Pavement automatic inspection robot path planning method and device based on artificial intelligence

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a road surface automatic inspection robot path planning method and device based on artificial intelligence, and the method comprises the steps: collecting road surface environment data in real time through various sensors, carrying out the feature extraction through a convolutional neural network based on the road surface environment data, and constructing a dynamic environment model through a graph convolutional network. According to the method, the change trend of potential road surface damage areas and obstacles is predicted, path planning and optimization are performed by using a deep reinforcement learning algorithm, after path planning is completed, the robot autonomously travels according to a planned path, and real-time adaptive adjustment is performed on the path through an online learning method so as to cope with environmental changes. Collected data and path planning information are fed back to the central control unit in real time to be stored, and a road surface health condition report is generated. According to the invention, the efficiency and the precision of the inspection task are improved, and the method has high environmental adaptability and is suitable for complex and dynamically changing road environments.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based path planning method and device for an automatic road inspection robot. Background Art

[0002] With the acceleration of urbanization, the management and maintenance of road infrastructure are particularly important. Traditional road inspection methods usually rely on manual inspections or simple mechanized equipment. Manual inspections are not only inefficient and costly, but are also easily affected by external factors such as weather and geography, resulting in unstable inspection quality. In order to improve the efficiency and accuracy of road inspections, automated inspection technology has gradually been widely used in recent years, especially the use of robots and sensors for road health testing. By integrating various sensor devices (such as lidar, ultrasonic sensors, high-definition cameras, etc.), it is possible to collect road environment data in real time to help accurately locate road damage areas.

[0003] Existing road inspection robot devices have problems such as insufficient path planning accuracy, poor adaptability, and low inspection efficiency. Traditional path planning algorithms usually rely on static environmental models and cannot cope with dynamically changing road environments (such as sudden obstacles or newly appeared road damage), which may cause the robot to fail to avoid obstacles in time and adjust the path slowly during the actual inspection process. In addition, existing technologies are mostly rule-driven, which makes it difficult to effectively cope with complex and dynamically changing environments, and lack the ability to adaptively adjust to environmental changes, resulting in the inspection process being inefficient and inefficient. Summary of the invention

[0004] The present invention provides an artificial intelligence-based path planning method and device for an automatic road inspection robot.

[0005] The path planning method of the automatic road inspection robot based on artificial intelligence includes the following steps:

[0006] S1, road environment data collection: real-time acquisition of road environment data through multiple synchronously working sensors, the sensors include laser radar, ultrasonic sensor and camera, the road environment data includes road surface flatness, road surface cracks, obstacle location and road surface damage degree;

[0007] S2, data preprocessing and feature extraction: preprocess the collected road environment data, including noise filtering and data deduplication, and extract features from the preprocessed road environment data. For the image data obtained by the camera, a convolutional neural network is used to extract features to identify road cracks and potholes. For the laser radar and ultrasonic sensor data, a deep neural network is used to fuse data to extract the location and shape information of road damage areas and obstacles.

[0008] S3, build dynamic environment model: Based on the extracted features, use graph convolutional network to build a dynamic environment model to predict the change trend of potential road damage areas and obstacles;

[0009] S4, path planning and optimization: Based on the constructed dynamic environment model, combined with the robot's current position and target position, a deep reinforcement learning algorithm is used for path planning;

[0010] S5, path tracking and adaptive adjustment: After the path planning is completed, the robot drives autonomously according to the planned path. When the environment changes or new obstacles are detected, the robot uses real-time feedback from sensors and adopts online learning methods to adaptively adjust the path;

[0011] S6, Data storage and analysis: The road environment data and path planning information collected during the inspection are fed back to the central control unit in real time for data storage and analysis to form a road health status report.

[0012] Optionally, the S1 includes:

[0013] S11, sensor selection and configuration: in the road inspection task, multiple sensors working synchronously are selected for data collection, and the sensors include laser radar, ultrasonic sensor and camera;

[0014] S12, sensor placement and data collection: Place lidar, ultrasonic sensors and cameras on the robot to ensure that the sensors cover the inspection area;

[0015] S13, road surface environment data collection: collecting road surface environment data in real time through multiple synchronously working sensors, the data including road surface flatness, road surface cracks, obstacle locations and road surface damage degree;

[0016] S14, data synchronization: The collected road environment data is time-stamped uniformly through a synchronization mechanism and transmitted to the data preprocessing module.

