Inspection robot and inspection method thereof

Abstract

The invention discloses an inspection robot and an inspection method thereof, and relates to the technical field of robots, and the inspection robot comprises a remote control center which is in communication connection with an environment perception and mapping module, an intelligent movement and obstacle avoidance control module, and an intelligent analysis and user interaction module; the environment sensing and mapping module integrates a laser radar, an ultrasonic sensor and an RGB-D camera, and realizes multi-dimensional environment data acquisition; the intelligent movement and obstacle avoidance control module adopts an SLAM algorithm to position and map, adopts an RRT algorithm to plan a path, adopts a CNN algorithm to recognize an obstacle, and adjusts movement through a PID controller and a kinematic model to realize intelligent obstacle avoidance; the intelligent analysis and user interaction module reduces delay through edge calculation, provides real-time data processing and a graphical user interface and displays the state and operation guide of the robot. On the whole, by integrating various sensors and advanced algorithms, efficient and intelligent inspection operation is realized, and the inspection efficiency and safety are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of robots, and in particular to an inspection robot and an inspection method thereof. Background Art

[0002] Inspection robots are intelligent devices that use artificial intelligence technology and autonomous navigation systems and have inspection, monitoring, alarm and other functions. Through high-precision sensors, cameras and advanced algorithms, they can comprehensively and accurately monitor equipment and process flows to ensure the safety of the production environment and improve production efficiency. Against the background of the continuous improvement of industrial automation, inspection robots have been widely used in industries such as industry, energy, transportation, and construction, and play an important role. They can inspect around the clock and are not restricted by environmental factors such as temperature and humidity. They can accurately identify potential failure risks and issue early warnings in a timely manner to effectively avoid production accidents. In addition, inspection robots can also be combined with drones and virtual reality technology to achieve more comprehensive and accurate monitoring and detection. In the future, with the continuous advancement of technologies such as artificial intelligence and machine learning, inspection robots will have stronger intelligence and autonomous decision-making capabilities, and will be able to more accurately judge equipment status and failure risks, and improve the efficiency and accuracy of maintenance. At the same time, with the rapid development of network technology and cloud computing, inspection robots can achieve remote monitoring, data sharing and real-time prediction, and provide comprehensive inspection services on a larger scale. In short, inspection robots and their inspection methods, as an innovative intelligent device and technology, are gradually changing the way and efficiency of inspection work, bringing great changes and advantages to various fields.

[0003] However, in actual use, the problem of insufficient obstacle avoidance of inspection robots is still prominent. When performing tasks, inspection robots often need to travel in complex environments, such as narrow passages and dense pipelines, which place high demands on the robot's obstacle avoidance ability. At present, some inspection robots often find it difficult to respond quickly and accurately when facing complex obstacles, resulting in collisions, jams and other problems, which not only affects the efficiency of inspections, but may also cause damage to the equipment itself. The reasons for insufficient obstacle avoidance may include low sensor accuracy, insufficient algorithm optimization, and poor environmental adaptability. In order to solve this problem, it is necessary to combine multiple perception technologies such as lidar, ultrasonic sensors, and visual recognition to build a three-dimensional map, enhance the understanding and reaction speed of the surrounding environment, and use deep learning algorithms to train robots to identify various types of obstacles so that they can better cope with various complex environments. In short, insufficient obstacle avoidance is an important problem facing the current development of inspection robots. Only by continuously improving the technical level can we ensure that inspection robots can perform tasks stably and efficiently in various environments and provide better support for the development of various fields. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides an inspection robot and an inspection method thereof, which solve the problems in the above-mentioned background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: an inspection robot, comprising a remote control center, wherein the remote control center is communicatively connected to an environment perception and mapping module, an intelligent movement and obstacle avoidance control module, and an intelligent analysis and user interaction module;

[0006] The environment perception and mapping module is used to obtain point cloud data of obstacle distance, ultrasonic data of close-range obstacle distance, and color image and depth information, wherein the point cloud data of obstacle distance is collected and measured by laser radar, the ultrasonic data of close-range obstacle distance is collected by ultrasonic sensor, and the color image and depth information are captured by RGB-D camera;

