D and ISO/SAE 21434.