Nvidia puts robot safety workloads on dedicated Halos hardware
Halos for Robotics gives robot makers a separate safety processor, safety software and an accredited test lab. Agility's Digit humanoid is the first to use it.
Nvidia Halos for Robotics, announced in June 2026, extends the company's self-driving car safety system to warehouse, factory and surgical robots.
Its IGX Thor computer has an independent processor that runs only safety tasks, next to the robot's main AI.
Agility Robotics is the first user, building Halos into the safe human detection system of its Digit humanoid.
An accredited Nvidia lab helps partners prepare robots for outside safety certification by groups such as UL Solutions and TÜV SÜD.
Quick read · 1 min
Nvidia Halos for Robotics is a safety system for robots that work near people. Announced in June 2026, it extends the Halos platform Nvidia launched for self-driving cars in 2025.
The key idea is a separate safety brain. Nvidia's IGX Thor computer has an independent processor that only handles safety tasks, running next to the robot's main AI so it can react in real time. Halos also includes safety software, a blueprint that uses outside cameras to help robots slow down near people, and an accredited lab that prepares robots for independent certification.
First user: Agility Robotics' Digit humanoid, already used by firms such as Amazon and GXO.
Why now: Nvidia says safety is the new bottleneck as robot AI improves.
Next: Agile Robots is exploring Halos for a research humanoid due in early 2027.
Nvidia thinks the next thing holding robots back isn’t brains or bodies. It’s safety. That’s why its Halos for Robotics system gives a robot a second, independent processor whose only job is keeping the people around it out of harm’s way.
This matters because robots are moving out of fenced-off cages and onto shared floors. A humanoid walking through a warehouse or factory may pass workers, forklifts and other machines all day. If you work in logistics or manufacturing, the safety systems inside those robots will soon shape your workplace. For Nvidia, physical AI is also big business: chief executive Jensen Huang said in March 2026 that it already brings in nearly $10 billion a year.
01
What is Nvidia Halos for Robotics?
Put simply, Nvidia Halos for Robotics is a full-stack safety system for robots, meaning it covers the chips, the software, the sensor data and the testing. Nvidia calls it the industry’s first system of its kind for physical AI, the term for machines that sense, decide and act in the real world.
It grew out of Halos for self-driving cars, which launched in 2025. Nvidia announced the robotics version in June 2026, aimed at mobile robots in warehouses, humanoids in factories and even surgical robots. Behind it, Nvidia says, sit more than 18,600 engineering years spent on autonomous vehicle safety.
Amit Goel, who leads Nvidia’s robotics ecosystem and edge computing work, says the AI models and robot hardware are now capable enough that safety has become the next bottleneck. Halos is Nvidia’s attempt to clear it.
02
Why the safety system gets its own processor
The core of the system is hardware. The Nvidia IGX Thor computing module, built for robots and industrial machines, includes an independent processor set aside for safety-related work. A Holoscan Sensor Bridge links cameras and other sensors to it with low delay.
Goel’s argument is that in physical AI, the robot’s normal functions and its safety system need to run on the same chip. That keeps the safety checks close to the AI making decisions, so they can react in real time. Separating the safety work onto its own processor means a busy main AI can’t crowd out the part that watches for danger.
03
The software and testing behind it
On top of the hardware sits Halos OS, the software layer for robot safety. It includes Halos Core, which handles safety-related operating functions.
Nvidia also offers an Outside-In Safety Blueprint. Instead of relying only on the robot’s own sensors, it uses cameras mounted around a site and AI agents to help robots change their behavior, for example slowing down near people. Industrial partners including KION Group, FORT Robotics and Inventec are building safety agents with it.
The final piece is the Halos AI Systems Inspection Lab. Nvidia says it is the first program for physical AI safety accredited by the ANSI National Accreditation Board. The lab helps partners get ready for independent certification from testing groups such as UL Solutions, SGS, TÜV SÜD, exida, CertX and TÜV Rheinland, which is inspecting IGX Thor.
04
Agility’s Digit humanoid is the first to use it
Agility Robotics is the first company to adopt Halos for Robotics. Its humanoid robot, Digit, already works in logistics, manufacturing and warehouses for customers including GXO, Toyota Motor Manufacturing Canada, Amazon and Schaeffler.
Agility is building IGX Thor and Halos Core into Digit’s own system for detecting humans safely. It will also use Nvidia’s inspection lab to check Digit’s safety software, AI parts and cybersecurity against standards such as IEC 61508, ISO 13849 and ISO/IEC TR 5469 before final outside certification. Agility chief executive Peggy Johnson called that kind of safety work a “non-negotiable requirement” for putting humanoids into industrial jobs.
A wider group of partners supports the system. Chip and sensor suppliers include Texas Instruments, Infineon, SICK, NXP and STMicroelectronics. QNX, Acontis and Amazon FreeRTOS provide real-time software, and Advantech and NexCobot build IGX-based computers for robots.
For workers, the hope is fewer surprises on the floor. A robot with a dedicated safety processor and outside cameras watching the space should be better at spotting a person and slowing or stopping in time. Independent certification against known standards also gives employers something firmer than a vendor’s promise.
It doesn’t make any robot automatically safe, though. Each company still has to build, test and certify its own machine, and those results take time. Nvidia also has a financial stake: it has invested in humanoid startups including Agility Robotics and Figure AI, and it has worked with China’s Unitree on an open humanoid reference design for researchers.
06
What happens next for Halos
More robot makers are lining up. In late October, at its GTC event in Berlin, Nvidia and Agile Robots unveiled the Agile ONE SN, a compact research humanoid running Nvidia’s Jetson Thor platform. Agile Robots says it is exploring Halos integration, but that work is still in development, and the robot is not certified for unsupervised industrial use. It is due in early 2027.
For Digit, the next milestone is third-party certification, which will show how well the Halos approach holds up under outside testing.
07
Quick answers
What does Nvidia Halos for Robotics do?
It is a safety system that combines a dedicated safety processor, safety software, outside camera monitoring and an accredited test lab to help robot makers build safer machines.
Which robots use Nvidia Halos?
Agility Robotics’ Digit humanoid is the first. Agile Robots is exploring Halos for its Agile ONE SN research humanoid.
Is Halos only for robots?
No. Halos started in 2025 as a safety system for self-driving cars before Nvidia extended it to robots in June 2026.
The machine learning veteran says Microsoft AI wanted bigger models, not research. Here is what happened to AI Frontiers, his cost-cutting idea and the AutoGen framework.
7 min read
We use cookies to understand how readers use Talk With Tech, so we can make it better. Is that OK? Privacy policy
The Daily Brief
Today's biggest tech stories, in 5 minutes
Every morning, the news that matters from AI, phones, apps and the people shaping tech. Explained in plain English. Free.
One email a day. No spam, unsubscribe anytime. Privacy policy