NVIDIA Launches Halos Software to Enable Safe Human-Robot Collaboration for Humanoid Robots
Claire Weston
Nvidia unveiled Halos, a safety software system that lets humanoid robots make autonomous safety decisions in real time around people — instead of simply shutting down when a human is detected. This means → robots shift from 'isolation safety' to 'collaborative safety,' a prerequisite for real-world deployment.
What's wrong with how robots handle safety today?
Current industrial robot safety runs on a simple rule: detect a human nearby, slow down or stop.
This means → robots are either caged off or effectively paralyzed the moment a person approaches — they cannot hand over objects, assist with carrying, or do anything requiring direct contact.
In plain terms = today's safety approach keeps robots safe but also keeps them nearly useless — the safer they are, the less they can do.
How does Halos solve this?
Halos is derived from Nvidia's autonomous-driving technology and runs on Nvidia's IGX Thor hardware — a high-compute chip platform designed for edge scenarios.
The core shift: the robot can perceive its surroundings in real time and make its own safety decisions, rather than defaulting to "see human, stop."
The system also supports external sensor networks — for example, an autonomous forklift in a warehouse can tap into ceiling cameras to check what's around a corner before deciding to proceed at full speed or slow down.
This means → safety judgment upgrades from "robot alone" to "robot + environment sensors" co-decision.
Why is safety design so much harder for humanoid robots?
Agility Robotics CTO Pras Velagapudi argues that humanoid robot safety is far harder than autonomous driving.
This reflects a fundamental difference — self-driving cars only need to avoid contact; humanoid robots must judge which objects they can touch, move, or apply force to, and how much force is appropriate.
In plain terms = if a robot is so weak it can only lightly brush a person, it lacks the strength to do any useful work — safety and capability must be solved together.
Who's using it, and how big is the market?
Agility Robotics is among the first customers. Its Digit humanoid robot is already deployed at a Toyota manufacturing plant in Canada and will integrate Halos.
The commercialization path: start with warehouse logistics, then extend into retail, healthcare, and construction — each new scenario is harder, but each is a massive market in its own right.
Barclays projects the humanoid robot market will reach $200 billion by 2035.
This means → whether Halos can turn "safety perception" from a technical concept into a mass-certifiable standard is the pivotal variable for this market to deliver on schedule.
What else is Nvidia doing to drive adoption?
Nvidia also announced a dedicated safety-testing lab where robot manufacturers and customers can test before applying for regulatory certification.
Nvidia engineers will provide pre-certification checks and technical tuning support.
This reflects Nvidia's broader playbook: not just selling software and chips, but making the certification process itself part of its ecosystem.
Content is for reference only, not financial advice.