The Humanoid Bottleneck Is Leaving the Cage
NVIDIA launched Halos safety infrastructure for physical AI deployments, with Agility’s Digit as its first humanoid participant.
The announcement concerns humanoid safety certification.
NVIDIA announced Halos for Robotics on June 22, extending its Halos safety framework from autonomous vehicles into robotics and physical AI. Agility is the first humanoid company to incorporate elements of it into Digit, its industrial humanoid with customer and deployment activity involving Amazon, GXO, Schaeffler, and Toyota Motor Manufacturing Canada. That pairing maps directly onto the problem Agility has been trying to solve in public: how to move Digit from controlled workcells into open environments where humans and robots share the same floor.
The cage problem
Industrial robots have traditionally been made safe by separation. Fenced cells, restricted zones, light curtains, and controlled access keep humans away from the robot's working envelope. The logic is simple: if the robot cannot reach a person, it cannot harm one.
Humanoids intended to move between tasks in warehouses, factories and loading docks need safety controls for navigation and work alongside people beyond a fixed cell.
If a human approaches the humanoid, the robot needs to take action to bring itself down to the ground, avoid harming the person, and avoid dropping whatever payload it is carrying.
Johnson also tied that transition directly to Agility’s product roadmap. She described the current Gen 4 robot as still operating around the safety-cage model, while the next Gen 5 architecture is expected to scale because it can operate outside the workcell, moving to areas such as loading docks and travelling among human workers.
Those details are interview-sourced, not formal certification claims. Agility has also said publicly that safety standards for dynamically stable industrial mobile robots are still developing.
What NVIDIA is doing
Halos for Robotics is positioned as a full-stack safety system spanning AI compute, sensor connectivity, system software, safety applications, and inspection support. The stack includes IGX Thor for industrial-grade AI compute with built-in safety capabilities, the Holoscan Sensor Bridge for sensor connectivity, Halos OS and Halos Core for safety-related operating functions, an Outside-In Safety Blueprint using external sensors, and the Halos AI Systems Inspection Lab. NVIDIA says the lab is the first program accredited by the ANSI National Accreditation Board for functional and AI safety assessment for physical AI, working with certification bodies including TUV Rheinland, UL Solutions, TUV SUD, exida, SGS, and CertX.
Agility is integrating IGX Thor and Halos Core into Digit's proprietary safe human detection system, with IGX Thor delivering industrial-grade AI compute and Halos Core supporting the software layer for safety-related operating functions. Together, Agility and NVIDIA plan to use the lab to help ensure Digit's safety-related software, AI components, and cybersecurity protections meet standards including IEC 61508, ISO 13849, and ISO/IEC TR 5469 before final third-party certification.
As Johnson said in NVIDIA's announcement: "For humanoids to deliver value at scale, safety has to be built into the robot and validated across the entire system."
Deployment beyond fixed cells
Agility’s RoboFab in Salem, Oregon, has stated capacity for more than 10,000 robots per year.
Have a robotics update Korthos should review? Send news, deployments, product releases, funding rounds, research, or media to [email protected] or reach out on X at @agkorthos.