Trossen and Stereolabs add stereo vision to Physical AI platforms

Trossen Robotics and Stereolabs, a wholly owned Ouster subsidiary, have factory-integrated ZED X Mini and ZED X Nano cameras into Trossen’s Workbench and Rivet platforms. The setup is designed to produce synchronized, training-grade visual data for robot learning, teleoperation and sim-to-real workflows.

Why it matters: - Physical AI systems need clean, synchronized visual data to train better models and reduce time spent on calibration and integration. - Trossen’s new camera stack is aimed at turning robot demonstrations into usable data for imitation learning, reinforcement learning and sim-to-real workflows. - The integration targets research teams and industrial deployments that need stable close-range and full-scene stereo vision.

What happened: - Trossen Robotics and Stereolabs announced factory integration of the ZED X Mini and ZED X Nano cameras into Trossen’s Workbench and Rivet platforms. - The announcement covers Workbench, a stationary bimanual manipulation platform, and Rivet, a mobile bimanual manipulation platform. - Stereolabs is a wholly owned subsidiary of Ouster, Inc. (Nasdaq: OUST). - Trossen said the suite will be demonstrated at Trossen’s Physical AI Residency in San Francisco from August 12–28, 2026, and at Actuate on August 18–19, 2026.

The details: - Workbench and Rivet each use dual WidowX Pro 6-DoF arms with 700–1,000 mm reach, 4–6 kg payload capacity and 1 mm repeatability. - Each platform includes NVIDIA Jetson AGX Orin 64 GB compute, local and internet teleoperation, and a factory-calibrated three-camera Stereolabs vision system. - A center-mounted ZED X Mini provides a global stereo-depth view. - Two wrist-mounted ZED X Nano cameras capture close-range views of grippers and objects for fine manipulation. - The ZED X Nano uses dual 2.3 MP, 1920x1200 global-shutter sensors and captures at up to 60 fps. - The ZED X Nano neural depth engine resolves geometry from 3 cm. - GMSL2 connectivity, locking EMI-resistant cabling and a zero-copy pipeline keep the three cameras synchronized on the Jetson during recording, encoding and inference. - An onboard IMU supports stable wrist views during arm motion. - Each teleoperated episode produces synchronized, multi-view RGB-plus-depth data. - The systems support native ROS 2, NVIDIA Isaac Sim and Isaac Lab. - Workbench anchors a broader suite that also includes Rivet, Glide passive leader arms and the Cockpit operator station. - Trossen Robotics develops modular hardware and software platforms for Physical AI research, data collection and deployment. - Stereolabs develops 3D vision and depth-perception technology for Physical AI, with ZED cameras and SDK tools for depth sensing, spatial mapping and AI-powered object understanding. - The post includes a LinkedIn social link: Trossen Robotics on LinkedIn.

Between the lines: - The pitch is less about a single camera and more about removing friction from the full data-collection pipeline. - By factory-calibrating the cameras into the robot platforms, Trossen is positioning the hardware as a ready-made training system instead of a custom integration project. - The emphasis on GMSL2 and synchronized capture signals a push toward more reliable data collection than consumer USB camera setups can typically provide.

What's next: - Trossen and Stereolabs will use the residency and conference demos to show the integrated platforms in action. - Adoption will likely depend on whether the combined system delivers consistent, training-grade data in real-world labs and industrial environments. - Ouster’s forward-looking statements note that product performance, customer adoption and integration outcomes remain subject to risk and uncertainty.

The bottom line: - Trossen is bundling cameras, compute and teleoperation into a calibrated Physical AI data-collection stack, with Stereolabs vision at the center.