[0007] The intelligent movement and obstacle avoidance control module is used to construct a map of the inspection area, identify different types of obstacles, and dynamically plan the inspection path to implement obstacle avoidance measures. The SLAM algorithm is used to synchronize positioning and map construction, the CNN algorithm is used to train the robot to identify different types of obstacles, the RRT algorithm is used to dynamically plan the path, the PID controller and kinematic model are used to adjust the robot speed and direction, and the obstacle avoidance command is responded to to avoid obstacles;

[0008] The intelligent analysis and user interaction module processes data instantly on-site through edge computing, establishes a graphical user interface, and provides real-time status feedback and operation guidance.

[0009] Furthermore, in the environment perception and mapping module, the process of collecting point cloud data for measuring obstacle distances through a laser radar includes:

[0010] In the inspection robot, a laser radar is deployed on the top of the inspection robot to obtain the distance information of surrounding objects by emitting laser beams and receiving reflected signals to form point cloud data;

[0011] The collected point cloud data is preprocessed to remove abnormal data points caused by environmental factors and sensor errors, and the data at different scanning angles and time points are integrated into the same coordinate system. Then, key geometric features are identified from the preprocessed point cloud data, including planes, edges and corners. The extracted feature information is then transmitted to the remote control center.

[0012] Furthermore, in the environment perception and mapping module, the process of collecting ultrasonic data for measuring the distance of close-range obstacles by using an ultrasonic sensor includes:

[0013] In the inspection robot, ultrasonic sensors are deployed on the outside of the inspection robot to cover the short-range detection range;

[0014] After the ultrasonic sensor transmits sound waves and receives reflected signals, it obtains distance measurement data and pre-processes the collected distance measurement data to remove erroneous readings caused by multiple reflections, compensate for temperature effects, and perform filtering;

[0015] This is achieved by setting a threshold to identify significant distance change points, thereby identifying plane boundaries and protruding objects. The distance change points represent the edges of obstacles, and the extracted feature information is then transmitted to the remote control center.

[0016] Furthermore, in the environment perception and mapping module, the process of capturing color images and depth information through an RGB-D camera includes:

[0017] In the inspection robot, an RGB-D camera is installed in front of the inspection robot to capture the three-dimensional structure of the environment and obtain color images and depth information;

[0018] The collected data is preprocessed. For color images, the preprocessing includes color correction, denoising and smoothing operations. For depth information, filtering is performed to remove invalid points that are too far or too close, and holes are filled. The color image and depth information are aligned, and high-level semantic features of the preprocessed image are extracted through edge detection and corner detection. For depth information, plane fitting and surface reconstruction techniques are used to identify geometric features in the environment, and the extracted feature information is then transmitted to the remote control center.

[0019] Furthermore, in the intelligent movement and obstacle avoidance control module, the process of synchronous positioning and map construction using the SLAM algorithm includes:

[0020] The features extracted from the preprocessed point cloud data, ultrasonic data, and RGB-D camera data are matched with the features in the imported map to determine the position changes of the inspection robot. The extended Kalman filter is used to combine the feature matching results and the odometer motion model to estimate the current position and posture of the robot. The RGB-D camera uses the visual inertial odometer to improve the positioning accuracy. According to the latest position estimation and observation data, the environmental map is continuously updated, new features are added to the global map, and the accumulated position error is corrected through closed-loop detection.

[0021] Furthermore, in the intelligent mobility and obstacle avoidance control module, the process of dynamically planning a path using the RRT algorithm includes:

[0022] The remote control center sets the inspection target point according to the task requirements. For each target point, the RRT algorithm starts to build a search tree and randomly expands the path nodes from the starting position to the target position. The RRT algorithm randomly selects sampling points in the environment and tries to connect to the nearest existing path node. If the new node does not collide with an obstacle, it is added to the path tree and the above process is repeated until a feasible path from the starting point to the target point is found.

