Essential details
Shipping:Express Delivery
Product Introduction
The RS-188 is a dedicated multi-modal data acquisition camera engineered for embodied AI robot large model training and embodied intelligence dataset construction. Delivering first-person egocentric perspective capture, this head-mountable stereo device integrates dual 2MP global shutter image sensors, a 6-axis inertial measurement unit and synchronized dual microphones, producing fully time-aligned visual, inertial and acoustic data required for modern embodied AI systems.
With a combined native RGB resolution of 3840×1200 and hardware-level synchronization across all sensing modules, the RS188 eliminates labor-intensive manual temporal alignment during dataset preprocessing. Factory-calibrated stereo intrinsic and extrinsic parameters enable out-of-the-box depth and spatial data output, making it perfectly suited for building training datasets for visual-motor policy learning, embodied perception, SLAM navigation and multi-modal embodied large models. Its standard UVC protocol and USB-C interface ensure seamless integration with robotics development platforms running Windows, Android and Linux.
Core Advantages for Embodied AI Large Model Training
- Global Shutter Imaging for Dynamic Embodied Interaction
Dual 1/2.8-inch global shutter sensors completely eliminate rolling shutter distortion, capturing geometrically precise, blur-free frames even during high-speed robot motion, object manipulation and dynamic environmental interaction. Featuring 2.8μm × 2.8μm pixels and 65.3dB wide dynamic range, it preserves fine-grained visual detail in high-contrast real-world scenes, delivering high-fidelity training data that improves the generalization and robustness of embodied perception models.
- Tightly Coupled Visual-Inertial Data for Motor Policy Learning
A built-in 6-axis IMU (3-axis accelerometer + 3-axis gyroscope) outputs real-time acceleration and angular velocity data, perfectly hardware-synchronized with the dual camera stream. This tightly coupled visual-inertial dataset is foundational for training robot pose estimation, motion control and end-to-end visual-motor policy models, removing the need for complex post-hoc temporal alignment and reducing dataset preprocessing workload for embodied AI teams.
- Multi-Modal Audio-Visual Data for Embodied Intelligence
Dual integrated microphones capture stereo environmental audio fully aligned with visual and motion data, enabling the construction of complete multi-modal datasets for embodied large models. This supports the development of multi-modal embodied AI systems capable of joint visual-audio scene understanding, human-robot interaction and acoustic event perception, expanding the perceptual dimension of robot intelligence.
- Offline Standalone Recording for Real-World Embodied Datasets
Onboard TF card storage up to 256GB supports long-duration standalone data recording without a host computer, ideal for real-world field data collection in unstructured environments. Embodied robot test platforms or wearable human demonstration rigs can capture hours of continuous multi-modal interaction data, supporting large-scale real-world dataset construction to improve the zero-shot generalization of embodied large models.
- Factory Stereo Calibration for Accelerated Model Iteration
Every RS188 unit undergoes factory calibration of binocular intrinsic and extrinsic parameters, delivering pixel-level stereo alignment out of the box. This eliminates time-consuming on-site calibration work, reduces data preprocessing overhead and allows embodied AI teams to immediately generate usable stereo depth and spatial data, significantly shortening the cycle from data collection to model training iteration.
Optical Design Optimized for Embodied Perception
The RS188 is equipped with high-precision 5G structured lenses in a standard M12×P0.5 mount, with a total optical track length of 15.56mm. The fixed-focus design delivers stable, consistent imaging across typical robot working and interaction distances, while the ultra-wide 125° diagonal field of view captures broad egocentric environmental footage — ideal for training embodied navigation, scene understanding and object manipulation models. Lenses are precision-aligned during manufacturing to maintain long-term stereo accuracy and consistent dataset quality.
Optical Design Optimized for Embodied Perception
The RS188 is equipped with high-precision 5G structured lenses in a standard M12×P0.5 mount, with a total optical track length of 15.56mm. The fixed-focus design delivers stable, consistent imaging across typical robot working and interaction distances, while the ultra-wide 125° diagonal field of view captures broad egocentric environmental footage — ideal for training embodied navigation, scene understanding and object manipulation models. Lenses are precision-aligned during manufacturing to maintain long-term stereo accuracy and consistent dataset quality.
Primary Applications in Embodied AI & Robot Large Models
- Multi-modal training dataset construction for embodied AI large models
- Human demonstration data collection for robot manipulation skill learning
- Visual-motor policy and end-to-end robot control model training
- Human-robot interaction scenario data acquisition for interactive embodied intelligence
- Embodied SLAM and autonomous navigation algorithm development and validation
- Real-world unstructured scene data for embodied model generalization improvement
- Egocentric vision dataset building for embodied perception research
Product details
