Orbbec and Robbyant, Ant Group’s embodied AI company, are deepening their collaboration on physical-world perception and spatial intelligence, based on the strategic partnership established this January. On July 7, Robbyant released its next-generation spatial perception model, LingBot-Depth 2.0, which has been certified by Orbbec’s Depth Vision Laboratory. Combined with Orbbec’s 3D vision cameras, the model further enhances spatial perception capabilities in complex real-world environments.
Meanwhile, the EGO RGB-D device, as one of the core products within Orbbec’s newly released Robot-Free Data Collection Hardware Platform, is being adapted to a customized LingBot-Depth 2.0 version for physical AI data collection scenarios. Moving forward, Orbbec and Robbyant plan to adopt an advanced commercial version to continuously optimize missing depth completion, object edges, and spatial structures, jointly building a precise, stable, and highly usable real-world data foundation for embodied AI training.
Advancing Real-World Data Collection for Physical AI
As physical AI moves from model validation toward real-world deployment, high-quality real-world interaction data has become essential for model training, iteration, and generalization.
Orbbec’s Robot-Free Data Collection Hardware Platform brings together multiple product forms, including EGO, UMI, and WristCam, covering key scenarios such as first-person observation, wrist-mounted near-field observation, and detailed hand-object interaction capture. The platform is designed for physical AI model companies, robotics companies, data collection service providers, data operators, and other algorithm-driven enterprises, offering standardized products as well as CM and JDM services.
Among these products, the EGO RGB-D device is designed for first-person interaction data collection. Jointly developed by Orbbec and Robbyant, the solution combines chip-level depth output from Orbbec’s 3D vision hardware with model-based depth enhancement powered by LingBot-Depth.
On the hardware side, EGO RGB-D integrates Orbbec’s Gemini 330 series stereo vision 3D camera, which is powered by Orbbec’s proprietary MX6800 depth engine chip for robotics applications. The camera can synchronously capture RGB images and high-precision depth data, delivering low-latency and stable depth output as a reliable data source for physical AI training.
On the model side, EGO RGB-D will support a version of LingBot-Depth 2.0 specifically customized by Robbyant for data collection scenarios. Built on the high-quality raw depth data output by Orbbec hardware, the model further completes missing depth information, refines object edges, and enhances spatial structure details. This is particularly valuable in complex scenarios involving reflective surfaces, transparent objects, occlusion boundaries, and other challenging perception conditions.
By combining 3D vision hardware with spatial perception models, Orbbec and Robbyant aim to improve the quality and usability of real-world interaction data, while accelerating the fusion of AI models and data collection hardware for physical AI applications.
From Model Validation to Integrated Product Deployment
In broader robotic vision applications, the deep integration of high-precision 3D vision hardware and advanced AI models at the edge is becoming another key direction for Orbbec and Robbyant’s collaboration.
The newly released LingBot-Depth 2.0 was trained on a dataset of 150 million samples, delivering upgrades in training data scale, edge clarity, small object recognition, long-range depth estimation, and robustness in complex environments. Notably, the model has been professionally certified by Orbbec’s Depth Vision Laboratory.
In benchmark testing, LingBot-Depth 2.0 achieves top rankings in 12 out of 16 depth completion evaluations. In the most challenging indoor scenarios with large-area missing depth, LingBot-Depth 2.0 halves the depth error compared to its predecessor, reducing RMSE from 0.132 to 0.062. The model also demonstrated strong performance in scenarios where traditional depth cameras may face limitations, including glass, mirrors, and transparent objects.
The collaboration between Orbbec and Robbyant dates back to the LingBot-Depth 1.0 stage. In January 2026, Robbyant released its first open-source spatial perception model, LingBot-Depth 1.0, using Orbbec’s Gemini 330 series camera for RGB-D data collection and performance validation. Building on this foundation, the two companies developed a LingBot Enhanced Depth Filter customized for the Gemini 330 series, targeting complex perception challenges such as transparent objects, low-texture or repetitive-pattern surfaces (such as white walls and fences), and highly reflective materials.
In collaboration with Robbyant, Orbbec will launch an SDK product that integrates the latest LingBot-Depth model capabilities, providing robotics customers with edge-side depth enhancement for robots using the Gemini 330 series cameras. Moreover, Orbbec plans to launch an integrated camera product featuring the commercial version of LingBot-Depth by year-end, enabling all-in-one delivery of 3D camera hardware and spatial perception capabilities.
Orbbec, as a leading provider of robotics and AI vision, has provided technologies and products to more than 5,000 enterprise customers worldwide. Moving forward, Orbbec and Robbyant will continue to expand their collaboration in physical-world perception, spatial intelligence, and related fields, supporting the large-scale application and real-world deployment of physical AI technologies.
Full product specs: orbbec.com/robot-free-data-collection/
