The landscape of industrial automation is shifting from restricted machine zones to collaborative human-robot environments. Fourteen years ago, Rodney Brooks’ Rethink Robotics introduced Baxter to the manufacturing environment. Baxter was designed to work outside of its cage. Rethink closed in 2018, but Baxter’s cage-free DNA laid a foundation for today’s cobot environment.
In 2012, the global collaborative robot market size was valued at $100 million, with players like Universal Robots and Rethink defining the emerging market. In 2025, the collab robot market was $2.9 billion and is expected to reach $4.0 billion in 2026 and $17.2 billion by 2033. Even with those numbers, the International Federation of Robotics reports that approximately 89% of industrial robots are still caged, leaving around 10.5% uncaged in 2023.
According to Knut Sandven, CEO of Sonair Robotics, software innovation has accelerated the development of robots in manufacturing settings, but it has also exposed a bottleneck: safety has not kept pace with how robots are used.
“Mobile robots make cages impossible by definition,” said Sandven. “A robot carrying goods from A to B simply can’t be fenced in. So, robots left their cages out of necessity, and in recent years, AI has sharply accelerated what they can do.”
Sandven says that today we envision robots working alongside people in almost any setting, even in our homes, but he warns that safety sensing hasn’t kept pace with their use. He says that the sensors in use today are laser-based and built for the stationary, predictable world of controlled light and material conditions, which limits their ability to serve as a reliable safety basis.
A 3D safety layer
Sonair’s solution to these challenges is a 3D safety layer for collaborative robots. Sonair manufactures Acoustic Detection and Range (ADAR) sensors to give collaborative robots real-time 3D vision. Sonair recently announced a regulatory milestone achieving the first independently safety-certified 3D ultrasonic sensor.
“Once you can certify detection in 3D, the question shifts from ‘can we detect a person across a plane’ to ‘can we account for a person anywhere in the working volume’. Sandven says this sets a higher, more honest bar for collab robots.
“This 3D certification takes certified safety for a mobile machine from 2D to 3D for the first time,” he said in an email interview. “Until now, a robot could only be certified to detect people in a flat plane. It could only see a single slice of the world. Certifying 3D ultrasonic sensing gives the industry a proven, deterministic way to detect a human anywhere in a volume, not just across a line on the floor.”
Sandven says that lets robots work closer to people, with protection throughout the entire workspace. “This is a safety layer the whole robotics industry can build on, whatever robot or AI is built on top of it,” he added.
Tearing Down Physical Cages
Sandven says the industry has assumed that heavy industrial machinery remains isolated behind cages because robots are inherently dangerous. But in reality, he said, they are cordoned off because their traditional safety hardware is essentially blind.
“Standard 2D laser systems, like LiDAR, only monitor a flat plane at shin-height. If a factory worker crouches down to pick up a tool or reaches over that single invisible scan line, the sensor fails to see them.”
Sandven says that this kind of certified safety is now available in 3D, which completes the shift from theory to practice.
“That should change how the whole industry thinks about virtual safety barriers,” he said. “Instead of drawing them on a flat plane, you can define them through a real volume around the machine. And once the protected space is something you can trust in three dimensions, you can finally start taking down the physical fences rather than working around them.”
Ultimately, the real breakthrough isn’t always about making robots smarter. “It’s about making them trustworthy, too,” said Sandven.
AI Blindspot
While modern robotics relies heavily on advanced machine intelligence to guide machines, Sandven warns that treating a perception stack in the AI brain as a safety guarantee is a critical engineering mistake.
“I think the biggest obstacle for many robot makers now is that they start too late to think about how they can make their robots safe,” said Sandven. “Part of the industry assumes that if perception just gets good enough, with more cameras and better AI, safety will follow.”
Sandven says the robots that earn their place among people are the ones that will be built with a certified safety layer from the start.
“While AI models excel at mapping and navigating environments, they operate on probabilistic models, meaning they calculate highly confident guesses rather than absolute certainty,” said Sandven.
He says that for an explicit safety function, a software-driven guessing game introduces unacceptable risks. “Sound waves, by contrast, are governed by unyielding physical laws. They travel at a known speed and reflect predictably, allowing engineers to mathematically bound and prove a machine’s worst-case response time.”
Sandven emphasizes that safety functions must be kept separate from the robot’s AI brain to ensure that an artificial intelligence glitch or failure can’t compromise human safety. He believes that the hard lesson will be that ‘very confident’ and ‘proven’ aren’t the same thing, and that you can’t bolt safety on as a last step before deployment.
Ultimately, the real breakthrough isn’t always about making robots smarter. “It’s about making them trustworthy, too.”
