Nvidia's Halos safety stack aims to make robotaxis and humanoid robots safer

[CMPunk88]

Ars Technica takes a look at Nvidia's big bet on physical AI, and in particular how it is trying to make robotaxis and humanoid robots safe. The centrepiece is Halos, a safety system that Nvidia first built for self driving cars and has now adapted for robots. Halos for Robotics launched in June, so the Ars piece is more of a deep dive than a new announcement. Nvidia describes Halos as the first and only full stack safety system for physical AI

There are several layers to it. Halos OS monitors hardware and software for faults and keeps safety critical workloads isolated from everything else. The Holoscan Sensor Bridge can flag corrupted sensor data and runs on microcontrollers and FPGAs. There are simulation tools for testing, and an inspection lab that reviews safety documentation, which Nvidia says is the first programme of its kind to be accredited under ISO/IEC 17020

On the robot side, the hardware is IGX Thor, which includes a dedicated Functional Safety Island. Agility Robotics is the first adopter, putting IGX Thor and Halos Core into its Digit humanoid. Agility CEO Peggy Johnson said that for humanoids to deliver value at scale, safety has to be built into the robot and validated across the entire system. Partners include KION, Infineon, NXP, STMicro and Texas Instruments, plus certification bodies such as TÜV Rheinland, TÜV SÜD and UL Solutions

For cars, Nvidia offers DRIVE AGX Thor, its Hyperion level 4 platform and its Alpamayo reasoning models. Uber, Grab and Lyft are using Hyperion for robotaxis, and Geely, Isuzu, Nissan and Einride are building level 4 ready vehicles on it. Nvidia cites more than 18,600 engineering years of safety work on autonomous vehicles. Industry forecasts it quotes suggest 49 million level 3 to 5 vehicles and about 60 million industrial robots could be deployed by 2035

One point the coverage makes well is that safety needs vary hugely. A robot vacuum and a warehouse forklift need very different levels of protection. Nvidia's VP of robotics Deepu Talla said physical AI is transforming how factories, warehouses and logistics operations work. I think a standard safety stack is a sensible idea, as long as companies do not treat it as a box ticking exercise

Would you trust a humanoid robot working near you? And should one company's safety platform become the industry standard?

Nvidia's big bet on physical AI aims for safer robotaxis, humanoid robotsarstechnica.com