Currently, industrial inspections, security patrols, specialized operations, and lightweight robotic platforms are accelerating their commercial deployment. Embodied intelligent devices are setting stringent requirements for visual perception systems—namely, lightweight integration, all-weather imaging capabilities, native AI analytics, and multi-channel synchronous transmission. The camera-plus-encoding-board combinations currently available on the market often suffer from poor imaging consistency, excessive size and weight, cumbersome protocol integration, failure to perform effectively in low-light conditions, and delays or packet loss during multi-channel video transmission—defects that directly hinder robots’ autonomous exploration.Core capabilities including measurement, intelligent analysis, and remote scheduling. Kemorei has launched the CM8805EM mini starlight all-in-one AI module, paired with the CM2004HU encoding control board, as a complete set of vision hardware. This setup establishes a comprehensive visual pipeline—“image acquisition—edge-side AI inference—encoding conversion—multi-terminal output”—providing a lightweight, highly stable, and easy-to-develop standardized vision infrastructure for various robotic AI applications.

I. CM8805EM Miniature Zoom Starlight Module: A Lightweight, All-Weather AI Perception Terminal for Robotics

The robot gimbal, end-effector, and small payload module all face stringent limitations on payload space and self-weight. The CM8805EM integrated mechanism weighs only 78 grams and measures just 40×47×74.1mm. Its compact size allows seamless integration into various compact robotic platforms, simplifying the overall mechanical design, reducing cable routing complexity, and lowering both assembly and subsequent maintenance costs. The imaging core is equipped with a Sony 1/2.8-inch STARVIS CMOS sensor, delivering exceptional imaging performance in ultra-low-light conditions—color images at 0.05 lux and black-and-white images at 0.005 lux. In low-light scenarios such as underground tunnels, nighttime factory sites, or unlit machine rooms, it produces pristine 8-megapixel 4K high-definition video. The device incorporates multiple advanced image optimization algorithms: electronic image stabilization effectively counteracts camera shake caused by robotic movement; temperature-compensated focusing ensures sharp images even under varying temperatures; and features such as wide dynamic range, strong-light suppression, electronic defogging, and thermal wave elimination work in tandem to maintain stable image quality in challenging conditions like backlighting, smog, and sudden temperature changes, providing AI algorithms with high-fidelity, raw image data. The optical lens offers a focal length range of 3.3–16.5mm, supporting 5x optical zoom plus 16x digital zoom. The horizontal field of view spans from a wide-angle coverage of 103.1° down to a telephoto mode of 19.1°. The wide-angle mode enables rapid scanning of the entire environment, while the telephoto mode captures fine details at long distances. A full zoom transition can be completed in as little as 2 seconds, eliminating the need for multiple lenses stacked together. The device features built-in deep-learning AI algorithms that natively recognize four types of targets: people, vehicles, smoking, and phone calls. It also supports perimeter detection functions including tripwire crossing, area intrusion, abandoned objects, loitering individuals, and crowd gatherings. All alarm signals are generated directly on the device side without consuming the robot’s main control unit’s computational resources, enabling autonomous AI-driven tasks such as self-tracking, risk prediction, and intelligent obstacle avoidance. The transmission channels employ a dual-output architecture featuring MIPI and network interfaces. The MIPI interface delivers lossless 1080P@30fps video streams for real-time inference by the local controller, while the network port can transmit up to 4K@30fps video streams to a remote management platform. The device is compatible with VISCA and PELCO gimbal control protocols, conforms to the national GB28181 and ONVIF standards, and provides an open HTTP CGI interface, significantly reducing secondary development and integration costs for robotic systems.

