LG Innotek is teaming up with NVIDIA to develop the sensing systems that humanoid robots will use as eyes and hands. The South Korean component maker said on October 22 that it had unveiled camera and composite sensing systems built on NVIDIA’s Holoscan Sensor Bridge technology at GTC Berlin, and that the two companies will expand their cooperation into next-generation vision sensing for humanoid robots and robot hands.
The announcement makes LG Innotek the latest Korean supplier to be pulled into NVIDIA’s physical AI orbit during the Berlin conference, but on the perception side of the stack rather than in chip packaging: what LG is selling is how a robot sees, not how its processor is mounted.
What LG Innotek Showed At GTC Berlin
The lineup has two parts. The first is a 3-megapixel camera module built on Holoscan Sensor Bridge (HSB). The second is a multimodal vision system, or MVS, that combines cameras with other sensors — including LiDAR and radar — into one perception package that processes the different feeds jointly, so a robot recognises objects and spaces more accurately than any single sensor would allow.
LG Innotek said the pair will expand their work from the products shown in Berlin to vision sensing for humanoid robots and for robot hands, developing sensing solutions that interoperate with NVIDIA’s next-generation platform. The follow-on systems are due to be unveiled at CES in Las Vegas and at NVIDIA’s GTC next year.
“This collaboration is a great opportunity to spread LG Innotek’s differentiated composite sensing solutions in the global market,” president Moon Hyuk-soo said. “As a ‘physical AI total solution provider’ encompassing not only sensing but also communication modules and semiconductor substrates, we will move quickly to secure leadership in the physical AI solutions market.”
How Holoscan Sensor Bridge Speeds Up Robot Perception
HSB is NVIDIA’s high-speed sensor-data path. Instead of routing what cameras, LiDAR and radar collect through the central processing unit first, it transmits the data directly to GPU memory on the AI computer. Shortening that path cuts information-processing latency, which is what lets robots and autonomous vehicles perceive their surroundings and judge a response quickly enough to act on it.
That latency is the practical bottleneck for physical AI. A humanoid that has to see a handhold, judge it and close its fingers around it cannot afford a perception pipeline that queues behind general-purpose computing, and LG’s pitch is that aligning its modules with NVIDIA’s platform standard from the outset raises the odds that robot makers building on that platform choose its cameras and sensor packages.
Why It Matters
LG Innotek’s deal matters because it shows where the contest for the humanoid supply chain is being fought: not only over models and computers, but over the standardised sensing layer every robot body needs. NVIDIA already used GTC Berlin to push its Isaac platform out to builders, including a reference humanoid built for Agile ONE SN, and to put its stack into working warehouses, as in the robotic returns loop Sereact is closing with NVIDIA and Nebius at CEVA and Zalando. A sensor supplier whose hardware speaks NVIDIA’s fastest data path natively becomes a default component in that ecosystem.
For LG Innotek, the cooperation is also a springboard it intends to use openly: the company says it will expand business opportunities with global physical AI companies that build on NVIDIA’s platform, and diversify what it supplies them beyond cameras and sensors to communication modules and semiconductor substrates. Perception is the entry product; the component basket is the prize.
Background
LG Innotek built its name on smartphone camera modules and autonomous-driving sensors, and has been repointing that optical and composite-sensing expertise at robotics for over a year: it agreed in May last year to cooperate with Boston Dynamics on vision sensing systems for robots, signed a memorandum of understanding with Japan’s TDK on physical AI cooperation in July this year, and began mass production of camera modules for a North American robotics customer at its Paju plant the same month. In autonomous driving it holds a strategic partnership with US LiDAR company Aeva and is pursuing joint development with Applied Intuition and Kakao Mobility. Its NVIDIA sensing work lands in a Berlin week when the chipmaker also launched its OpenShell reference architecture for safe AI agents, underlining how much of the physical AI stack NVIDIA wants running on its standards.
Source: Chosun Biz, The Korea Times
0 Comments