Orbbec Robot-free Data Collection Hardware Platform

The Orbbec Data Collection Kit is designed for robot learning data collection, human demonstration data collection, imitation learning, VLA training, and embodied AI development. It helps teams collect structured human operation data before a robot body, end-effector, or full robotic system is ready.

(1) For customers with an existing device design, our own factory can handle professional CM(Contract Manufacturing) services. Orbbec Factory features industry leading calibration expertise and mass production capacity for millions of multi-sensor vision devices.

(2) For customers seeking a complete device solution, Orbbec provides JDM(Joint Design Manufacturing) services. Products can be launched under the customer’s own brand, while Orbbec manages the full hardware development and production process. Orbbec has rich experience in robotic vision deployment to fully understand your needs.

Orbbec combines multi-sensor synchronization, timestamp synchronization, precise camera calibration, multi-camera consistency, and stable data pipeline design to help maintain aligned, consistent, and high-quality data output. This supports more reliable robot learning data collection across long-duration tasks and complex real-world operation scenarios.

Yes. Orbbec provides SDK and API support for robot-free data collection, as well as support for on-device application deployment. This helps customers integrate data collection hardware into their own Physical AI training, robot learning, or data capture and processing workflows.

Based on the open-source, cross-platform Orbbec SDK, developers can build and customize applications for Android, iOS, Windows, and Linux. Orbbec SDK also supports data access and parameter configuration for depth cameras, RGB cameras, and IMU sensors, along with camera-IMU frame synchronization, point cloud data, and stream processing capabilities.

For specific data collection projects, Orbbec can further provide support for device control, multimodal data capture, calibration data access, data recording, and customized data export based on project requirements.

Ego: Designed for egocentric vision data collection, capturing first-person visual perception, task flow, and real-world scene context. It helps embodied AI and robot learning systems understand how humans perceive environments, perform tasks, and interact with the physical world.

UMI: Designed for manipulation data collection, capturing human demonstrations such as grasping, moving, placing, assembling, and organizing objects. It focuses on how a task is completed and provides valuable manipulation data for embodied AI, robot learning, and real-world task training.

WristCam: Designed for wrist-level, near-field interaction capture, focusing on detailed hand-object and tool-object interactions. It complements egocentric or global views by capturing near-field visual details, making it suitable for fine-grained manipulation, contact-rich tasks, and high-quality embodied AI training data.

It depends on the task complexity and specific data type requirements. (For more guidance on using a single-module, please see the Question above.)

While each module can be used independently, Orbbec recommends a kit-based setup for teams that need to cover multiple data collection scenarios, improve configuration flexibility, and capture more complete human operation data across scene context, manipulation process, and hand-object interaction.

The kit supports diverse data collection scenarios, including household manipulation, industrial assembly, warehouse picking, sorting, tool-use tasks, bimanual operation, human demonstration recording, and VLA training data collection. For scenarios with special wearing structures, sensor layouts, camera angles, or operation environments, Orbbec can also evaluate CM/JDM customization options.