(TÜV Rheinland: Chicago) -- TÜV Rheinland North America has announced its collaboration with NVIDIA to support the certification pathway for AI-enabled safety systems in robotics and physical AI based on NVIDIA Halos.

As industries increasingly explore the deployment of autonomous systems that operate alongside people, TÜV Rheinland is applying its independent testing, inspection, and certification expertise to help assess whether these emerging technologies can meet recognized international functional safety expectations.

As part of this effort, TÜV Rheinland is inspecting NVIDIA IGX Thor, NVIDIA Halos OS, and NVIDIA Holoscan Sensor Bridge for functional safety certification readiness. This work is intended to help robotics developers, industrial automation companies, and system integrators advance AI-enabled safety concepts toward deployable systems that can be evaluated against global standards, including IEC 61508 and ISO 13849.

A significant milestone in the collaboration has already been achieved. TÜV Rheinland North America assessed the previous generation NVIDIA IGX Orin and an IEC 61508-compliant subsystem (NVIDIA Halos Outside-In Safety) using AI to detect human presence and transmit a “safe-to-mute” signal to an autonomous robot. The evaluation incorporated considerations beyond traditional functional safety topics, including emerging guidance from ISO/IEC AWI TS 22440 for AI/ML-based functional safety.

The assessment examined factors such as system architecture robustness, fault tolerance, failure detection and correction, resilience to adversarial input, algorithm transparency and explainability, as well as the verification of algorithm maturity and performance. The work demonstrates how established safety principles can evolve to address the unique characteristics of AI-enabled decision-making.

“Industrial organizations are eager to capture the productivity benefits of physical AI, but large-scale adoption depends on trust,” says Sven Nowak, vice president of cybersecurity and functional safety at TÜV Rheinland. “Independent assessment provides evidence that AI-enabled safety functions behave predictably under real-world conditions and align with internationally recognized safety expectations. That confidence will be essential to moving physical AI from experimentation to widespread deployment.”

Autonomous robots are rapidly moving beyond controlled demonstrations and fixed safety cages into dynamic industrial environments, where humanoids, autonomous forklifts, and mobile manipulators are expected to increasingly operate near people. As these technologies evolve, demonstrating that safety functions remain reliable, robust, and understandable becomes critical for deployment.

NVIDIA Halos for Robotics provides a unified safety architecture that connects AI computing, sensor input, safety software, and applications to help companies address safety throughout the development life cycle. TÜV Rheinland’s participation supports the NVIDIA Halos AI Systems Inspection Lab, helping partners prepare Halos integrations for independent third-party certification assessments.

With decades of experience assessing safety-critical technologies across industrial sectors, TÜV Rheinland is extending its expertise to the emerging field of AI-enabled functional safety, helping organizations navigate the intersection of innovation, regulation, and trust.

“For decades, functional safety standards have guided the deployment of industrial automation,” says Nowak. “We are now entering a new chapter in which those same principles must evolve to address AI-enabled decision-making. Organizations cannot treat AI safety as a black box. Demonstrating robustness, transparency, and fault tolerance will become essential requirements for industrial deployment.”

The NVIDIA Halos Outside-In Safety Blueprint also supports autonomous forklift applications, including trailer loading, high-speed door crossing, and proximity-zone monitoring. In one assessed use case involving trailer loading operations, external cameras monitor entrances and designated safety zones.

The blueprint targets IEC 61508 SIL 2 and ISO 13849 PL d compliance, with concept inspection complete and certification activities currently in progress with TÜV Rheinland.

“We have a unique opportunity in North America to shape the future of AI in safety systems alongside NVIDIA and other leading tech companies,” says Nowak. “This work demonstrates that TIC companies can take a front-and-center role in driving innovation.”

A key part of TÜV Rheinland’s role is helping translate AI innovation into certifiable safety. For industrial customers, this means supporting a structured pathway from NVIDIA Halos evaluation to broader functional safety assessment under IEC and ISO standards. TÜV Rheinland’s work also supports the NVIDIA Halos AI Systems Inspection Lab, which helps partners prepare Halos integrations for third-party certification.

The collaboration has already produced an initial assessment milestone. TÜV Rheinland North America assessed the previous generation NVIDIA IGX Orin that demonstrated IEC 61508 SC3 and SIL 2 capability and ISO 13849 Pl D capability. TÜV Rheinland North America also assessed an IEC 61508-compliant subsystem using AI to detect human presence and transmit a “safe-to-mute” signal to an autonomous robot. In a confined trailer-loading environment, the NVIDIA Halos Outside-In Safety mechanism uses external cameras to monitor entrances and key zones. When a worker enters the monitored area, the system disables the safe-to-mute signal, so the robot performs a safe stop; once the area is clear, the robot can resume operations automatically.

Because the assessed safety function relies on AI/ML for human detection and location, TÜV Rheinland incorporated requirements beyond traditional IEC 61508 topics, including emerging guidance from ISO/IEC AWI TS 22440 for AI/ML-based functional safety. The evaluation considered system architecture robustness, fault tolerance, failure detection and correction, resilience to adversarial input, algorithm transparency and explainability, and verification of algorithm maturity and performance.

NVIDIA Halos for Robotics combines NVIDIA IGX Thor and NVIDIA Holoscan Sensor Bridge for industrial-grade AI compute and sensor connectivity, NVIDIA Halos OS for robotics safety functions and applications, and the NVIDIA Halos AI Systems Inspection Lab to help partners prepare for third-party certification. TÜV Rheinland’s participation helps give industrial customers a standards-driven pathway for physical AI systems that need to sense, decide, and act safely in the real world.

“Physical AI will only scale if companies can demonstrate that AI-enabled safety functions are reliable, explainable, and aligned with recognized standards,” says Nowak. “Through its work with NVIDIA, TÜV Rheinland is helping make advanced AI systems safe for deployment in real-world industrial environments.”

For more information about NVIDIA HALOS for Robotics, read NVIDIA’s official announcement here.