[0023] As the inspection robot moves along the planned path, it continuously receives new data from the environmental perception module. When encountering unforeseen obstacles, the RRT algorithm re-evaluates the current path and generates obstacle avoidance instructions to find new detour solutions. The RRT algorithm smoothes and optimizes the generated path. The final path is sent to the intelligent movement and obstacle avoidance control module to guide the robot to move along the optimal route.

[0024] Furthermore, in the intelligent movement and obstacle avoidance control module, the process of using the CNN algorithm to train the robot to recognize different types of obstacles includes:

[0025] The images and distance information of different types of obstacles collected by RGB-D cameras, lidars, and ultrasonic sensors are annotated, the obstacle types in each image are labeled, and a CNN model is trained using the annotated data set. The CNN model includes convolutional layers, pooling layers, and fully connected layers.

[0026] The convolutional layer first organizes the color image and depth information from the RGB-D camera into three color channels of red, green, and blue plus a depth channel. It then converts the point cloud data collected by the lidar into a 2D grid map and performs a convolution operation with the RGB-D image. The close-range obstacle distance data provided by the ultrasonic sensor is used as supplementary information and added to the feature map at the front end of the network.

[0027] The pooling layer selects the maximum value in each 2x2 window as the output through the maximum pooling method, which keeps the key visual features of the RGB-D image while reducing the computational burden and simplifies the geometric feature representation of long-distance obstacles in the lidar point cloud data. The pooling strategy is used to remove noise from low-resolution ultrasonic sensor data and retain close-range information.

[0028] The fully connected layer receives the high-dimensional feature vectors flattened from the convolution and pooling layers, integrates the multimodal sensor data through dense connections, and forms a comprehensive feature representation. Each neuron is connected to all neurons in the previous layer, learns the complex relationship between features, and outputs obstacle classification and distance estimation results. It adopts a lightweight network architecture and pruning technology to reduce the number of parameters and computational burden, uses backpropagation to optimize the loss function, minimizes the classification error, and applies the softmax activation function for multi-category probability output. It adapts to new environmental conditions through transfer learning, and analyzes early, intermediate, and late fusion strategies. After training, the CNN model is deployed to the inspection robot to perform reasoning tasks online. When the robot moves, it continuously receives multimodal sensor data and quickly identifies the type of obstacles in the image through the trained CNN model.

[0029] Furthermore, in the intelligent movement and obstacle avoidance control module, the process of adjusting the robot speed and direction through the PID controller and the kinematic model and responding to the obstacle avoidance instruction includes:

[0030] After receiving the obstacle avoidance command, the PID controller adjusts the linear speed and angular velocity of the inspection robot, and calculates the control command to drive the motor by comparing the error between the target speed and the actual measured value. Among them, the proportional term directly reflects the current error, the integral term accumulates past errors and eliminates static deviations, and the differential term predicts future changes and suppresses overshoot and oscillation, thereby achieving accurate and smooth speed and direction adjustment. The PID controller works based on a predefined kinematic model, which describes the geometric relationship of the robot's wheels and its influence on the overall motion, so that the control command is converted into physical action to make the robot move according to the expected trajectory.

[0031] Furthermore, in the intelligent analysis and user interaction module, the process of processing data instantly on-site through edge computing, establishing a graphical user interface, and providing real-time status feedback and operation guidance includes:

[0032] The edge computing device is deployed locally on the inspection robot to process raw data from lidar, ultrasonic sensors and RGB-D cameras. The edge computing device is equipped with a high-performance processor and dedicated algorithms to perform data preprocessing, feature extraction and advanced analysis locally. The real-time operating system runs on the edge computing node of the edge computing device to prioritize obstacle avoidance command response and path planning tasks. The graphical user interface is generated and updated through local edge computing, which directly displays the processed environmental perception information and robot status, and provides operation guides and real-time feedback.

