Urban underground utility tunnels are the “underground lifelines” of a city’s operations—spaces that are enclosed, humid, unevenly lit, filled with moisture, and characterized by narrow confines. Traditional manual inspections of these tunnels are inefficient and pose significant safety risks, including exposure to toxic gases, standing water, and potential collapses. AI-powered inspection robotic dogs can venture into these confined spaces—areas that are difficult or even impossible for humans to reach—and carry out routine inspections. However, challenging operating conditions such as low light levels, interference from moisture, fluctuating wireless networks, and constant movement and vibration place extremely high demands on visual data acquisition, intelligent analysis, and efficient encoding and transmission. The CM8810E starlight smart imaging module, paired with the CM204UE encoding control unit, forms a comprehensive visual solution that integrates image capture, AI-based analysis, encoding and transmission, and protocol adaptation. This solution empowers AI inspection robotic dogs to see clearly, make accurate judgments, and transmit data reliably, thereby driving the transition toward unmanned, highly precise operation and maintenance of underground utility tunnels.

I.Star-level imaging performance, suitable for the complex and variable lighting conditions in tunnel corridors.

As the core onboard perception unit for robotic dogs, the CM8810E is equipped with a 1/2.8" Sony Starvis CMOS sensor that delivers color imaging at an ultra-low illumination level of 0.01 Lux and black-and-white imaging down to 0.001 Lux. In environments such as tunnel corridors where lighting is either absent or sparse, this camera can capture clear images without the need for additional illumination. The ICR infrared filter enables seamless switching between day and night modes, supporting independent configuration of separate parameter sets for day and night operations, thus effectively handling dramatic changes in brightness caused by direct sunlight or large areas of shadow within the corridor.

Equipped with 4.8-48The mm lens features 10x optical zoom, enabling wide-angle scans for comprehensive coverage of tunnel corridors. The telephoto lens allows for long-distance magnification, making it possible to detect subtle defects such as pipe corrosion, interface leaks, and meter readings with precision—without requiring the robotic dog to approach hazardous areas. Equipped with electronic anti-fog, strong-light suppression, backlight compensation, and wide dynamic range functions, this device effectively counteracts image degradation caused by moisture, light glare, and extreme contrast between bright and dark areas. Electronic image stabilization compensates for vibrations and jolts encountered when the robotic dog navigates obstacles, ensuring a steady and stable image output. The entire device weighs only 152g.-30Operates within a wide temperature range of ℃~+60℃; its compact body is well-suited for the limited installation space of robotic dog pods and is ideal for use in tunnel corridors with high humidity and significant temperature fluctuations.

II.Edge-side intelligent analysis empowers robotic dogs to autonomously identify on-site risks.

The CM8810E features powerful built-in behavioral intelligence analysis capabilities, reducing the computational load on backend servers in tunnel inspection scenarios. It natively supports algorithms for tripwire detection, area intrusion detection, loitering detection, detection of abandoned or moved objects, and crowd gathering detection, and can identify targets such as people and vehicles. During autonomous cruising, if the robotic dog detects unauthorized intrusions or abnormal lingering, the device can immediately issue an intelligent alarm signal.

The triple-stream design adapts to fluctuations in the bandwidth of tunnel wireless networks, while H.265’s efficient encoding reduces the burden on the stream bitrate. It supports local storage up to 1 TB via TF card, enabling inspection videos to be saved locally even in deep sections of tunnels with poor network signals, and allowing data to be uploaded once the network is restored. Compatible with ONVIF and GB28181 standard protocols, the device seamlessly integrates PTZ control, alarm events, and video streams into operation and maintenance platforms, thereby reducing the complexity of overall system integration and development for robotic dogs.

III.CM204UE encoding control unit, establishing the data transmission link for tunnel corridor scenarios.

The CM204UE serves as the data hub for the entire vision system. This 50x50mm miniature board card receives image signals from the CM8810E, performs encoding conversion, supports multi-channel output, and enables device control, thereby addressing the pain points of single video output and inconvenient debugging associated with robotic dogs used in underground utility tunnels. It supports input resolutions up to 4K, allowing it to receive high-definition images from the core processor without any loss of quality. It also supports 4K.-Dual independent video outputs via HDMI and IP network; the parameters for each output can be configured separately. The local HDMI port is used for on-site debugging, while the 4K stream over the IP network is transmitted back to the remote operations and maintenance backend. A USB 3.0 expansion interface is also provided.

Supports H.264/H.265/MJPEG encoding and triple-stream output, making it suitable for unstable network environments in tunnel corridors. The protocol covers VISCA Over IP, ONVIF, and GB28181, and enables precise coordinated control of lens zoom and focus via multiple hardware interfaces. It supports RTMP streaming and custom OSD information overlay, including marking inspection points and time stamps. Powered by DC12V, with controllable power consumption, it is compatible with the onboard power supply system of robotic dogs.

IV.Integrated software-hardware collaboration accelerates the implementation and deployment of AI-powered tunnel inspection projects.

CM8810E and CM204UE form a complete acquisition system.—Analysis—Encoding—Implement a closed-loop control system to reduce debugging costs associated with assembling multiple components, simplify the structural design of the robotic dog’s pod, and shorten the overall product development cycle. The system closely aligns with the real-world inspection workflows for utility tunnels: during the cruising phase, it performs wide-angle, full-field scanning; when a suspicious target is detected, it quickly switches to high-resolution zoom mode to capture detailed images; in low-light and humid environments, it ensures superior image quality; in case of network disconnection, it locally caches video recordings; and in the background, it receives video feeds and alerts to complete the closed-loop process for addressing potential hazards.

In addition to underground utility tunnels, this solution is also well-suited for enclosed, high-risk environments such as cable tunnels and chemical corridors. By leveraging AI-powered inspection robotic dogs to replace human workers in hazardous areas, the solution significantly reduces personnel safety risks and increases both inspection frequency and the detection rate of potential hazards. With its highly reliable and easily integrable complete set of vision hardware, Kemorui helps underground infrastructure achieve intelligent and safe operations and maintenance.