How Boston is becoming a physical AI hub

5 Min Read

A live stereo depth stream (left) and an RGB color stream (right) from an Intel RealSense D585 Pro AI-native 3D depth camera system.RealSense

Boston is emerging as a center for robotics and physical AI, with major players including Boston Dynamics, Amazon Robotics and, as of this month, Google DeepMind developing new industrial and domestic robots in the region.

The area is also attracting investment in the technologies that help robots perceive and navigate the physical world. Last month, RealSense, an Intel spinoff developing AI-powered computer vision tech, was bought by Natick, Mass.-based machine vision vendor Cognex. The $500 million deal folds one of the most widely used robot vision companies under a Massachusetts parent, adding to the area’s momentum.

In this Q&A, RealSense CEO Nadav Orbach, who will lead Cognex's new robotics division, spoke about the role of 3D perception in autonomous systems, the factors driving Boston's robotics ecosystem and how hub fits into the wider U.S. push to strengthen domestic supply chains.

Related:Robot company bets touch, sound can give bots an edge over vision AI

RealSense develops 3D cameras to help robots understand the geometry of their surroundings. Why is this important as robots move into more complex environments?

Nadav Orbach: A conventional camera captures pixels representing colors, whereas a 3D camera also measures the distance from the camera to each point, giving you the geometry of the environment. That’s important for robotic perception because if a robot needs to pick something up or avoid hitting a wall, it needs to understand the full physicality of its surroundings.

We initially explored applications in laptops and tablets, but it was with the rise of robotics that we found product-market fit. Today, our cameras are used across industrial robotics, with around 40% of our business concentrated in industrial and warehouse robotics, including autonomous mobile robots, automated forklifts and inventory-management systems. Another quarter or so is humanoids and quadrupeds.

Are 3D cameras becoming standard in autonomous robots, or are there alternatives?

Orbach: There are other approaches, including radar and proximity sensors, but vision is the dominant perception technology. Autonomy is built on two things: perception and reasoning. Sensors handle perception, while AI running on a general-purpose processor handles reasoning.

Think about early domestic robots, such as robotic vacuum cleaners. They could bump into a wall or sofa and then correct their direction. For a relatively simple task, that was acceptable. But if you have a 100-pound humanoid walking around, you don't want it bumping into furniture and correcting itself afterward. It needs a much more accurate understanding of its surroundings.

Related:Robots can't talk to each other yet -- that's a big AI bottleneck

What is Boston's role in the emerging physical AI landscape?

Orbach: If you look at robotics today, I would say the Bay Area and Boston are the two biggest hubs in the U.S.

These start out with strong universities and professors who are innovating, which creates an early-stage startup ecosystem that can evolve into a broader industry. Over time, the ecosystem develops a life of its own because the talent pool is there -- companies attract people who want to work in robotics, and that talent attracts more companies.

I think these hubs are much more about talent than manufacturing. When a robotics company starts out, it typically does everything in-house: it buys a processor and a camera, builds the mechanics and deploys its robots with a small number of customers. As it scales, it has to decide whether to manufacture domestically or work with a large contract manufacturer in Asia. There isn't one answer; some companies manufacture internally, while others outsource.

But crucially, while manufacturing can move among locations, what you cannot move so easily is the talent pool. That's the real advantage of a place like Boston.

Does this reflect a broader shift toward building more of the robotics stack in the U.S.?

Related:China leads dexterous hand supply for humanoids

Orbach: I think there is a fairly clear division in the ecosystem today. The U.S. is a major hub for autonomy and the software stack that makes robots autonomous. Companies such as OpenMind, Skild AI and Field AI are developing the systems that act as the brain of the robot.

From a mechanical side, a lot of manufacturing is still concentrated in Asia. But I think some of the mechanics will become commoditized over time, while a lot of the innovation will happen in AI and software.

How are you seeing this push to diversify supply chains (such as the FCC’s ban on foreign-made robotics components) affect U.S. robotics companies?

Orbach: I haven't seen a direct impact from the FCC restrictions yet, although my customers may have. What I have seen are shifts driven by tariffs and other restrictions over the past few years. These have pushed some U.S. companies to move production out of China, either to the U.S. or elsewhere, and given a greater focus on localized manufacturing than we’ve seen even two or three years ago.

We're a platform company, so we're neutral, but our customers build the full robot, which is very sensitive in terms of its application and the regulations around it, especially as robots move into more complex environments.

Now, we’re seeing a lot more regulation around humanoids because there's growing concern about AI’s prevalence. I think governments are rightfully cautious, and they’re taking measures to examine a product's origins as a result.

Are data concerns growing as robots become more capable of perceiving their surroundings?

Orbach: I would say yes, because on a fundamental level the data collected by robots is huge. They are continuously perceiving and generating a lot of visual and audio data, which can be sensitive for customers. But the concern isn't only about data: robots also have physical capabilities and can act on the world, so people are cautious.

Some of that caution is justified, while some is just fear of the unknown -- it reflects the fact that robotics is a relatively new industry that isn't yet regulated like more established ones. If you look at the automotive industry, cars also have multiple cameras, continuously perceive their surroundings and can cause harm in the physical environment, but we've built regulatory and safety frameworks around them. That means people don’t have the same reservations.

Robotics is at the start of a similar journey. As regulation and safety standards develop, the industry will be better positioned to deploy robots at scale.

Editor’s note: This interview has been edited for clarity and conciseness.

About the Author