[0017] Optionally, S2 includes:

[0018] S21, data preprocessing: preliminary preprocessing of the collected road environment data, including noise filtering and data deduplication;

[0019] S22, data standardization and normalization: standardize the data collected by different sensors;

[0020] S23, feature extraction: extracting features from the pre-processed road environment data, including:

[0021] S231, image data feature extraction: for the image data collected by the camera, a convolutional neural network is used to extract features. The convolution operation is expressed as:

[0022] F(x,y)=∑ i,j I(i,j)·K(xi,yj);

[0023] Among them, I(i,j) is the pixel value of the input image, K(xi,yj) is the convolution kernel, and F(x,y) is the convolution result, which represents the feature response of the image at a specific position;

[0024] S232, LiDAR and ultrasonic sensor data fusion: Perform deep neural network fusion on the data collected by LiDAR and ultrasonic sensors. By fusing these sensor data, more accurate location and shape information of road damage areas and obstacles can be extracted.

[0025] Optionally, S2 further includes:

[0026] S24, crack and pothole recognition of image data: crack and pothole recognition is performed on the image features processed by the convolutional neural network, and a segmentation method based on a specific threshold is used to extract the location and shape information of the cracks and potholes in the image;

[0027] S25, road damage area and obstacle location identification: Based on the fusion data of LiDAR and ultrasonic sensors, a deep neural network model is used to identify road damage areas and obstacle locations;

[0028] S26, feature vector generation and annotation: annotate and fuse the features in the image data and sensor data to generate a feature vector.

[0029] Optionally, the S3 includes:

[0030] S31, feature map construction: constructing a road environment feature map based on the generated feature vector;

[0031] S32, graph convolutional network model construction: Use the graph convolutional network to process the constructed road environment feature map and build a preliminary dynamic environment model;

[0032] S33, Optimization and updating of dynamic environment model: Based on the initially established dynamic environment model, the dynamic environment model is optimized by dynamically adjusting network parameters and updating node characteristics.

[0033] Optionally, the S4 includes:

[0034] S41, definition of path planning objectives: before path planning, define the objective function of path planning, which is used to quantify the efficiency of the robot inspection task;

[0035] S42, Application of deep reinforcement learning algorithm: Deep reinforcement learning algorithm using deep Q network as path planning;

[0036] S43, path optimization and adjustment: During the path planning process, the path is dynamically adjusted in combination with the robot’s real-time feedback information.

[0037] Optionally, S4 further includes:

[0038] S44, deep Q network training process: using deep Q network for path planning training, including experience replay and Q value update;

[0039] S45, dynamic path adjustment: During the robot inspection process, when the environmental state changes (such as obstacle movement, crack expansion, etc.), the deep Q network adjusts the path planning through real-time feedback;

[0040] S46, evaluation and optimization of path planning results: After the path planning is completed, the robot verifies it through actual inspection tasks and evaluates and optimizes the path based on the inspection results.

[0041] Optionally, the S5 includes:

[0042] S51, path tracking: After the path planning is completed, the inspection robot drives autonomously according to the planned path, and obtains the distance deviation between the robot and the path in real time through sensors installed on the inspection robot (such as laser radar, ultrasonic sensor, camera, etc.);

[0043] S52, environmental change detection: The inspection robot monitors the surrounding environment in real time through sensors and detects changes that affect the path (such as new obstacles, road damage, cracks, etc.);

[0044] S53, Adaptive path adjustment: When the environment changes or new obstacles are detected, the inspection robot uses the real-time feedback of road environment data and online learning methods to adaptively adjust the path;

[0045] S54, path replanning and execution: After the path adjustment is completed, the robot re-controls the motion according to the new path and continues to move along the optimized path.

[0046] Optionally, the S6 includes:

[0047] S62, data storage and management: storing the collected road environment data and path planning information in a database of the central control unit;

[0048] S63, data analysis and report generation: After the data is stored, the central control unit analyzes the road environment data and path planning information and generates a road health status report.

[0049] The artificial intelligence-based automatic road inspection robot path planning device is used to implement the artificial intelligence-based automatic road inspection robot path planning method, including the following modules:

[0050] Data acquisition module: acquires road surface environment data in real time through multiple synchronously working sensors, including laser radar, ultrasonic sensor and camera. The road surface environment data includes road surface flatness, road surface cracks, obstacle location and road surface damage degree;

[0051] Data preprocessing module: preprocess the collected road environment data, including noise filtering and data deduplication, and extract features from the preprocessed road environment data. For the image data obtained by the camera, a convolutional neural network is used to extract features to identify road cracks and potholes. For the laser radar and ultrasonic sensor data, a deep neural network is used to fuse data to extract the location and shape information of road damage areas and obstacles.

