When the largest platform company in physical AI says safety has to be built across hardware, software, AI behaviour and the whole lifecycle, a small team building a robot safety layer should read that as good news. It means the problem is real, and it is being taken seriously.
What Halos is
NVIDIA describes Halos as a full-stack safety system for physical AI. For robots it includes IGX Thor, an industrial module with a dedicated Functional Safety Island, Halos Core for fault detection and monitoring between sensors, actuators and safety components, and the Holoscan Sensor Bridge for flagging invalid sensor data. Simulation in Isaac Lab and Omniverse supports testing. The Halos AI Systems Inspection Lab is accredited by ANAB as an ISO/IEC 17020 inspection body, and TÜV Rheinland is inspecting IGX Thor, Halos OS and the sensor bridge for functional-safety certification readiness.
IGX is designed to support systems built to IEC 61508 and ISO 13849. Those are the same standards we have put on our own roadmap.
Why we see it as complementary
NVIDIA brings certified compute, certified processes and a large ecosystem. We bring a modular robot base, sector safety profiles and a fixed fallback layer that is designed to work when the AI stack, or the platform above it, cannot be trusted. Those are different layers of the same stack.
- Compute: the ZIV compute unit is designed around a Jetson-class module.
- Safety path: Inztinct Core sits below that compute, with its own sensing and its own drive cut-off.
- Testing: simulation tools such as Isaac Lab are a natural place to test safety profiles before they reach a real robot.
What we are and are not
We are a startup, a member of NVIDIA Inception, and we build on Jetson. Membership is not an endorsement or a partnership, and nothing here is a certification. NVIDIA, Jetson and Halos are trademarks of NVIDIA Corporation.