[0033] Furthermore, a patrol robot inspection method is implemented based on the above-mentioned patrol robot, comprising the following steps:

[0034] S1, collects point cloud data for measuring obstacle distances through lidar, collects ultrasonic data for measuring close-range obstacle distances through ultrasonic sensors, and captures color images and depth information through RGB-D cameras;

[0035] S2, using SLAM algorithm to synchronize positioning and map construction, and RRT algorithm to dynamically plan the path;

[0036] S3, use CNN algorithm to train the robot to recognize different types of obstacles;

[0037] S4, adjust the robot speed and direction through PID controller and kinematic model to respond to obstacle avoidance instructions;

[0038] S5. Process data instantly on-site through edge computing, establish a graphical user interface, and provide real-time status feedback and operation guides.

[0039] The present invention provides an inspection robot and an inspection method thereof, which have the following beneficial effects:

[0040] First, by integrating laser radar, ultrasonic sensor and RGB-D camera, the environmental perception and mapping module can collect environmental data in an all-round and high-precision manner, providing a solid foundation for the robot's autonomous navigation and obstacle avoidance. This not only improves the robot's adaptability in complex environments, but also ensures the accuracy and safety of inspection operations. Secondly, the intelligent movement and obstacle avoidance control module adopts advanced SLAM algorithm for synchronous positioning and map construction, combined with RRT algorithm for dynamic path planning, and CNN algorithm for obstacle recognition. The application of this series of algorithms enables the robot to autonomously navigate in unknown and dynamically changing environments, effectively avoid collisions, and greatly improve the inspection efficiency. In addition, the intelligent analysis and user interaction module realizes real-time data processing through edge computing technology, significantly reduces delays, and improves user experience. The establishment of a graphical user interface allows users to intuitively understand the real-time status of the robot and remotely control it according to the operating guide, thereby realizing seamless interaction between man and machine. In summary, the present invention integrates a variety of advanced technologies and algorithms, not only improves the autonomous navigation and obstacle avoidance capabilities of the inspection robot, but also improves the efficiency and safety of inspection operations, while enhancing user experience. These beneficial effects make the inspection robot of the present invention have broad promotion value and application prospects in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a block diagram of an inspection robot according to an embodiment of the present application.

[0042] Figure 2 This is a flow chart of an inspection robot according to an embodiment of the present application. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] like Figure 1 and Figure 2 As shown, the present invention provides a technical solution: a patrol robot, comprising a remote control center, wherein the remote control center is communicatively connected to an environment perception and mapping module, an intelligent movement and obstacle avoidance control module, and an intelligent analysis and user interaction module;

[0045] The environment perception and mapping module is used to obtain point cloud data of obstacle distance, ultrasonic data of close-range obstacle distance, and color image and depth information, wherein the point cloud data of obstacle distance is collected and measured by laser radar, the ultrasonic data of close-range obstacle distance is collected by ultrasonic sensor, and the color image and depth information are captured by RGB-D camera;

[0046] The intelligent movement and obstacle avoidance control module is used to construct a map of the inspection area, identify different types of obstacles, and dynamically plan the inspection path to implement obstacle avoidance measures. The SLAM algorithm is used to synchronize positioning and map construction, the CNN algorithm is used to train the robot to identify different types of obstacles, the RRT algorithm is used to dynamically plan the path, the PID controller and kinematic model are used to adjust the robot speed and direction, and the obstacle avoidance command is responded to to avoid obstacles;

[0047] The intelligent analysis and user interaction module processes data instantly on-site through edge computing, establishes a graphical user interface, and provides real-time status feedback and operation guidance.

[0048] In the environment perception and mapping module, the process of collecting point cloud data for measuring obstacle distances through a laser radar includes:

[0049] In the inspection robot, a laser radar is deployed on the top of the inspection robot to obtain the distance information of surrounding objects by emitting laser beams and receiving reflected signals to form point cloud data;

[0050] The collected point cloud data is preprocessed to remove abnormal data points caused by environmental factors and sensor errors, and the data at different scanning angles and time points are integrated into the same coordinate system. Then, key geometric features are identified from the preprocessed point cloud data, including planes, edges and corners. The extracted feature information is then transmitted to the remote control center.

