Halos, a safety architecture NVIDIA originally built for autonomous vehicles, is being extended to bipedal robots that will share warehouse and factory floors with human workers, with Agility's Digit named as the first deployment.
Humanoid robots are starting to share factory and warehouse floors with human workers, and the question of who writes the safety rulebook for that shared space is being answered early, by chip and platform vendors rather than by independent standards bodies. NVIDIA said it is extending Halos, a safety architecture it originally built for autonomous vehicles, into humanoid robotics, with Agility's Digit named as the first deployment partner.
The transfer is the story, not the product launch. An autonomous-vehicle safety stack is designed for a machine that moves through predictable lanes and never shares its workspace with a pedestrian. A bipedal robot working shoulder-to-shoulder with a person on a logistics floor is a different safety problem. NVIDIA's framing of Halos as "the industry's first full-stack, comprehensive safety system" for physical AI is the company's own characterization, and the underlying question of whether AV-derived architecture actually maps onto human-collaborative humanoids remains unanswered by anyone outside the vendor and its launch partners.
Halos for Robotics is structured in three layers: NVIDIA IGX Thor plus the Holoscan Sensor Bridge for AI compute and sensor connectivity, the Halos OS software stack for safety functions and applications, and the NVIDIA Halos AI Systems Inspection Lab, which the company describes as a third-party certification pathway. Halos originated as an AV safety stack covering vehicle architecture, AI models, chips, software, tools, and services, and is being ported into robotics. The associated halos-outside-in-safety repository signals an outside-in, sensor-perception safety methodology inside the offering. None of these layers currently carry independent functional-safety certifications tied to humanoid deployments; the standards mapping for ISO 13849 performance levels, IEC 61508 safety integrity levels, or the industrial-robot-specific ISO 10218 and ISO/TS 15066 has not been established in the public material reviewed.
The first test case is Agility Robotics' Digit. According to Technology Magazine, elements of the Halos stack are being integrated into Digit, with end customers Amazon, GXO, Schaeffler, and Toyota Motor Manufacturing Canada cited by Agility rather than confirmed directly by those companies. Agility CEO Peggy Johnson, in an attributed statement, said "safety must be a core component of deployment rather than an afterthought" and described the NVIDIA partnership as extending Agility's "responsible-automation leadership." Agility separately describes Digit as the first humanoid robot in production deployment, a company claim rather than an independently verified industry fact. Digit's fleet management runs on Arc, a multi-cloud management platform from Microsoft.
Deepu Talla, NVIDIA's vice president of robotics and edge AI, framed the move as a response to a scaling problem: physical AI is reshaping factories, warehouses, and logistics, and needs a unified safety architecture before autonomous systems can move into human-collaborative environments at scale. That framing is consistent with the vendor's stated design goal. It is not, on its own, evidence that the architecture holds up in human-collaborative settings. The Inspection Lab is positioned by NVIDIA as the third-party validation layer, but the lab is an NVIDIA-operated facility, and which independent bodies, if any, will sign off on Halos-certified humanoids is not yet public.
A wider ecosystem is forming around the announcement. Third-party reporting describes a roughly 43-partner Halos ecosystem that includes Agility, Boston Dynamics, and lidar maker Hesai, with Agility framed as already integrating Halos into Digit, a figure and characterization that originate in flash copy rather than the NVIDIA newsroom release and should be treated as preliminary. Boston Dynamics has separately expanded its collaboration with NVIDIA to accelerate AI capabilities in humanoid robots, signaling a second major humanoid partner inside the same architecture.
What to watch next is the answer to the transfer question the lede leaves open. The first independent safety, robotics, or standards voice to publicly weigh in on whether an AV-derived architecture can carry a bipedal humanoid through a shared human workspace will determine whether Halos lands as a reference architecture or as a vendor-led proof of concept. The first Agility customer to publish operational safety data on Digit, whether Amazon, GXO, Schaeffler, or Toyota Motor Manufacturing Canada, would do more to settle that question than any NVIDIA press release.