[0052] Dynamic environment model building module: Based on the extracted features, a dynamic environment model is built using a graph convolutional network to predict the changing trends of potential road damage areas and obstacles;

[0053] Path planning module: Based on the constructed dynamic environment model, combined with the robot's current position and target position, a deep reinforcement learning algorithm is used for path planning;

[0054] Path tracking and adaptive adjustment module: After the path planning is completed, the robot drives autonomously according to the planned path. When the environment changes or new obstacles are detected, the robot uses real-time feedback from sensors and adopts online learning methods to adaptively adjust the path.

[0055] Data storage and analysis module: The road environment data and path planning information collected during the inspection process are fed back to the central control unit in real time for data storage and analysis to form a road health status report.

[0056] Beneficial effects of the present invention:

[0057] The present invention combines multiple synchronously working sensors (lidar, ultrasonic sensors and cameras) to collect road environment data, and uses a deep reinforcement learning algorithm (deep Q network) for path planning and optimization, ensuring that the robot can cover all important sections of the inspection area in the shortest time while avoiding obstacles, cracks and other damaged areas. This process can significantly improve the efficiency of inspection tasks through real-time path adjustment and adaptive learning, reduce redundant driving and invalid paths, and thus greatly improve the overall effectiveness and accuracy of inspection work.

[0058] The present invention uses a graph convolutional network to construct a dynamic environment model and combines it with a deep Q network for path planning, so that the robot can make dynamic adjustments based on real-time environmental changes (such as the emergence of new obstacles or road damage). Through the online learning mechanism, the robot can correct the path in time according to sensor feedback during the actual inspection process, avoid unnecessary collisions or driving errors, and ensure that the robot always inspects along the optimal path. This adaptive path adjustment capability not only improves the robot's adaptability to complex environments, but also ensures the continuity and efficiency of inspection tasks.

[0059] The present invention feeds back the road environment data and path planning information collected during the inspection process to the central control unit in real time, stores and analyzes the data, and generates a detailed road health status report. The report can accurately reflect key information such as road damage, crack location, obstacle existence and distribution, and provide a scientific basis for subsequent road maintenance and repair. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0061] Figure 1 A schematic diagram of a method flow chart of an embodiment of the present invention;

[0062] Figure 2 Schematic diagram of the device flow of an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0064] It should be noted that the references to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0065] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0066] like Figure 1 As shown, the path planning method of the automatic road inspection robot based on artificial intelligence includes the following steps:

[0067] S1, road environment data collection: real-time acquisition of road environment data through multiple synchronously working sensors, including lidar, ultrasonic sensors and cameras. Road environment data includes road surface flatness, road cracks, obstacle locations and road damage degree;

[0068] S2, data preprocessing and feature extraction: preprocess the collected road environment data, including noise filtering and data deduplication, and extract features from the preprocessed road environment data. For the image data obtained by the camera, a convolutional neural network is used to extract features to identify road cracks and potholes. For the laser radar and ultrasonic sensor data, a deep neural network is used to fuse data to extract the location and shape information of road damage areas and obstacles.

[0069] S3, build dynamic environment model: Based on the extracted features, use graph convolutional network to build a dynamic environment model to predict the change trend of potential road damage areas and obstacles;

[0070] S4, path planning and optimization: Based on the constructed dynamic environment model, combined with the robot's current position and target position, a deep reinforcement learning algorithm is used for path planning;

[0071] S5, path tracking and adaptive adjustment: After the path planning is completed, the robot drives autonomously according to the planned path. When the environment changes or new obstacles are detected, the robot uses real-time feedback from sensors and adopts online learning methods to adaptively adjust the path;

[0072] S6, Data storage and analysis: The road environment data and path planning information collected during the inspection are fed back to the central control unit in real time for data storage and analysis to form a road health status report.

[0073] S1 includes:

[0074] S11, sensor selection and configuration: In the road inspection task, multiple sensors working synchronously are selected for data collection. The sensors include lidar, ultrasonic sensor and camera, among which:

[0075] LiDAR is used to obtain 3D point cloud data of the road surface, accurately measure the height changes of the road surface, and identify the road surface flatness and obstacles;

[0076] Ultrasonic sensors are used to detect subtle changes in the road surface, such as cracks and pothole depths;

[0077] The camera is used to collect high-resolution image data and provide location information of road cracks, damage and obstacles;

[0078] The selection of sensors is based on comprehensive considerations of factors such as the requirements of the inspection environment, road type, and inspection accuracy;

[0079] S12, sensor placement and data collection: Place lidar, ultrasonic sensors and cameras on the robot to ensure that the sensors cover the inspection area and avoid blind spots;

[0080] The laser radar is placed in front of or at the bottom of the robot to ensure accurate data on road surface smoothness and obstacle locations;

[0081] Ultrasonic sensors are placed on the road surface to detect road surface damage in real time;

[0082] The camera is placed in front of the robot to obtain high-definition road images;

[0083] Multiple sensors work synchronously to ensure that the acquired road environment data is timely and spatially consistent;

[0084] S13, road surface environment data collection: The road surface environment data is collected in real time through multiple synchronously working sensors, including road surface flatness, road surface cracks, obstacle locations, and road surface damage degree, among which:

[0085] The road surface flatness is assessed by obtaining the three-dimensional point cloud data of the road surface through LiDAR scanning and calculating the normal vector, surface curvature and other parameters of the point cloud.

