Halos: A Humanoid, a Crate and a Colleague

Picture a humanoid robot carrying a crate of parts down a factory aisle. The crate is for a technician at the end of the aisle, but before the robot gets there, she steps out from behind a shelf two meters ahead, her eyes on her tablet. What should the robot do?

Situations like this are what we design for. They look simple, but they touch on almost every safety question a humanoid raises. 

For robot arms, most of these questions were answered long ago, and we have spent years building industrial robots with safety designed in from the start. Their safety functions limit speed, torque, power and force, and the technology behind them is certified by TÜV SÜD.

That experience covers both sides of the problem: turning safety requirements into functions that pass certification, and building those functions reliably at scale. More than 20,000 of our automation solutions, made in our own factories, are in use around the world today. 

The humanoid in the aisle has a harder job than any robot arm, because it is walking. The crate in its hands adds momentum, and it has to stay balanced with every step. 

That also undermines the classic safety reflex. When a robot arm stops, it is safe. When a humanoid freezes mid-stride with a load in its hands, it may topple over and create the very hazard the stop was supposed to avoid.

Instead, the robot has to respond more gradually, slowing down as the technician approaches, adjusting its gait and steering around her. It also has to know exactly where she is. If the technician is within reach of its arm, the risk is a touch or a bump. If she is standing where the robot could fall onto her, the risk is a different one, and the robot has to react accordingly. 

Then the technician looks up, sees her delivery and reaches for the crate. This time contact is intended, which raises a new question: how much force is acceptable?

People answer that intuitively. We know that a pat on the shoulder is harmless, while the same force against someone's temple is not. A humanoid needs the same understanding, but in a form that can be verified.

This is why we are building contact limits directly into the robot's control. Depending on which part of the body it might touch, and in what situation, the robot adjusts its force and speed. For a humanoid, those limits also have to factor in movement, payload and balance. Because this logic sits inside the robot itself, companies that deploy humanoids would not have to define every contact limit from scratch for each new application. That could make collaborative applications faster to set up and easier to validate. 

Contact limits like these need a place in the robot's overall safety architecture. That is where our work with NVIDIA comes in: together, we are developing safety applications and architecture concepts for collaborative humanoids.  

The basis is NVIDIA Halos, a safety platform for physical AI that brings computing, system software, sensors and validation tools together in one architecture and builds on NVIDIA's experience in autonomous driving. One question we want to explore together is how our contact limits could run as safety functions on this platform, combining our experience with certified robot safety with NVIDIA technology for physical AI. 

The safety standards for humanoids in industry are being written right now, and our safety experts are contributing to two new ISO standards. One covers safety requirements for industrial mobile robots. The other is written specifically for robots with actively controlled stability, meaning robots that need continuous control to stay upright and could become unstable if they lose power. That includes everything from bipedal humanoids to quadrupeds and self-balancing wheeled platforms, and it describes exactly the robot carrying the crate down our aisle. 

Both standards are aimed at industrial environments with restricted public access. Taking part in this work keeps us close to the questions the whole industry will have to answer and lets us bring our experience with certified robot safety to the table. 

All of this work comes back to the moment in the aisle. The technician takes the crate, nods and carries it to her workstation, and the robot turns back the way it came. For her, it was just an ordinary handover. Making moments like this safe, and proving that they are, is what we are working toward.

Header image: A humanoid robot walking alongside a person. For illustrative purposes only.