1. Introduction With the development of robotics technology, automated cleaning of high-rise building curtain walls has become possible. The "Development of a Cleaning Robot for Complex Curved Curtain Walls" project, funded by the 863 Program, focuses on cleaning the metal and glass roof of the National Centre for the Performing Arts (NCPA) located on Chang'an Avenue in Beijing. Situated in the heart of Beijing, the NCPA is poised to become a landmark building, and given the harsh climate of northern China, the cleaning of its exposed exterior walls is crucial. Applying the robot system to high-rise curtain wall cleaning operations under challenging conditions necessitates higher safety and reliability for engineering applications. Therefore, a robust control system is essential. This paper, focusing on the characteristics of high-altitude curved surface operations, introduces the mechanical components of a self-climbing robot and details the robot's control hardware and software structure. 2. Structural Characteristics of the National Centre for the Performing Arts and Robot Design The NCPA's main structure is semi-ellipsoidal, with exterior walls covered by glass and titanium alloy panels, totaling 36,000 m². The titanium panels on the NCPA come in seven different specifications, ranging in width from 2.2 m to 1.5 m. Closed aluminum guide rails are distributed between each layer of titanium plates along the spherical latitude direction. The guide rails are 25mm wide and 40mm high above the plate surface, and the guide rails on the glass and titanium plates are continuous. There are horizontal and vertical gaps between the titanium plates, and decorative hemispherical structures are installed in the gaps. The self-climbing robot scheme is proposed based on the structural characteristics of the building. The robot body is divided into climbing mechanism, driving mechanism, cleaning mechanism and pitch adjustment mechanism according to function. The robot prototype [2] is shown in Figure 1. It is about 3 meters long, 1 meter wide and 0.5 meters high. The body is built of lightweight aluminum profiles. The main frame and the rear housing generate relative climbing motion through synchronous belt drive. The main drive is installed at the front end of the main frame, and the auxiliary drive is installed between two sliding guide rails 1000mm long at the rear end of the main frame. The auxiliary drive can passively slide within this range to adapt to changes in the length of the building panels. The main and auxiliary drives cooperate with the building guide rails during robot movement, providing driving force through friction wheels. The brush module can move up and down relative to the rear housing and the main frame perpendicular to the wall, and can move synchronously with the rear housing along the main frame. The front and rear pitch support adjustment mechanisms are mainly used to adjust the robot's posture during climbing to adapt to changes in the angle between each floor panel of the building, and serve as a movable force fulcrum to improve force distribution during robot climbing. The bottom of the main frame adopts a boat-shaped structure, which matches the shape of the sliding guide rod installed on the building guide rails. With the coordinated cooperation of the front and rear clamping components, the safety of robot climbing and movement is further ensured. The front and rear clamping components have the same structure. When this mechanism is activated, the robot can reliably grip the sliding guide rod and mechanically lock it to ensure the safety of the robot working at heights. The main technical specifications of the robot are as follows: maximum working height 50m; maximum crawling speed 200mm/s; cleaning efficiency >800m²/day; robot body weight <150Kg. 3. Design of the Self-Climbing Robot Control System. The CAN bus is a serial data communication protocol developed by Bosch in Germany in the early 1980s to solve the data exchange problem between numerous control and testing instruments in modern automobiles. Its excellent characteristics, high reliability, and unique design make it particularly suitable for interconnecting industrial process monitoring equipment. Therefore, it has received increasing attention from the industry and is recognized as one of the most promising fieldbuses. Based on the motion function requirements and the robot's mechanical structure, the robot controller's hardware system adopts a distributed CAN bus network structure, as shown in Figure 2. The system is divided into 6 parts and 5 CAN bus nodes: main frame general control node, rear housing control node, climbing and pitch control node, main and auxiliary drive control node, main control computer node, and remote control operation part. 3.1 Control System Structure Distribution The main control computer (onboard IPC), as the central node, forms the core of the entire system. It utilizes a PCM-9575 single-board computer, which is powerful, low-power, and compact. It embeds a low-power VIA Ezra 800M processor, capable of operating at 60℃ without a fan, with a typical power consumption of 14W, and supports the PC/104 bus. A PCM3680 CAN interface adapter is used to connect to the PCM-9575 via the PC/104 bus, and wireless communication uses an ADAM-4550 transceiver module. The main control computer is a high-level intelligent module and does not directly participate in the lower-level control. It receives motion commands from the remote control box in real time, performs motion planning and control scheduling for the robot, and sends commands to the four lower-level nodes, including the main frame controller, via the CAN bus to control and coordinate the work of these four nodes. On the other hand, it feeds back the robot's status information to the operator for monitoring. The main control computer is also equipped with an image processing card to process the video signals provided by the robot's CCD camera, used to detect the cleanliness of the building's guide rails and surfaces. The remote control box also uses the ADAM-4550 module for wireless communication with the main control computer. The main functions of the control box are simple planning and monitoring. It simultaneously receives and displays image information captured by the wireless CCD camera, allowing the operator