[0051] In the environment perception and mapping module, the process of collecting ultrasonic data for measuring the distance of close-range obstacles through ultrasonic sensors includes:

[0052] In the inspection robot, ultrasonic sensors are deployed on the outside of the inspection robot to cover the short-range detection range;

[0053] After the ultrasonic sensor transmits sound waves and receives reflected signals, it obtains distance measurement data and pre-processes the collected distance measurement data to remove erroneous readings caused by multiple reflections, compensate for temperature effects, and perform filtering;

[0054] This is achieved by setting a threshold to identify significant distance change points, thereby identifying plane boundaries and protruding objects. The distance change points represent the edges of obstacles, and the extracted feature information is then transmitted to the remote control center.

[0055] In the environment perception and mapping module, the process of capturing color images and depth information through an RGB-D camera includes:

[0056] In the inspection robot, an RGB-D camera is installed in front of the inspection robot to capture the three-dimensional structure of the environment and obtain color images and depth information;

[0057] The collected data is preprocessed. For color images, the preprocessing includes color correction, denoising and smoothing operations. For depth information, filtering is performed to remove invalid points that are too far or too close, and holes are filled. The color image and depth information are aligned, and high-level semantic features of the preprocessed image are extracted through edge detection and corner detection. For depth information, plane fitting and surface reconstruction techniques are used to identify geometric features in the environment, and the extracted feature information is then transmitted to the remote control center.

[0058] In the intelligent movement and obstacle avoidance control module, the process of synchronous positioning and map construction using the SLAM algorithm includes:

[0059] The features extracted from the preprocessed point cloud data, ultrasonic data, and RGB-D camera data are matched with the features in the imported map to determine the position changes of the inspection robot. The extended Kalman filter is used to combine the feature matching results and the odometer motion model to estimate the current position and posture of the robot. The RGB-D camera uses the visual inertial odometer to improve the positioning accuracy. According to the latest position estimation and observation data, the environmental map is continuously updated, new features are added to the global map, and the accumulated position error is corrected through closed-loop detection.

[0060] In the intelligent mobility and obstacle avoidance control module, the process of dynamically planning the path using the RRT algorithm includes:

[0061] The remote control center sets the inspection target point according to the task requirements. For each target point, the RRT algorithm starts to build a search tree and randomly expands the path nodes from the starting position to the target position. The RRT algorithm randomly selects sampling points in the environment and tries to connect to the nearest existing path node. If the new node does not collide with an obstacle, it is added to the path tree and the above process is repeated until a feasible path from the starting point to the target point is found.

[0062] As the inspection robot moves along the planned path, it continuously receives new data from the environmental perception module. When encountering unforeseen obstacles, the RRT algorithm re-evaluates the current path and generates obstacle avoidance instructions to find new detour solutions. The RRT algorithm smoothes and optimizes the generated path. The final path is sent to the intelligent movement and obstacle avoidance control module to guide the robot to move along the optimal route.

[0063] In the intelligent movement and obstacle avoidance control module, the process of using the CNN algorithm to train the robot to recognize different types of obstacles includes:

[0064] The images and distance information of different types of obstacles collected by RGB-D cameras, lidars, and ultrasonic sensors are annotated, the obstacle types in each image are labeled, and a CNN model is trained using the annotated data set. The CNN model includes convolutional layers, pooling layers, and fully connected layers.

[0065] The convolutional layer first organizes the color image and depth information from the RGB-D camera into three color channels of red, green, and blue plus a depth channel. It then converts the point cloud data collected by the lidar into a 2D grid map and performs a convolution operation with the RGB-D image. The close-range obstacle distance data provided by the ultrasonic sensor is used as supplementary information and added to the feature map at the front end of the network.

[0066] The pooling layer selects the maximum value in each 2x2 window as the output through the maximum pooling method, which keeps the key visual features of the RGB-D image while reducing the computational burden and simplifies the geometric feature representation of long-distance obstacles in the lidar point cloud data. The pooling strategy is used to remove noise from low-resolution ultrasonic sensor data and retain close-range information.

