NVIDIA Halos Robotics Safety Stack Debuts with Agility Digit Humanoid

NVIDIA Halos Robotics: Standardizing Safety for Scaling Humanoids

On June 22, 2026, at the Automate show in Chicago, NVIDIA unveiled Halos for Robotics, positioning it as the industry's first open, full-stack safety architecture purpose-built for physical AI and humanoid platforms. The announcement arrives at a critical inflection point where regulatory scrutiny and insurer requirements are increasingly dictating deployment timelines for robots sharing space with humans.

From Autonomous Vehicles to Factory Floors

NVIDIA is leveraging more than a decade of autonomous vehicle safety engineering—encompassing roughly 18,600 engineering years and billions of safety transistors—to create a unified framework. The robotics version layers safety across compute, sensor connectivity, operating system functions, and third-party certification pathways. This addresses a persistent bottleneck: fragmented safety implementations that force each humanoid maker to reinvent functional safety validation from scratch.

Core Architecture Layers

At the hardware foundation sits the NVIDIA IGX Thor platform paired with the Holoscan Sensor Bridge. IGX Thor delivers industrial-grade AI inference with an integrated safety island capable of supporting IEC 61508 SIL 3 requirements through dedicated diagnostic coverage and error-handling mechanisms. The Holoscan Sensor Bridge provides deterministic, low-latency Ethernet-based connectivity for multimodal sensors, enabling real-time data paths that feed directly into safety-critical decision loops without CPU bottlenecks.

Halos Core OS then supplies the software runtime for safety-related functions, application isolation, and monitoring. The stack culminates in the NVIDIA Halos AI Systems Inspection Lab, now ANAB-accredited for functional safety and AI safety assessments. Partners can use the lab to prepare for third-party audits against standards including IEC 61508, ISO 13849, and emerging AI-specific guidelines before handing off to bodies such as TÜV Rheinland.

Agility Digit: First Named Integrator

Agility Robotics becomes the initial commercial platform to publicly commit to the stack. Digit will incorporate IGX Thor for its compute backbone and Halos Core to bolster the company's existing proprietary safe human-detection system. The integration targets logistics, manufacturing, and warehouse environments where Digit already operates. Agility will also participate in the inspection lab workflow to align software, AI models, and cybersecurity elements with certification requirements.

Manufacturing and Certification Bottlenecks

While the architecture promises standardization, several hurdles remain visible. Sensor Bridge latency budgets and deterministic Ethernet timing must hold across diverse third-party camera and lidar modules. Software certification for AI components—particularly learned perception models—remains an evolving discipline with limited precedent at scale. The inspection lab's throughput will determine whether the ecosystem can move beyond early adopters quickly enough to satisfy 2027–2028 volume forecasts from multiple humanoid programs.

Market Implications

Safety certification is rapidly becoming the gating item for insurance, regulatory approval, and customer acceptance in shared workspaces. NVIDIA's move compresses the validation timeline for smaller robotics teams while raising the baseline for all participants. Expect follow-on announcements from other humanoid developers and autonomous mobile robot fleets as they evaluate the stack against in-house or competing open-source safety approaches.

The June 22 reveal signals that physical AI safety is no longer an afterthought bolted on during integration; it is now a first-class architectural layer with measurable hardware and software commitments.