[0086] The image data of road cracks obtained by the camera is used to identify the type, size and distribution of cracks through image processing algorithms (such as edge detection, image segmentation, etc.);

[0087] Obstacle location: LiDAR and ultrasonic sensors provide information on the location, shape, and distance of obstacles to assist in identifying obstacles on the road.

[0088] The degree of road damage is assessed by using ultrasonic sensors to detect the depth of road surface damage and combining it with the image processing results of camera images;

[0089] S14, data synchronization: the collected road environment data is time-stamped uniformly through a synchronization mechanism and transmitted to a data preprocessing module;

[0090] Through the above S11 to S14, it is ensured that the selection and arrangement of sensors are reasonable, the data collection process is comprehensive and accurate, and information such as road surface flatness, cracks, obstacles and road damage degree can be accurately obtained, providing a reliable data basis for subsequent data processing and analysis.

[0091] S2 includes:

[0092] S21, data preprocessing: preliminary preprocessing of the collected road environment data, including noise filtering and data deduplication, wherein;

[0093] Noise filtering: Use filtering algorithms (such as median filtering or Gaussian filtering) to remove random noise from sensor data, especially camera and lidar data, to improve data quality and reliability;

[0094] Data deduplication: Deduplication of collected road environment data, especially when multiple sensors collect similar data, to ensure that the data set does not contain redundant information to avoid affecting subsequent analysis and calculation;

[0095] S22, data standardization and normalization: standardize the data collected by different sensors to ensure that all types of data are analyzed at the same scale. Through the normalization method, the data of different sensors are adjusted to a unified standard range (such as the interval from 0 to 1) to avoid inconsistencies caused by data scale differences in subsequent processing;

[0096] S23, feature extraction: extracting features from the pre-processed road environment data, including:

[0097] S231, image data feature extraction: for the image data collected by the camera, a convolutional neural network is used to extract features. Through the hierarchical structure of the convolutional neural network, important features in the image, such as road cracks, potholes and other damaged areas, are automatically extracted. The specific network structure may include multiple convolutional layers, pooling layers and fully connected layers, and finally a feature vector for identifying cracks and potholes is obtained. The convolution operation is expressed as:

[0098] F(x,y)=∑ i,j I(i,j)·K(xi,yj);

[0099] Among them, I(i,j) is the pixel value of the input image, K(xi,yj) is the convolution kernel, and F(x,y) is the convolution result, which represents the feature response of the image at a specific position;

[0100] S232, LiDAR and ultrasonic sensor data fusion: The data collected by LiDAR and ultrasonic sensors are fused by deep neural network. By fusing these sensor data, more accurate information about the location and shape of road damage areas and obstacles can be extracted. The data fusion model can use a multi-layer perceptron structure, in which each layer learns the features from LiDAR and ultrasonic data and merges them into a unified feature representation. The output of the data fusion model is represented as:

[0101] y=f(x1,x2,…,x n );

[0102] Among them, x1,x2,…,x n are the features obtained from different sensors, f represents the deep neural network model, and y is the fused output feature, which includes the location information of the road damage area and obstacles.

[0103] S2 also includes:

[0104] S24, crack and pothole recognition of image data: crack and pothole recognition is performed on the image features processed by the convolutional neural network, and a segmentation method based on a specific threshold is used to extract the location and shape information of cracks and potholes in the image, which is expressed as:

[0105] S = {(x,y)|I(x,y)>T};

[0106] Where I(x,y) is the pixel value of the image at position (x,y), T is the threshold, and S is the area of ​​cracks and potholes;

[0107] S25, road damage area and obstacle location recognition: Based on the fusion data of LiDAR and ultrasonic sensors, a deep neural network model is used to recognize road damage areas and obstacle locations. By training the neural network model, it can automatically recognize the location and shape of road damage areas (such as cracks, potholes, etc.) and obstacles (such as road bumps, stones, etc.) from the sensor data;

[0108] S26, feature vector generation and annotation: annotate and fuse the features in the image data and sensor data to generate feature vectors. Each feature vector includes the location coordinates, shape description and other related attributes of road cracks, potholes, damaged areas and obstacles. These feature vectors will serve as input for subsequent dynamic environment modeling and path planning.