[0067] The fully connected layer receives the high-dimensional feature vectors flattened from the convolution and pooling layers, integrates the multimodal sensor data through dense connections, and forms a comprehensive feature representation. Each neuron is connected to all neurons in the previous layer, learns the complex relationship between features, and outputs obstacle classification and distance estimation results. It adopts a lightweight network architecture and pruning technology to reduce the number of parameters and computational burden, uses backpropagation to optimize the loss function, minimizes the classification error, and applies the softmax activation function for multi-category probability output. It adapts to new environmental conditions through transfer learning, and analyzes early, intermediate, and late fusion strategies. After training, the CNN model is deployed to the inspection robot to perform reasoning tasks online. When the robot moves, it continuously receives multimodal sensor data and quickly identifies the type of obstacles in the image through the trained CNN model.

[0068] In the intelligent movement and obstacle avoidance control module, the process of adjusting the robot speed and direction through the PID controller and the kinematic model and responding to the obstacle avoidance command includes:

[0069] After receiving the obstacle avoidance command, the PID controller adjusts the linear speed and angular velocity of the inspection robot, and calculates the control command to drive the motor by comparing the error between the target speed and the actual measured value. Among them, the proportional term directly reflects the current error, the integral term accumulates past errors and eliminates static deviations, and the differential term predicts future changes and suppresses overshoot and oscillation, thereby achieving accurate and smooth speed and direction adjustment. The PID controller works based on a predefined kinematic model, which describes the geometric relationship of the robot's wheels and its influence on the overall motion, so that the control command is converted into physical action to make the robot move according to the expected trajectory.

[0070] In the intelligent analysis and user interaction module, the process of processing data on-site in real time through edge computing, establishing a graphical user interface, and providing real-time status feedback and operation guidance includes:

[0071] The edge computing device is deployed locally on the inspection robot to process raw data from lidar, ultrasonic sensors and RGB-D cameras. The edge computing device is equipped with a high-performance processor and dedicated algorithms to perform data preprocessing, feature extraction and advanced analysis locally. The real-time operating system runs on the edge computing node of the edge computing device to prioritize obstacle avoidance command response and path planning tasks. The graphical user interface is generated and updated through local edge computing, which directly displays the processed environmental perception information and robot status, and provides operation guides and real-time feedback.

[0072] A patrol robot inspection method is implemented based on the above-mentioned patrol robot, comprising the following steps:

[0073] S1, collects point cloud data for measuring obstacle distances through lidar, collects ultrasonic data for measuring close-range obstacle distances through ultrasonic sensors, and captures color images and depth information through RGB-D cameras;

[0074] S2, using SLAM algorithm to synchronize positioning and map construction, and RRT algorithm to dynamically plan the path;

[0075] S3, use CNN algorithm to train the robot to recognize different types of obstacles;

[0076] S4, adjust the robot speed and direction through PID controller and kinematic model to respond to obstacle avoidance instructions;

[0077] S5. Process data instantly on-site through edge computing, establish a graphical user interface, and provide real-time status feedback and operation guides.

[0078] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. Without further restrictions. The sentence "comprises an element defined by ... does not exclude the existence of other identical elements in the process, method, article or device including the element". Although embodiments of the present invention have been shown and described, it is understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the attached claims and their equivalents.

Claims

1. A patrol robot, characterized in that: It includes a remote control center, which is communicatively connected to an environment perception and mapping module, an intelligent movement and obstacle avoidance control module, and an intelligent analysis and user interaction module; The environment perception and mapping module is used to obtain point cloud data of obstacle distance, ultrasonic data of close-range obstacle distance, and color image and depth information, wherein the point cloud data of obstacle distance is collected and measured by laser radar, the ultrasonic data of close-range obstacle distance is collected by ultrasonic sensor, and the color image and depth information are captured by RGB-D camera; The intelligent movement and obstacle avoidance control module is used to construct a map of the inspection area, identify different types of obstacles, and dynamically plan the inspection path to implement obstacle avoidance measures. The SLAM algorithm is used to synchronize positioning and map construction, the CNN algorithm is used to train the robot to identify different types of obstacles, the RRT algorithm is used to dynamically plan the path, the PID controller and kinematic model are used to adjust the robot speed and direction, and the obstacle avoidance command is responded to to avoid obstacles; The intelligent analysis and user interaction module processes data instantly on-site through edge computing, establishes a graphical user interface, and provides real-time status feedback and operation guidance.

