Guangjian Technology Launches Embodied AI Visual Perception Solution with Libra Binocular Vision Modules

Guangjian Technology announced on July 20 the launch of an embodied AI visual perception solution, a comprehensive vision system centered on AI-powered binocular vision technology that integrates intelligent perception algorithms with efficient computing architecture. The system is designed to provide reliable visual perception for robots operating in real-world environments including homes, factories, and commercial spaces.

As multimodal large models and VLA (Vision-Language-Action) technology rapidly advance, robots are gaining stronger reasoning and decision-making capabilities. However, Guangjian notes that what ultimately determines embodied AI's real-world utility is not just how smart the brain is, but whether it can accurately perceive and understand the physical world — autonomous navigation, precision grasping, and human-robot interaction all depend on reliable 3D vision perception.

The company integrates AI algorithms into binocular stereo matching through a proprietary high-performance algorithm, achieving a fundamental upgrade in spatial perception. Unlike traditional binocular vision that relies on local texture matching, AI binocular vision better understands scene structure and object relationships, maintaining stable depth recovery even with transparent glass, highly reflective metal, black surfaces, and other challenging materials, achieving millimeter-level detail reconstruction.

The solution adopts a universal perception design that supports depth sensing, SLAM positioning, object detection, and other tasks on a unified hardware platform with OTA upgrade capability. By optimizing the perception computing architecture, it significantly reduces compute requirements, freeing up resources for core tasks like VLA inference and SLAM positioning.

The accompanying Libra series includes two AI binocular vision modules: the Libra 1000 features a 2cm baseline for wrist-mounted precision operations such as robotic arm grasping, dexterous hand manipulation, teleoperation, and data collection, supporting fisheye lenses for over 180-degree horizontal FOV; the Libra 3000 features a 7cm baseline for head-mounted spatial perception, targeting mid-to-long-range obstacle avoidance, SLAM positioning, and spatial mapping, with integrated spatial perception algorithms. Both modules support GMSL high-speed output and are compatible with mainstream robot platforms.

This launch marks Guangjian's strategic push into the physical AI era. As embodied AI expands into real-world scenarios, visual perception is evolving from a single hardware capability into critical infrastructure that supports autonomous perception, decision-making, and execution for robots.