For years, the hard problem in robotics was capability: could a robot perceive its environment, plan a path, manipulate an object? That problem is largely being solved. The new hard problem is trust — proving that an AI-driven machine can operate at full speed in a warehouse aisle, next to a human being, without ever becoming a hazard.
NVIDIA Halos, announced in June 2026, is the most serious attempt yet to industrialise that proof. Rather than treating functional safety as something each robotics team bolts on at the end, Halos delivers it as a full stack: the IGX Thor system-on-module with a dedicated Functional Safety Island in silicon, Halos OS running Linux and QNX for hard real-time control, safety middleware and open-source application blueprints on GitHub — all designed to be inspected and certified together. Agility Robotics' Digit will be the first production humanoid shipping with Halos OS on board.
Two ideas in the platform are worth understanding. Inside-out safety is the classic approach: the robot's own sensors manage its immediate safety envelope. Outside-in safety is the newer one: infrastructure-mounted cameras watch the environment from above, drawing virtual fences and dynamic zones around people and machines. In a trailer-loading concept inspected by TÜV Rheinland, an autonomous forklift runs at full efficiency when the zone is clear and drops into safety mode the instant a person steps in — including in blind spots the robot itself cannot see.
The certification story matters just as much as the technology. Halos comes with an ANAB-accredited AI Systems Inspection Lab and an ecosystem of notified bodies, meaning the path from prototype to a certifiable production deployment is a structured process with named auditors — not a research project.
For companies planning automation in warehouses, factories and logistics, the message is clear: the question is shifting from “can a robot do this job?” to “how do we deploy it safely, and prove it?”. At Svapna we have been building computer vision and real-time video analytics systems for years — and outside-in safety is, at its core, exactly that discipline: cameras, edge inference, zoning logic and alerting, engineered with production rigour. If robots are entering your operations, the safety layer is where the real work begins. Dream it. Do it.