2. The inspection robot according to claim 1, characterized in that: In the environment perception and mapping module, the process of collecting point cloud data for measuring obstacle distances through a laser radar includes: In the inspection robot, a laser radar is deployed on the top of the inspection robot to obtain the distance information of surrounding objects by emitting laser beams and receiving reflected signals to form point cloud data; The collected point cloud data is preprocessed to remove abnormal data points caused by environmental factors and sensor errors, and the data at different scanning angles and time points are integrated into the same coordinate system. Then, key geometric features are identified from the preprocessed point cloud data, including planes, edges and corners. The extracted feature information is then transmitted to the remote control center.

3. The inspection robot according to claim 2, characterized in that: In the environment perception and mapping module, the process of collecting ultrasonic data for measuring the distance of close-range obstacles through ultrasonic sensors includes: In the inspection robot, ultrasonic sensors are deployed on the outside of the inspection robot to cover the short-range detection range; After the ultrasonic sensor transmits sound waves and receives reflected signals, it obtains distance measurement data and pre-processes the collected distance measurement data to remove erroneous readings caused by multiple reflections, compensate for temperature effects, and perform filtering; This is achieved by setting a threshold to identify significant distance change points, thereby identifying plane boundaries and protruding objects. The distance change points represent the edges of obstacles, and the extracted feature information is then transmitted to the remote control center.

4. The inspection robot according to claim 3, characterized in that: In the environment perception and mapping module, the process of capturing color images and depth information through an RGB-D camera includes: In the inspection robot, an RGB-D camera is installed in front of the inspection robot to capture the three-dimensional structure of the environment and obtain color images and depth information; The collected data is preprocessed. For color images, the preprocessing includes color correction, denoising and smoothing operations. For depth information, filtering is performed to remove invalid points that are too far or too close, and holes are filled. The color image and depth information are aligned, and high-level semantic features of the preprocessed image are extracted through edge detection and corner detection. For depth information, plane fitting and surface reconstruction techniques are used to identify geometric features in the environment, and the extracted feature information is then transmitted to the remote control center.

5. The inspection robot according to claim 4, characterized in that: In the intelligent movement and obstacle avoidance control module, the process of synchronous positioning and map construction using the SLAM algorithm includes: The features extracted from the preprocessed point cloud data, ultrasonic data, and RGB-D camera data are matched with the features in the imported map to determine the position changes of the inspection robot. The extended Kalman filter is used to combine the feature matching results and the odometer motion model to estimate the current position and posture of the robot. The RGB-D camera uses the visual inertial odometer to improve the positioning accuracy. According to the latest position estimation and observation data, the environmental map is continuously updated, new features are added to the global map, and the accumulated position error is corrected through closed-loop detection.

6. The inspection robot according to claim 5, characterized in that: In the intelligent mobility and obstacle avoidance control module, the process of dynamically planning the path using the RRT algorithm includes: The remote control center sets the inspection target point according to the task requirements. For each target point, the RRT algorithm starts to build a search tree and randomly expands the path nodes from the starting position to the target position. The RRT algorithm randomly selects sampling points in the environment and tries to connect to the nearest existing path node. If the new node does not collide with an obstacle, it is added to the path tree and the above process is repeated until a feasible path from the starting point to the target point is found. As the inspection robot moves along the planned path, it continuously receives new data from the environmental perception module. When encountering unforeseen obstacles, the RRT algorithm re-evaluates the current path and generates obstacle avoidance instructions to find new detour solutions. The RRT algorithm smoothes and optimizes the generated path. The final path is sent to the intelligent movement and obstacle avoidance control module to guide the robot to move along the optimal route.

