New Sensing Hardware Speeds Up Humanoid Robot Processing
LG Innotek and NVIDIA aim to slash perception latency using a new sensor bridge that bypasses the CPU for faster AI processing.
Updated on Oct. 22, 2026 in Robotics
Live Poll
Are you optimistic that the rapid development of physical AI robots will improve society?
LG Innotek and NVIDIA have announced a partnership to advance physical AI through integrated sensing solutions. The collaboration features a 3-million-pixel camera module and a multimodal system that consolidates data from cameras, LiDAR, and radar.
Why it matters
By integrating high-speed data transfer directly into robot perception systems, this partnership aims to reduce latency for real-time spatial awareness. This advancement is critical for improving object recognition accuracy in next-generation humanoid robots.
The sensing suite utilizes the NVIDIA Holoscan Sensor Bridge, which bypasses the CPU to move raw sensor data directly into GPU memory. This hardware-level integration supports high-speed processing for multimodal systems that combine camera, LiDAR, and radar inputs.
The players
LG Innotek
A South Korean electronics components manufacturer specializing in optical modules, substrates, and sensor technologies for the global robotics and automotive markets.
NVIDIA
A computing platform company that designs GPUs and hardware-software stacks for high-performance AI, simulation, and robotics applications.
Boston Dynamics
An engineering and robotics company known for developing advanced mobile humanoid and quadrupedal robot platforms.
The details
The multimodal vision system functions by fusing concurrent data streams from disparate sensor types to build a coherent environmental model. By using the Holoscan Sensor Bridge — a specialized interface that enables direct memory access between sensors and AI-processing units — the system minimizes bottlenecks associated with CPU-based data routing. This architecture allows robots to interpret complex spatial information more efficiently than conventional serialized processing methods.
Timeline
May 2025: LG Innotek initiated vision sensing collaboration with Boston Dynamics.
October 20, 2026: The GTC Berlin conference opened.
October 22, 2026: The GTC Berlin conference concluded.
2027: Planned unveiling of further sensing solutions at CES in Las Vegas.
The Tech Race
This partnership formalizes a deepening integration between hardware component manufacturers and NVIDIA's compute-centric robotics stack. It marks a push to move beyond general-purpose computing toward specialized sensor bridges tailored for the real-time constraints of physical AI.
This collaboration primarily impacts developers and manufacturers building autonomous humanoid systems that require high-speed environmental data. While currently at the announcement stage, these modules are designed to integrate into future robot architectures, with mass production of camera components already underway in Paju.
The takeaway
The move signifies a shift toward dedicated hardware pathways for sensor data, moving away from CPU-bound architectures in robotics. Watch for the 2027 CES disclosures to see if these sensing modules achieve improved object-recognition speeds in real-world environments.
What happens next
The companies are scheduled to showcase further developments of these sensing solutions at the 2027 CES event in Las Vegas.
Further reading
Explore ongoing advancements in machine vision and hardware integration on our Robotics page.
Source note: This article includes information reported by 조선일보.
Live Poll
Are you optimistic that the rapid development of physical AI robots will improve society?
