Sensing as the Foundation of Physical AI
A great deal of public attention around artificial intelligence has focused on software, language models, and data. Yet there is a parallel revolution unfolding in the physical world — one where machines must not only think, but see, move, and act. Everything that moves is, in essence, becoming robotic. From mining machines and humanoids to drones surveying critical infrastructure, the spread of autonomy depends on a single quiet enabler: perception. Without reliable eyes, intelligent systems are blind, and blind systems cannot be trusted with the tasks we want them to perform.
This is why sensing and perception have become foundational layers of physical AI. Whether the application sits in robotics, automotive, or heavy industry, the underlying need is the same — a sensor stack that can describe the surrounding environment with enough fidelity that a machine can navigate it intelligently and safely.
The Limits of Black-and-White Vision
Lidar has long been a workhorse of autonomous systems. It produces remarkably accurate three-dimensional maps of an environment and is reliable enough that human lives can be entrusted to it. But it has historically suffered from one major blind spot: it sees the world in black and white.
A monochromatic 3D model can identify the shape of a stop sign, but it cannot read the red color that gives that sign its meaning. It cannot tell the difference between an illuminated brake light and an unlit one. It cannot distinguish the state of a traffic signal, or interpret the text painted on a sign. And yet most of the meaningful information that a driver — or an autonomous system — uses to navigate the world is encoded in color. To describe a scene fully, depth alone is not enough.
Until recently, the workaround has been to bolt cameras onto lidar units and try to fuse the two streams of data after the fact. This is technically feasible but enormously difficult: aligning two different sensor modalities in space and time, calibrating them, and reconciling their outputs is a complex engineering problem that absorbs scarce attention from the teams building the actual robots. Every minute spent stitching sensors together is a minute not spent on the application itself.
Native Color: A Single Silicon Solution
The breakthrough now reshaping the field is the integration of color and depth pixels onto the same silicon chip. Instead of two separate sensors that must be synchronized after the fact, a single device captures both simultaneously, producing what can be described as a colorized point cloud.
This shift is not merely an engineering convenience — it is a paradigm change. Customers no longer need to architect bespoke sensor-fusion pipelines. Color and 3D depth simply emerge from the device as a unified data stream, available "for free" out of every sensor. That liberation lets builders concentrate on what they do best: designing the end application, whether that is a humanoid robot, a mining vehicle, or an inspection drone.
The innovation, importantly, is rooted in silicon. By embedding color and depth pixels directly into a chip, the sensor inherits the scaling characteristics of semiconductors. This is a meaningful competitive moat — one that turns sensing from a fragmented integration problem into a vertically integrated product.
Why Color Translates Directly into Safety
It is tempting to think of color perception as a nicety. In reality, it is fundamental to safe autonomy. A robo-taxi on the highway needs to know — instantly and unambiguously — whether the car ahead is braking. A self-driving vehicle needs to read the speed limit posted on a sign. A delivery robot must be able to interpret traffic signals.
Each of these tasks is essentially impossible without color, and each is essential for safety. The principle is straightforward: the more context a robot has about its environment, the more intelligently and cautiously it can navigate that environment. Color perception is therefore not a feature; it is a precondition for trustworthy autonomy.
Reliability of the hardware itself is the other half of the safety equation. Sensors that power autonomous systems must be functionally safe, automotive grade, and ruggedized — engineered so that a single data stream is dependable enough that human lives can rest on it. That standard of reliability is no small bar, and it is one of the reasons the autonomy stack is built on a relatively narrow set of trusted components.
Toward a Unified Perception Platform
The trajectory of the industry now points toward consolidation: a unified sensing and perception platform that can serve every flavor of physical AI. Recent acquisitions in the space — including the addition of stereo-camera technology to lidar portfolios — are extending this vision. By combining colorized point clouds with stereo vision and other modalities, a single platform can address the full spectrum of robotic perception needs.
Commercial momentum reflects the underlying demand. Strong revenue growth in the sector — including double-digit consecutive quarters of expansion and product revenue acceleration of more than fifty percent year over year — signals that customers are beginning to standardize on integrated perception solutions. Hundreds of thousands of cameras and lidars have already been shipped, with customer counts now in the tens of thousands across robotics, automotive, and industrial segments.
The Earliest Innings
For all this progress, it is important to keep perspective on where we sit in the adoption curve. Most people have never received an autonomous delivery from a drone. Most have never encountered a humanoid robot in everyday life. Robo-taxis exist — they can be hailed in cities such as San Francisco — but they remain geographically constrained novelties rather than a routine part of daily life.
That gap between today's reality and the autonomous future being built is enormous, and it is exactly why this moment is so consequential. We are in the earliest innings of physical AI. The sensors, the silicon, and the perception platforms being built now are the foundation upon which a much larger world of autonomous systems will rest. The future may not look quite like a Jetsons cartoon, but it will be unmistakably robotic — and the eyes that allow those machines to see, both in depth and in color, will be the unsung infrastructure beneath it all.