New Camera Software Makes Robots See Faster Without Lag
This could make self-driving cars and robots react quicker and safer.
StereoLabs just updated its ZED camera software to version 5.6, adding a feature that lets cameras share video data more efficiently with robots and AI systems. Normally, when a camera captures images, the data has to travel from the graphics processor (GPU) to the main processor (CPU), get converted, and then go back to the GPU for processing. That round trip adds delay. The new version uses a special buffer that keeps data on the GPU, so compatible programs can grab it directly. This is like passing a note directly to a friend instead of mailing it through a central post office.
The update is built on technology contributed by NVIDIA to ROS 2 Lyrical, the latest version of the Robot Operating System (a common software framework for robots). It's designed for NVIDIA's Isaac ROS 5, a set of tools for building AI-powered robots. With this release, the ZED camera node publishes data straight from GPU memory, so GPU-accelerated subscribers no longer waste time on the round trip. In tests on a Jetson AGX Thor developer kit, latency from publish to data available on the subscriber's GPU was measured, showing improvement. The exact numbers depend on the hardware and settings.
For the average person, this means robots and autonomous machines could become more responsive. Think of a self-driving car that needs to spot a pedestrian instantly, or a warehouse robot that must avoid obstacles. Faster data transfer means quicker reactions, which can improve safety and efficiency. However, this update is quite technical and mainly benefits developers and companies building such systems. The catch is that it requires specific hardware (like Jetson Thor or Orin) and software (ROS 2 Lyrical and Isaac ROS 5). Also, intra-process subscribers still get CPU data unless you disable IPC, so not all setups will see the speedup. The wrapper supports all StereoLabs cameras, including ZED X and ZED 2 families, and the ZED Box 2 series. Developers can try the new examples to reproduce benchmarks on their own hardware.
- ZED cameras can now send video directly from GPU memory, skipping the slow CPU round trip.
- This reduces latency for robots and AI systems, making them react faster.
- Requires specific hardware (Jetson Thor/Orin) and software (ROS 2 Lyrical, Isaac ROS 5).
Why It Matters
Faster robot vision could lead to safer self-driving cars and more efficient automated machines.