[0109] Through steps S24 to S26, the features in the image data and sensor data are further deeply analyzed and annotated, providing comprehensive and high-precision road damage and obstacle information for the subsequent construction of the dynamic environment model, effectively improving the accuracy of path planning and the safety of robot driving.

[0110] S3 includes:

[0111] S31, feature map construction: based on the generated feature vector, a road environment feature map is constructed. The feature map is based on the output information of each sensor, including road surface flatness, crack depth, obstacle location, road damage area and other features. Through data fusion and spatial mapping, this information is converted into the connection relationship between nodes and edges on the image structure. Each node represents an environmental feature, and the edges between nodes represent the spatial or association relationship between features.

[0112] S32, graph convolutional network model construction: Use the graph convolutional network to process the constructed road environment feature map and build a preliminary dynamic environment model. The specific steps include:

[0113] Node feature initialization: Initialize the feature vector for each node. The vector contains information such as road damage degree, crack length, obstacle location, etc. Each feature vector is standardized or normalized to ensure the uniform scale of the data;

[0114] Adjacency matrix construction: Based on the data collected by the sensors and the physical spatial relationship, an adjacency matrix is ​​constructed. The adjacency matrix is ​​used to represent the relationship between nodes and is set based on spatial proximity, feature correlation, or physical connectivity.

[0115] Graph convolution operation: Graph convolution operation is used to transfer and aggregate node features, and to mine the spatial correlation and structural information between nodes, thereby achieving modeling of the road environment;

[0116] S33, Optimization and update of dynamic environment model: Based on the initially established dynamic environment model, the dynamic environment model is optimized by dynamically adjusting network parameters and updating node features, including:

[0117] Model update: Update the weights and node features of the graph convolutional network based on real-time collected road environment data, adjust model parameters, and ensure that the model can accurately reflect the changing trend of the road environment;

[0118] Environmental prediction: Use the updated model to predict the environment and the changing trends of potential road damage areas and obstacles, thereby providing support for subsequent path planning and decision-making.

[0119] S4 includes:

[0120] S41, definition of path planning objectives: Before path planning, define the objective function of path planning. The objective function is used to quantify the efficiency of the robot inspection task and includes the following optimization objectives:

[0121] Shortest path: Select the shortest path from the robot's current position to the target position to minimize inspection time;

[0122] Obstacle avoidance: During path planning, ensure that the path avoids obstacles and road damage areas (such as cracks, potholes, etc.) and that the robot does not collide with obstacles;

[0123] Inspection coverage: Maximize the coverage of the inspection area to ensure that the robot can fully check the road health in the designated area;

[0124] The mathematical expression of the objective function is:

[0125] J(π)=w1·f distance (π)+w2·f obstacle (π)+w3·f inspection (π);

[0126] Among them, f distance (π) is the total length of the path, f obstacle (π) represents the path obstacle avoidance performance, f inspection (π) is the inspection coverage area, w1, w2, w3 are weight coefficients, representing the priority of different targets;

[0127] S42, Application of Deep Reinforcement Learning Algorithm: A deep reinforcement learning algorithm using a deep Q network as path planning. The deep Q network learns to select the best action from the current state during the training process to maximize the long-term cumulative reward. The training process includes:

[0128] State space definition: The state space includes the robot's current position, target position, and current environment state (such as obstacle position, cracks, road damage area, etc.);

[0129] Action space definition: The action space represents the actions that the robot can choose, such as forward, left, right, turn, etc.

[0130] Reward function design: During the path planning process, the reward function is used to evaluate the quality of the robot's selected actions. The reward function includes:

[0131] R(s t ,a t )=-w1·distance_penalty(s t ,a t )-w2·collision_penalty(s t ,a t )+w3·inspection_coverage(s t ,a t );

[0132] Among them, distance_penalty is the distance penalty of the path, collision_penalty is the collision penalty, inspection_coverage is the inspection coverage reward, and the weight coefficients w1, w2, and w3 are used to balance the weights of different goals;

[0133] S43, path optimization and adjustment: During the path planning process, the robot's real-time feedback information (such as sensor data) is combined to dynamically adjust the path. The deep Q network updates the Q value according to real-time environmental changes (such as the appearance of obstacles, the expansion of cracks, etc.) and selects the optimal action strategy. The robot adjusts the path in real time during the inspection process to avoid the appearance of new obstacles or road damage areas.