II. CM2004HU Encoding Control Board: The Central Hub for Transferring Multi-Format Video Signals in Robots

A single camera module can only handle image acquisition. For robotic AI applications, there’s a need for multi-channel video distribution, unified serial-port control, and synchronized preview capabilities across multiple terminals. The CM2004HU encoding and control board serves as the complementary intermediate hardware in this complete solution, offering perfect compatibility with the entire CM series of camera modules and Sony 4K imaging devices, thereby addressing critical gaps in visual-link transmission and scheduling. The video interfaces fully support 4K high-definition specifications: HDMI supports 4K@30Hz input and loop-out, while the USB2.0 port delivers up to 4K@30Hz video output. This allows simultaneous distribution of multiple video streams, meeting the triple demands of local AI computation on the robot, real-time preview at the ground control station, and cloud-based video storage. The board features a dedicated onboard analog serial port that synchronously transmits back camera zoom parameters, gimbal attitude data, and AI alarm event information. Image, control, and alert signals share a single set of wiring, simplifying the internal wiring layout of the robot. Measuring just 50mm×50mm, the board is equipped with standard 2.1mm mounting holes, enabling direct, flush installation into gimbals or robot body housings without adding extra weight to the device. Thanks to its standardized hardware architecture, no complex driver debugging is required—simply power it on, and the board will seamlessly perform the entire workflow, including image encoding, signal forwarding, and peripheral device integration, significantly shortening both prototype debugging and mass-production cycles. To ensure reliable performance in the challenging electromagnetic environments typical of outdoor mobile robotics applications, the board has been optimized for anti-interference, stably handling 4K high-definition data streams without any transmission lag or frame drops, thus guaranteeing uninterrupted AI object recognition and continuous tracking.

III. A comprehensive software-hardware integration solution that addresses the four major pain points in the practical implementation of robotic AI vision.

Currently, the implementation of robot vision projects commonly faces four major core challenges: First, small onboard spaces limit the deployment of traditional zoom lenses, which are too bulky to simultaneously support wide-range inspections and detailed recognition at long distances; second, image quality deteriorates in nighttime, vibrating, or backlit environments, directly reducing the accuracy of on-device AI recognition; third, separate devices for data acquisition, encoding, and transmission lead to complicated cabling and prolonged integration and debugging cycles; fourth, a single video output channel cannot simultaneously meet the dual demands of local real-time computation and remote backend monitoring. The CM8805EM+CM2004HU modular kit from Kaimorui precisely addresses these pain points. The CM8805EM features a miniature, lightweight body, all-weather starlight imaging, and built-in edge-side AI capabilities, enabling comprehensive environmental perception across all scenarios. Meanwhile, the CM2004HU establishes seamless connections among multiple types of video output channels and a unified control link. The software and hardware of these two modules are deeply integrated, requiring no additional modifications or adaptations—thus, a single set of hardware can cover the entire visual workflow of robots. This solution is suitable for a variety of applications, including factory inspection robots, campus security patrol robots, lightweight payload platforms, and visual inspection equipment for industrial robotic arms. The entire system supports stable operation over a wide temperature range—from -30℃ to +60℃—making it ideal for diverse operational environments such as outdoor open spaces, low-temperature workshops, and enclosed equipment compartments.

IV. Fully open standardized interfaces to empower rapid iteration and mass production of robotics AI projects.

The entire visual module adopts an open and compatible design. The CM8805EM is equipped with an SDK, HTTP CGI, and national standard protocols, and supports local storage via MicroSD cards, enabling event capture and abnormal video recording. The CM2004HU features a standardized board interface that is compatible with mainstream robot control units available on the market. The hardware provides ample room for custom expansion, allowing flexible adjustments to AI detection rules, stream specifications, and video output channels according to project requirements, making it well-suited for various customized robot AI development projects. From R&D prototype testing and small-batch pilot production to large-scale project implementation, this complete module set effectively shortens the development, debugging, and integration cycles of vision systems, while significantly reducing the cost of modifying the overall robot hardware. Leveraging four core advantages—starlight-level all-weather imaging, native edge-side intelligent analysis, multi-channel parallel transmission, and lightweight integration—the CM8805EM+CM2004HU dual-module solution from Kaimorui has become the preferred hardware solution for robotic AI visual perception, providing stable and reliable underlying visual support for the large-scale deployment of embodied intelligence applications.