7. The inspection robot according to claim 6, characterized in that: In the intelligent movement and obstacle avoidance control module, the process of using the CNN algorithm to train the robot to recognize different types of obstacles includes: The images and distance information of different types of obstacles collected by RGB-D cameras, lidars, and ultrasonic sensors are annotated, the obstacle types in each image are labeled, and a CNN model is trained using the annotated data set. The CNN model includes convolutional layers, pooling layers, and fully connected layers. The convolutional layer first organizes the color image and depth information from the RGB-D camera into three color channels of red, green, and blue plus a depth channel, then converts the point cloud data collected by the lidar into a 2D grid map, performs convolution operations synchronously with the RGB-D image, and uses the close-range obstacle distance data provided by the ultrasonic sensor as supplementary information and adds it to the feature map at the front end of the network; The pooling layer selects the maximum value in each 2x2 window as the output through the maximum pooling method, which keeps the key visual features of the RGB-D image while reducing the computational burden and simplifies the geometric feature representation of long-distance obstacles in the lidar point cloud data. The pooling strategy is used to remove noise from low-resolution ultrasonic sensor data and retain close-range information. The fully connected layer receives the high-dimensional feature vectors flattened from the convolution and pooling layers, integrates the multimodal sensor data through dense connections, and forms a comprehensive feature representation. Each neuron is connected to all neurons in the previous layer, learns the complex relationship between features, and outputs obstacle classification and distance estimation results. It adopts a lightweight network architecture and pruning technology to reduce the number of parameters and computational burden, uses backpropagation to optimize the loss function, minimizes the classification error, and applies the softmax activation function for multi-category probability output. It adapts to new environmental conditions through transfer learning, and analyzes early, intermediate, and late fusion strategies. After training, the CNN model is deployed to the inspection robot to perform reasoning tasks online. When the robot moves, it continuously receives multimodal sensor data and quickly identifies the type of obstacles in the image through the trained CNN model.

8. The inspection robot according to claim 7, characterized in that: In the intelligent movement and obstacle avoidance control module, the process of adjusting the robot speed and direction through the PID controller and the kinematic model and responding to the obstacle avoidance command includes: After receiving the obstacle avoidance command, the PID controller adjusts the linear speed and angular velocity of the inspection robot, and calculates the control command to drive the motor by comparing the error between the target speed and the actual measured value. Among them, the proportional term directly reflects the current error, the integral term accumulates past errors and eliminates static deviations, and the differential term predicts future changes and suppresses overshoot and oscillation. The PID controller works based on a predefined kinematic model, which describes the geometric relationship of the robot's wheels and its influence on the overall motion, so that the control command is converted into physical action to make the robot move according to the expected trajectory.

9. The inspection robot according to claim 8, characterized in that: In the intelligent analysis and user interaction module, the process of processing data on-site in real time through edge computing, establishing a graphical user interface, and providing real-time status feedback and operation guidance includes: Edge computing devices are deployed locally on the inspection robot to process raw data from lidar, ultrasonic sensors, and RGB-D cameras. The edge computing devices are equipped with high-performance processors and dedicated algorithms to perform data preprocessing, feature extraction, and advanced analysis locally. Running a real-time operating system on the edge computing node of the edge computing device gives priority to obstacle avoidance command response and path planning tasks. A graphical user interface is generated and updated through local edge computing, which directly displays the processed environmental perception information and robot status, and provides operation guides and real-time feedback.

10. A patrol robot inspection method, implemented based on the patrol robot according to any one of claims 1 to 9, characterized in that: The following steps are involved: S1, collects point cloud data for measuring obstacle distances through lidar, collects ultrasonic data for measuring close-range obstacle distances through ultrasonic sensors, and captures color images and depth information through RGB-D cameras; S2, using SLAM algorithm to synchronize positioning and map construction, and RRT algorithm to dynamically plan the path; S3, use CNN algorithm to train the robot to recognize different types of obstacles; S4, adjust the robot speed and direction through PID controller and kinematic model to respond to obstacle avoidance instructions; S5. Process data instantly on-site through edge computing, establish a graphical user interface, and provide real-time status feedback and operation guides.

Images