[0134] S4 also includes:

[0135] S44, deep Q network training process: using the deep Q network for path planning training, including experience replay and Q value update, where;

[0136] Experience replay: During the interaction between the robot and the environment, the robot collects the state, action, reward, and next state quadruple and stores these experiences in the experience pool. Through the experience replay technology, the experience data is randomly extracted for training to break the correlation of the data and improve the learning effect of the model.

[0137] Q value update: In each round of training, the deep Q network selects an action based on the current state and updates the Q value based on the environmental feedback. The Q value update formula is as follows:

[0138]

[0139] Among them, R(s t ,a t ) is the current reward, γ is the discount factor that controls the weight of future rewards, α is the learning rate, Q(s t ,a t ) is the Q value of the current state and action, max a′ Q(s t+1 ,a′) is the maximum Q value of the next state;

[0140] S45, dynamic path adjustment: During the robot inspection process, when the environmental state changes (such as obstacle movement, crack expansion, etc.), the deep Q network adjusts the path planning through real-time feedback to ensure that the robot avoids new obstacles and optimizes the inspection efficiency. Specifically, the deep Q network recalculates the Q value based on the real-time road environment data, thereby adjusting the robot's travel path to ensure that the inspection task is completed on time;

[0141] S46, path planning result evaluation and optimization: After the path planning is completed, the robot verifies it through actual inspection tasks and evaluates and optimizes the path based on the inspection results. The optimization strategies include:

[0142] Path smoothing: Smoothing the path to reduce unnecessary turns and emergency stops, ensuring a smoother inspection process;

[0143] Efficiency improvement: Based on the comparison between the path planning results and the actual inspection time, the path planning algorithm is optimized to further improve the inspection efficiency.

[0144] S5 includes:

[0145] S51, path tracking: After the path planning is completed, the inspection robot drives autonomously according to the planned path, and obtains the distance deviation between the robot and the path in real time through the sensors installed on the inspection robot (such as laser radar, ultrasonic sensor, camera, etc.). The path tracking algorithm adjusts the robot's motion control by comparing the difference between the robot's current position and the target path to ensure that the robot always moves along the predetermined path;

[0146] The path tracking control algorithm uses a PID controller to adjust the robot's speed, steering angle, and acceleration in real time to ensure the accuracy and stability of path tracking;

[0147] S52, environmental change detection: The inspection robot monitors the surrounding environment in real time through sensors and detects changes that affect the path (such as new obstacles, road damage, cracks, etc.);

[0148] S53, Adaptive path adjustment: When the environment changes or new obstacles are detected, the inspection robot uses the real-time feedback of road environment data and online learning methods to adaptively adjust the path. The specific steps include:

[0149] Application of online learning algorithms: Use online reinforcement learning (such as Q-learning, DQN, etc.) to optimize the path. The online learning method can update the robot's path decision based on new environmental data, thereby adjusting the path to avoid obstacles or road damage areas;

[0150] Adaptive path generation: The robot recalculates a new path to avoid obstacles in real time based on environmental changes and adaptive adjustment strategies, and adjusts the robot motion according to the path:

[0151] The formula is as follows, assuming the current path is π current , the new avoidance path is π new , the goal is to minimize the path deviation and obstacle avoidance cost:

[0152]

[0153] Among them, D distance (π) is the new path length, D obstacle (π) is the obstacle avoidance cost, D inspection (π) is the inspection efficiency, w2, w3 are weight coefficients;

[0154] S54, path replanning and execution: After the path adjustment is completed, the robot will re-control the movement according to the new path, continue to drive along the optimized path, and ensure to avoid new obstacles or damaged areas. During the path execution process, the robot will continue to make continuous adjustments through real-time feedback to ensure that the inspection task is completed on time.

[0155] S6 includes:

[0156] S62, data storage and management: storing the collected road environment data and path planning information in the database of the central control unit. The data storage includes:

[0157] Data formatting and organization: Format the data and convert the data from different sensors into a standardized format to ensure that data from different sources can be stored compatibly in the database;

[0158] Data backup and security: regularly back up stored data to ensure data security and reliability, and take encryption measures to protect data privacy and prevent data loss;

[0159] S63, data analysis and report generation: After the data is stored, the central control unit analyzes the road environment data and path planning information and generates a road health status report. The specific steps are as follows:

[0160] Data analysis algorithm: Analyze the collected data using statistical analysis methods to identify road damage areas, potential risk areas, and sections that require special attention;

[0161] Health status assessment: Based on the analysis results, the health status of the entire road surface is assessed and a detailed report is generated. The report includes the road damage degree and distribution map, road inspection coverage, damage prediction of key sections, and optimization suggestions for the robot's driving path;

[0162] The generated report is fed back to relevant managers through the central control unit to provide a basis for subsequent maintenance decisions and road maintenance.

