Three engineers who built the depth sensors inside a billion devices left the Cupertino company in 2021 with a different question: what if robots could see the way iPhones do?
LyteVision fuses 4D sensing, RGB imaging, and motion data into a single perception stack for robots
CES 2026 named it Best of Innovation in Robotics — the industry's first vertically integrated perception platform
Alexander Shpunt, Arman Hajati, and Yuval Gerson had spent the prior decade making cameras understand depth. At PrimeSense, which Shpunt co-founded and the Cupertino company acquired in 2013, they helped create the 3D sensing pipeline that became Face ID. At Microsoft, they worked on Kinect. They knew how to make machines see.
The problem they saw after leaving was a different one entirely.
Robots in 2021 did not lack sensors. They lacked integration. A factory robot might carry a LiDAR from one vendor, a stereo camera from another, and an inertial measurement unit from a third — and then spend months in calibration hell stitching the outputs together. McKinsey estimates that over 60% of industrial companies lack the internal capability to implement robotic automation at all. The bottleneck was not hardware. It was perception software that did not ship as a system.
The PrimeSense lineage
Shpunt founded PrimeSense in 2005, years before depth sensing became a consumer technology. The startup's infrared structured-light system powered the first Microsoft Kinect, selling 24 million units and becoming the fastest-selling consumer electronics device at the time. Apple acquired the startup in 2013 for $360 million, and its depth-sensing technology became the foundation of Face ID — now embedded in over a billion iPhones and iPads.
Hajati led sensing architecture at the same company. Gerson led camera module engineering. Together they held dozens of patents in optical sensing, depth computation, and sensor fusion.
"Apple was a good school," Shpunt told Bloomberg. "We realize that perception — and more generally, having robots understand what they do, be safe and immediately react to the world — is something we would like to solve."
They started Lyte in 2021 and worked in stealth for four years.
The LyteVision stack
Lyte's core product, LyteVision, combines three data modalities into one hardware-software block. A 4D sensor measures not just distance but how objects move over time. An RGB camera captures color and texture. An inertial measurement unit tracks motion and orientation. The fusion happens at the edge, inside the sensing module, before data reaches the robot's main processor.
The result is a single connection that delivers spatial and visual data simultaneously — a "visual brain," as the company describes it, rather than a collection of eyes and reflexes.
This matters most in unstructured environments. A robot arm on a factory line can operate with pre-programmed coordinates. A mobile robot navigating a hospital corridor cannot — doors open, people cross, carts block the path. The robot needs to interpret motion, anticipate obstacles, and decide in real time. LyteVision is designed for that second category.
Teams today assemble perception from multiple vendors, then spend months calibrating sensors and writing fusion software. Lyte's vertically integrated stack eliminates that cycle entirely.
The platform supports autonomous mobile robots, robotic arms, quadrupeds, humanoids, and robotaxis — a range that spans from warehouse pallet movers to passenger vehicles.
$107 million and a chairman from semiconductors
Lyte raised its capital from a group that includes Fidelity Management & Research, Atreides Management, Exor Ventures, Key1 Capital, and Venture Tech Alliance. The founding investor and chairman is Avigdor Willenz, a semiconductor entrepreneur who previously founded Annapurna Labs (acquired by Amazon) and Habana Labs (acquired by Intel).
"Lyte is building at the right layer, at the right moment," Willenz said. "What stands out here is the depth of the team and the discipline to solve perception as a system — where lasting value is created."
The company has not disclosed its valuation, but a vertically integrated hardware-software play with $107 million in the bank and a founding team that has already scaled one sensing platform to a billion users carries weight.
Industry recognition and the market window
Lyte debuted publicly at CES 2026, where it won Best of Innovation in Robotics and honors in Vehicle Tech and Advanced Mobility — selected from a record 3,600 submissions. The timing aligns with a market that Grand View Research projects will reach $125 billion by 2030 for AI-driven robotics.
That figure reflects a structural shift. Industrial automation is no longer optional for manufacturers facing labor shortages and rising wage costs. But adoption depends on reducing integration complexity — exactly the problem Lyte targets at the perception layer.
Lyte's first design-win announcements with robot OEMs — proof that OEMs adopt vertically integrated perception over multi-vendor stacks
Competing perception platforms from established sensor vendors — validation of the category
Deployment metrics: time from integration to production deployment vs. the industry baseline of months
What comes next
Lyte plans to scale its platform into North America, Europe, and Asia-Pacific by the end of 2026, targeting logistics, manufacturing, and transportation. The company operates a subscription model for software updates and support layered on top of hardware sales — a structure that aligns recurring revenue with the long product cycles of industrial robotics.
The open question is whether robot makers will abandon the multi-vendor approach they have used for years in exchange for a unified stack from a single startup. Lyte's founding team has bet their careers that the answer is yes. They have the track record, the capital, and a Best of Innovation trophy to make the argument credible.
Robots can already see. The question is whether they can see well enough to work alongside humans without constant supervision. Lyte is trying to make that distinction disappear.