Ouster Unit Stereolabs Launches Its ZED Box 2 Series Of AI Computers Engineered For Autonomous Mobile Robots, Robot Vision And Spatial AI Systems

ZED Box 2 Max takes vision AI computing to the next level. Powered by NVIDIA® Jetson Thor™ with up to 2,070 FP4 TFLOPS and 128 GB of memory, ZED Box 2 Max is built to run generative AI, vision-language-action (VLA) models and multi-sensor fusion directly on the robot, enabling advanced perception and real-time decision-making at the edge.

Robots are moving from pilots to fleets, and each machine now carries multiple cameras and sensors. At that scale, the interface is a key aspect: GMSL2 has become the interface of choice for robot vision due to its low latency and reliability. Additionally, robotics teams want a complete system instead of separate components: AI compute, hardware synchronized sensors, and software that keeps working release after release.

Up to 8x Cameras for Full Surround Perception

Designed for complex visual environments where a single stereo sensor is insufficient, the ZED Box 2 Series natively supports up to eight GMSL2 camera inputs via dual Quad Fakra-Z connectors. This enables integration of up to eight ZED X cameras for full 360-degree surround vision, multi-angle robotic cell manipulation, and multi-sensor fusion with high resolution digital lidar.

To ensure accurate temporal alignment across high-speed camera streams, the ZED Box 2 Series features hardware-level frame synchronization via a dedicated SYNC + IO interface, supporting internal synchronization and external trigger modes across multiple devices. Capabilities include:

360° perception for AMRs navigating dynamic environments
Multi-view perception for humanoids and manipulation
Large-scale multi-camera video analytics
Robots combining camera vision with lidar and other sensors
Robotic workcells combining viewpoints to eliminate occlusions
Two Models for Different AI Workloads

Both ZED Box 2 models share the same camera interface and software, but differ in compute and cooling. Powered by NVIDIA Jetson Orin™ NX 16GB with up to 157 TOPS, ZED Box 2 runs neural depth, tracking, and multi-camera fusion in real time from a fanless enclosure only 55 mm tall, small enough to disappear inside an AMR or a robot cell. With the NVIDIA Jetson Thor T4000 or T5000 and up to 2070 FP4 TFLOPS, ZED Box 2 Max runs vision-language-action and foundation models directly on the robot. It adds 10 GbE and extra control I/O for humanoids and larger autonomous machines.

Full Range of I/O for All Robot Needs

A robot needs more than cameras to understand its environment. It should be able to connect to lidar, GNSS, IMUs, motor controllers, and fleet networks, and keep all its sensors on a common timeline. The ZED Box 2 Series brings connectivity, synchronization and onboard computing together in one compact platform, without extra adapters or custom carrier boards.

The ZED Box 2 Series is built for every sensor in your robot. Both models feature two Gigabit Ethernet ports, HDMI for on-site debugging, and USB 3 for GNSS receivers, external IMUs and peripherals. ZED Box 2 features two CAN bus ports, sync and GPIO. For more demanding systems, ZED Box 2 Max expands connectivity with four CAN bus ports, four USB ports, four PWM outputs for direct actuator control, and a QSFP port providing four 10 GbE links for lidar and other high-bandwidth sensors.