[0163] like Figure 2 As shown, the path planning device of the automatic road inspection robot based on artificial intelligence is used to implement the path planning method of the automatic road inspection robot based on artificial intelligence, and includes the following modules:

[0164] Data acquisition module: acquires road surface environment data in real time through multiple synchronously working sensors, including laser radar, ultrasonic sensor and camera. The road surface environment data includes road surface flatness, road surface cracks, obstacle location and road surface damage degree;

[0165] Data preprocessing module: preprocess the collected road environment data, including noise filtering and data deduplication, and extract features from the preprocessed road environment data. For the image data obtained by the camera, a convolutional neural network is used to extract features to identify road cracks and potholes. For the laser radar and ultrasonic sensor data, a deep neural network is used to fuse data to extract the location and shape information of road damage areas and obstacles.

[0166] Dynamic environment model building module: Based on the extracted features, a dynamic environment model is built using a graph convolutional network to predict the changing trends of potential road damage areas and obstacles;

[0167] Path planning module: Based on the constructed dynamic environment model, combined with the robot's current position and target position, a deep reinforcement learning algorithm is used for path planning;

[0168] Path tracking and adaptive adjustment module: After the path planning is completed, the robot drives autonomously according to the planned path. When the environment changes or new obstacles are detected, the robot uses real-time feedback from sensors and adopts online learning methods to adaptively adjust the path.

[0169] Data storage and analysis module: The road environment data and path planning information collected during the inspection process are fed back to the central control unit in real time for data storage and analysis to form a road health status report.

[0170] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0171] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A path planning method for an automatic road inspection robot based on artificial intelligence, characterized in that: The following steps are involved: S1, road environment data collection: real-time acquisition of road environment data through multiple synchronously working sensors, the sensors include laser radar, ultrasonic sensor and camera, the road environment data includes road surface flatness, road surface cracks, obstacle location and road surface damage degree; S2, data preprocessing and feature extraction: preprocess the collected road environment data, including noise filtering and data deduplication, and extract features from the preprocessed road environment data. For the image data obtained by the camera, a convolutional neural network is used to extract features to identify road cracks and potholes. For the laser radar and ultrasonic sensor data, a deep neural network is used to fuse data to extract the location and shape information of road damage areas and obstacles. S3, build dynamic environment model: Based on the extracted features, use graph convolutional network to build a dynamic environment model to predict the change trend of potential road damage areas and obstacles; S4, path planning and optimization: Based on the constructed dynamic environment model, combined with the robot's current position and target position, a deep reinforcement learning algorithm is used for path planning; S5, path tracking and adaptive adjustment: After the path planning is completed, the robot drives autonomously according to the planned path. When the environment changes or new obstacles are detected, the robot uses real-time feedback from sensors and adopts online learning methods to adaptively adjust the path; S6, Data storage and analysis: The road environment data and path planning information collected during the inspection are fed back to the central control unit in real time for data storage and analysis to form a road health status report.

2. The path planning method of the automatic road inspection robot based on artificial intelligence according to claim 1 is characterized in that: The S1 includes: S11, sensor selection and configuration: in the road inspection task, multiple sensors working synchronously are selected for data collection, and the sensors include laser radar, ultrasonic sensor and camera; S12, sensor placement and data collection: Place lidar, ultrasonic sensors and cameras on the robot to ensure that the sensors cover the inspection area; S13, road surface environment data collection: collecting road surface environment data in real time through multiple synchronously working sensors, the data including road surface flatness, road surface cracks, obstacle locations and road surface damage degree; S14, data synchronization: The collected road environment data is time-stamped uniformly through a synchronization mechanism and transmitted to the data preprocessing module.

3. The path planning method of the automatic road inspection robot based on artificial intelligence according to claim 2 is characterized in that: The S2 includes: S21, data preprocessing: preliminary preprocessing of the collected road environment data, including noise filtering and data deduplication; S22, data standardization and normalization: standardize the data collected by different sensors; S23, feature extraction: extracting features from the pre-processed road environment data, including: S231, image data feature extraction: for the image data collected by the camera, a convolutional neural network is used to extract features. The convolution operation is expressed as: F(x,y)=∑ i,j I(i,j)·K(x-i,y-j); Among them, I(i,j) is the pixel value of the input image, K(xi,yj) is the convolution kernel, and F(x,y) is the convolution result, which represents the feature response of the image at a specific position; S232, LiDAR and ultrasonic sensor data fusion: Perform deep neural network fusion on the data collected by LiDAR and ultrasonic sensors to extract the location and shape information of road damage areas and obstacles.

