Trossen Robotics and Stereolabs are advancing Physical AI data collection through a new robotics hardware collaboration. The companies are integrating Stereolabs cameras into Trossen’s latest Physical AI platforms. Trossen Robotics provides research and data-collection platforms for Physical AI. Stereolabs SAS operates as a wholly owned subsidiary of Ouster, Inc. (Nasdaq: OUST).
Stereolabs develops the ZED series of AI stereo cameras. Through the collaboration, Trossen will integrate the ZED X Mini and ZED X Nano cameras into its Physical AI hardware suite. The new lineup targets robot learning and data collection applications. It includes the Trossen Workbench and Rivet manipulation platforms.
Physical AI systems depend heavily on the quality of their training data. Most of that training data comes from visual information collected during robot demonstrations. However, many teams still build custom data-collection rigs. They often combine robot arms with consumer-grade USB cameras and custom driver software.
This approach can create several data-quality challenges. Low-resolution images can reduce the value of collected demonstrations. In addition, inconsistent calibration can affect data accuracy across different camera views. Motion blur can also reduce the quality of visual information during fast robotic movements.
Timing drift creates another challenge when systems combine multiple data streams. Therefore, robotics teams need more reliable and synchronized vision systems.
Trossen Platforms Combine Robotics and Stereo Vision
The Trossen Workbench and Rivet provide unified development platforms for Physical AI research. Both systems feature dual WidowX Pro 6-DoF arms. The robotic arms offer reach ranging from 700 mm to 1000 mm. They also provide payload capacities between 4 kg and 6 kg.
Each platform delivers 1 mm repeatability for robotic manipulation tasks. Furthermore, each system includes onboard NVIDIA Jetson AGX Orin 64 GB computing. The platforms support teleoperation across local networks and the internet. They also include factory-calibrated three-camera vision systems.
Trossen built the vision architecture entirely with Stereolabs hardware. The configuration uses one ZED X Mini and two ZED X Nano cameras. The ZED X Mini sits at the center of the workspace. It provides a broad scene view and stereo-depth information across the operating area.
Meanwhile, each ZED X Nano mounts near a robotic wrist. These cameras capture close-range views of the gripper and objects. This setup reflects how advanced manipulation policies receive visual information. The system combines workspace context with detailed views needed for precise manipulation.
Consequently, robotics teams can collect synchronized visual data during teleoperated demonstrations.
ZED X Nano Supports Training-Grade Robot Data
Stereolabs designed the ZED X Nano for demanding robotic applications. Its dual 2.3 MP global-shutter sensors capture images at up to 60 frames per second. The global-shutter design helps reduce motion blur during rapid robotic movements. This capability provides cleaner visual information than many rolling-shutter USB cameras.
The camera also uses a neural depth engine to resolve geometry from distances as close as 3 cm. Such close-range depth information supports robotic grasping and manipulation tasks. Furthermore, the cameras use GMSL2 connectivity with locking and EMI-resistant cabling. This configuration keeps the three-camera system synchronized with the onboard Jetson computer.
The architecture also uses a zero-copy pipeline. Therefore, teams can record, encode, and run inference at the same time. The system helps prevent silent frame losses during individual demonstration episodes. An onboard vibration-resistant IMU also supports wrist-mounted camera operation.
These capabilities improve the consistency of visual data collected during robot movements. As a result, teams can build stronger datasets for Physical AI model development. Every teleoperated episode collected through the Workbench and Rivet produces synchronized multi-view training data. The data combines RGB and depth information.
Trossen and Stereolabs Target Industrial Physical AI
Teams can use these datasets for imitation learning and reinforcement learning workflows. The platforms also support sim-to-real development. Native support for ROS 2 further connects the systems with established robotics development workflows. The platforms also support NVIDIA Isaac Sim and Isaac Lab.
“The Physical AI community is migrating to GMSL2 because USB can’t handle the long cable runs from the end effector to compute that real robots demand,” said Matt Trossen, CEO from Trossen Robotics. “Stereolabs ZED X Nano gives us the signal stability, image quality, and throughput to take Physical AI from the lab into hardened industrial deployments. Teams should spend their time collecting demonstrations and training policies, not integrating and calibrating camera rigs.”
The collaboration focuses on reducing the complexity involved in robotic data collection. Instead of assembling separate camera and robotics components, teams receive an integrated platform.
“Trossen has done what few others have: put the camera at the heart of a complete, calibrated data-collection system,” said Cecile Schmollgruber, President of Stereolabs. “The Workbench with Stereolabs ZED X Mini and ZED X Nano turns every demonstration into training-grade data, and that’s what will move Physical AI forward.”
The Workbench forms part of Trossen’s wider Physical AI hardware portfolio. The portfolio also includes the RIVET mobile manipulation platform. It also includes GLIDE passive leader arms and the COCKPIT operator station. Together, these systems support scalable teleoperated data collection.
Trossen will showcase the hardware suite at its Physical AI Residency in San Francisco. The program runs from August 12 through August 28, 2026. The company will also demonstrate live data collection and policy evaluation at the Actuate conference. The event takes place in San Francisco from August 18 through August 19, 2026.
The collaboration brings together robotic manipulation hardware and high-fidelity stereo vision. Therefore, the partnership aims to improve the quality and reliability of data used to train Physical AI systems.
Explore IT Tech News for the latest advancements in Information Technology & insightful updates from industry experts!
News Source: Businesswire.com