to monitor and intervene in the robot based on the images and the information returned by the main control computer. The main frame controller, rear housing controller, climbing and pitch controller, and main and auxiliary drive controller are low-level control modules responsible for driving the corresponding DC motors. These modules are primarily coordinated via CAN bus communication. The four nodes and the human-machine interface circuit in the control box were developed in-house, using a P80C592 microcontroller as the core to form a distributed node controller. The hardware structure of each controller differs slightly to allow for future expansion of system functionality. The P80C592 features abundant input and output ports; it employs an 80C51 central processing unit (CPU); external ROM expandable to 64kB; 2 × 256 bytes of on-chip RAM, expandable externally to 64kB; an 8-channel analog input 10-bit ADC converter; and an on-chip monitor-track timer (WDT). These features fully meet the design requirements of the node function. The built-in CAN controller increases system reliability, further enhancing integration and reducing controller size. Taking the climbing and pitch controller as an example, the hardware structure block diagram of the controller node is shown in Figure 3. The climbing and pitch controller controls three DC servo motors of the climbing mechanism and the front and rear pitch support adjustment mechanisms. The climbing controller encapsulates all the necessary detection information to complete the control function, mainly including the detection of the guide rail during climbing, the detection of the upper and lower limit positions of the climbing motion, the detection of the contact state between the main frame and the sliding rod, the detection of the fit between the boat-shaped plate and the sliding rod groove, the detection of the parallelism between the main frame and the curtain wall, the detection of the contact between the front and rear adjustment mechanisms and the curtain wall, and corresponding processing. 3.2 Sensor Configuration The system employs multiple sensors to detect the relative state between the robot and its environment, as well as the robot's motion. The controller achieves motion positioning and completes local autonomous intelligent control by comprehensively processing and analyzing (fusion) sensor signals such as distance, material, obstacles, and displacement, ensuring effective cleaning and smooth movement. The configuration of the robot's external sensors is directly related to the working environment. External information on buildings mainly includes aluminum guide rails, sliding guide rods, decorative lights, tall obstacles, large areas of dirt, and small amounts of water stains. Considering the actual requirements of the operation, due to the adoption of a climbing mechanism, the robot frame is in a non-contact state with the wall surface, except for the pitch adjustment support. Obstacles such as decorative lights do not affect the robot's operation. Small amounts of water stains will not have an adverse effect on cleaning, while dirt and dust are the direct targets of the cleaning operation. Therefore, the external information processed by the system mainly includes aluminum guide rails, sliding guide rods, and tall obstacles. Through the detection of these three types of obstacles, the controller processes the data to complete the geometric reconstruction of the working environment. In the control system, contact switches and photoelectric sensors are used to detect tall obstacles, while ultrasonic sensors and CCD cameras are used to detect aluminum guide rails, sliding guide rods, and suspended objects. Additionally, the CCD camera allows operators to monitor the robot's overall operational status in real time via wireless image transmission, enabling necessary intervention. Internal sensors are used to measure the robot's own state. Displacement sensors employ incremental encoders, with interrupt-based input via channels on the control board; proximity switches are used to detect various motion limit states. The energy system is an indispensable component. The National Centre for the Performing Arts has a massive structure, and obstacles exist along its longitudinal axis, such as dragging cables, which would inevitably result in long, heavy, and easily damaged cables, potentially causing pollution and scratches to the building's surface. Therefore, lithium batteries are chosen as the robot's power source due to their cleanliness, convenience, high energy storage efficiency, and light weight. A power monitoring section is also added to the detection system. 4. Robot Control Software The robot's operation process is divided into four parts: self-check, initialization, operation, and information feedback. The software structure is shown in Figure 4. The motion planning and operation management section, based on the requirements of the upper-level planning and accurate local environmental information, makes decisions based on a comprehensive judgment of the working environment information, selects the most reasonable trajectory to respond to the environmental state for synthesis, and calls the motion module. The local environment model provided by sensor detection information is the basis for selection or scheduling. The motion control adopts a modular design, and the motion modules are relatively independent. Each module can be independently coded, tested, debugged, or modified, thereby simplifying complex tasks. All movements required for the robot to complete the task can be combined from basic action modules according to certain logical relationships. After planning, the software enters the output drive control stage, specifically allocating, executing, and managing the action sequence, and ultimately forming various state information, which serves as the basis for fault diagnosis and handling. While the system is processing, all operation results are output to the operation box for visualization. 4. Conclusion This paper introduces the control system structure, controller and hardware composition, and software structure characteristics of the National Centre for the Performing Arts curtain wall cleaning robot. Engineering experiments show that it has high reliability, good performance, and safe operation.