4. The path planning method for an automatic road inspection robot based on artificial intelligence according to claim 3 is characterized in that: The S2 further includes: S24, crack and pothole recognition of image data: crack and pothole recognition is performed on the image features processed by the convolutional neural network, and a segmentation method based on a specific threshold is used to extract the location and shape information of cracks and potholes in the image; S25, road damage area and obstacle location identification: Based on the fusion data of LiDAR and ultrasonic sensors, a deep neural network model is used to identify road damage areas and obstacle locations; S26, feature vector generation and annotation: annotate and fuse the features in the image data and sensor data to generate a feature vector.

5. The path planning method for an automatic road inspection robot based on artificial intelligence according to claim 4 is characterized in that: The S3 includes: S31, feature map construction: constructing a road environment feature map based on the generated feature vector; S32, graph convolutional network model construction: Use the graph convolutional network to process the constructed road environment feature map and build a preliminary dynamic environment model; S33, Optimization and updating of dynamic environment model: Based on the initially established dynamic environment model, the dynamic environment model is optimized by dynamically adjusting network parameters and updating node characteristics.

6. The path planning method for an automatic road inspection robot based on artificial intelligence according to claim 5 is characterized in that: The S4 includes: S41, definition of path planning objectives: before path planning, define the objective function of path planning, which is used to quantify the efficiency of the robot inspection task; S42, Application of deep reinforcement learning algorithm: Deep reinforcement learning algorithm using deep Q network as path planning; S43, path optimization and adjustment: During the path planning process, the path is dynamically adjusted in combination with the robot’s real-time feedback information.

7. The path planning method of the automatic road inspection robot based on artificial intelligence according to claim 6 is characterized in that: The S4 further comprises: S44, deep Q network training process: using deep Q network for path planning training, including experience replay and Q value update; S45, dynamic path adjustment: During the robot inspection process, when the environmental state changes, the deep Q network adjusts the path planning through real-time feedback; S46, evaluation and optimization of path planning results: After the path planning is completed, the robot verifies it through actual inspection tasks and evaluates and optimizes the path based on the inspection results.

8. The path planning method for an automatic road inspection robot based on artificial intelligence according to claim 7 is characterized in that: The S5 includes: S51, path tracking: after the path planning is completed, the inspection robot drives autonomously according to the planned path, and obtains the distance deviation between the robot and the path in real time through the sensors installed on the inspection robot; S52, environmental change detection: The inspection robot monitors the surrounding environment in real time through sensors and detects changes that affect the path; S53, Adaptive path adjustment: When the environment changes or new obstacles are detected, the inspection robot uses the real-time feedback of road environment data and online learning methods to adaptively adjust the path; S54, path replanning and execution: After the path adjustment is completed, the robot re-controls the motion according to the new path and continues to move along the optimized path.

9. The path planning method for an automatic road inspection robot based on artificial intelligence according to claim 8 is characterized in that: The S6 includes: S62, data storage and management: storing the collected road environment data and path planning information in a database of the central control unit; S63, data analysis and report generation: After the data is stored, the central control unit analyzes the road environment data and path planning information and generates a road health status report.

10. An artificial intelligence-based automatic road inspection robot path planning device, used to implement the artificial intelligence-based automatic road inspection robot path planning method according to any one of claims 1 to 9, characterized in that: Includes the following modules: Data acquisition module: acquires road surface environment data in real time through multiple synchronously working sensors, including laser radar, ultrasonic sensor and camera. The road surface environment data includes road surface flatness, road surface cracks, obstacle location and road surface damage degree; Data preprocessing module: preprocess the collected road environment data, including noise filtering and data deduplication, and extract features from the preprocessed road environment data. For the image data obtained by the camera, a convolutional neural network is used to extract features to identify road cracks and potholes. For the laser radar and ultrasonic sensor data, a deep neural network is used to fuse data to extract the location and shape information of road damage areas and obstacles. Dynamic environment model building module: Based on the extracted features, a dynamic environment model is built using a graph convolutional network to predict the changing trends of potential road damage areas and obstacles; Path planning module: Based on the constructed dynamic environment model, combined with the robot's current position and target position, a deep reinforcement learning algorithm is used for path planning; Path tracking and adaptive adjustment module: After the path planning is completed, the robot drives autonomously according to the planned path. When the environment changes or new obstacles are detected, the robot uses real-time feedback from sensors and adopts online learning methods to adaptively adjust the path. Data storage and analysis module: The road environment data and path planning information collected during the inspection process are fed back to the central control unit in real time for data storage and analysis to form a road health status report.

Images