Highway inspection robot path planning method, device, equipment, medium and product

Abstract

The application discloses a highway inspection robot path planning method, device, equipment, medium and product, relates to the field of robot path planning, and comprises the following steps: determining an inspection section of a highway inspection robot, and constructing an environment grid map corresponding to the inspection section; adopting an improved path planning method according to the environment grid map and the inspection section to obtain a path planning result of the highway inspection robot. The application can flexibly fuse a local action prior distribution and a path planning method, only relies on a local environment state, and realizes efficient and reliable highway inspection robot path planning.

Description

Technical Field

[0001] This application relates to the field of robot path planning, and in particular to a method, apparatus, equipment, medium and product for path planning of a highway inspection robot. Background Technology

[0002] Intelligent safety inspection robots for highway engineering are mainly used for daily safety inspections and emergency checks of highways, urban expressways, and their roadside facilities (such as road surfaces, guardrails, protective nets, bridge structures, tunnels, and slopes). Compared to indoor or regular venues, these scenarios are characterized by long road mileage, diverse structural forms, complex traffic participants, and frequent temporary construction and sudden obstacles. When performing inspection tasks, the robot needs to plan a safe and efficient travel path in long-distance, dynamically changing road networks to achieve precise arrival and return by road segment, structure, and work point.

[0003] Existing highway inspection robots mostly employ traditional path planning algorithms based on grid maps or road network graphs, such as deterministic graph search algorithms like Dijkstra's algorithm and A* algorithm, as well as stochastic or swarm intelligence algorithms like Ant Colony Optimization (ACO), genetic algorithms, and particle swarm optimization. These methods suffer from several drawbacks in the large-scale, long-term, and partially observable real-world scenarios of highway engineering, including strong reliance on global maps, high computational overhead on large-scale roads, high sensitivity to human inspiration and parameter configuration, difficulty in systematically utilizing historical prior experience, and a lack of a unified prior guidance mechanism across algorithms. Summary of the Invention

[0004] The purpose of this application is to provide a path planning method, device, equipment, medium and product for highway inspection robots, which flexibly integrates local action prior distribution with path planning methods, and achieves efficient and reliable path planning for highway inspection robots by relying solely on local environmental conditions.

[0005] To achieve the above objectives, this application provides the following solution.

[0006] Firstly, this application provides a path planning method for a highway inspection robot, comprising: determining the inspection segment of the highway inspection robot and constructing an environmental grid map corresponding to the inspection segment; the inspection segment includes an inspection start point and an inspection end point; based on the environmental grid map and the inspection segment, an improved path planning method is used to obtain the path planning result of the highway inspection robot; the improved path planning method introduces a local action prior distribution in the node transfer mechanism of the path planning method to determine the transfer node of the current node; the local action prior distribution is a probability distribution function obtained by using a strategy model based on the environmental state corresponding to the current node in the environmental grid map; the environmental state includes obstacle distribution and passable area; the strategy model is a probability model obtained in advance by using an imitation learning method based on multiple sample environmental states and sample local actions corresponding to each sample environmental state; each element in the probability distribution function corresponds to the transfer probability of a transfer node.

[0007] Secondly, this application provides a path planning device for a highway inspection robot, comprising: a task construction module, used to determine the inspection segment of the highway inspection robot and construct an environmental grid map corresponding to the inspection segment; the inspection segment includes an inspection start point and an inspection end point; a path planning module, used to obtain the path planning result of the highway inspection robot by adopting an improved path planning method based on the environmental grid map and the inspection segment; the improved path planning method introduces a local action prior distribution in the node expansion mechanism of the path planning method to determine the transition node of the current node; the local action prior distribution is a probability distribution function obtained by adopting a strategy model based on the environmental state corresponding to the current node in the environmental grid map; the environmental state includes obstacle distribution and passable area; the strategy model is a probability model obtained in advance by adopting an imitation learning method based on multiple sample environmental states and sample local actions corresponding to each sample environmental state; each element in the probability distribution function corresponds to the transition probability of a transition node.

[0008] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described path planning method for a highway inspection robot.

[0009] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described path planning method for a highway inspection robot.

[0010] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned path planning method for a highway inspection robot.

[0011] Based on the specific embodiments provided in this application, the following technical effects are disclosed.

[0012] This application provides a path planning method, device, equipment, medium, and product for a highway inspection robot. By determining the inspection section of the highway inspection robot and constructing an environmental grid map corresponding to the inspection section, the continuous road environment is discretized into computable grid cells. Simultaneously, environmental states such as obstacle distribution and passable areas are integrated, enabling the highway inspection robot to accurately plan its inspection path. By pre-building a policy model using imitation learning based on multiple sample environmental states and the corresponding local actions of each sample environmental state, and directly training the probability model using expert sample data, training time and sample collection costs are significantly reduced. Furthermore, by obtaining the prior distribution of local actions based on the environmental state corresponding to the current node in the environmental grid map using the policy model, and utilizing… Probabilistic models output local action distributions rather than deterministic decisions, exhibiting better fault tolerance when dealing with sensor noise and environmental uncertainties. Furthermore, the traceability of expert decisions preserved through imitation learning allows highway inspection robots to behave more in accordance with empirical operating procedures. By introducing local action prior distributions into the node transfer mechanism of the path planning method to determine the transfer node of the current node, an improved path planning method is obtained. This transforms blind expansion into biased sampling, significantly reducing the number of invalid nodes explored and shortening the time to find feasible or optimal paths. Simultaneously, the prior distribution integrates environmental conditions and expert experience, enabling path planning to proactively avoid high-risk areas or prioritize highly passable road sections, maintaining path planning stability in highway inspection scenarios. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is an application environment diagram of a path planning method for a highway inspection robot according to an embodiment of this application.

[0015] Figure 2 This is a flowchart illustrating a path planning method for a highway inspection robot, provided as an embodiment of this application.

[0016] Figure 3This is a schematic diagram of the overall framework of a path planning method for a highway inspection robot provided in one embodiment of this application.

[0017] Figure 4 This is a schematic diagram of the functional modules of a road inspection robot path planning device provided in an embodiment of this application.

[0018] Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] The path planning method for highway inspection robots provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 101 communicates with server 102 via a network. A data storage system can store the data that server 102 needs to process. The data storage system can be set up independently, integrated into server 102, or placed in the cloud or on another server. Terminal 101 can send the inspection sections of the highway inspection robot and the corresponding environmental grid map to server 102. Server 102 receives the inspection sections and the corresponding environmental grid map, and uses an improved path planning method to obtain the path planning result for the highway inspection robot. Server 102 can then feed back the obtained path planning result to terminal 101. Furthermore, in some embodiments, the path planning method for the highway inspection robot can also be implemented separately by the server 102 or the terminal 101. For example, the terminal 101 can directly use an improved path planning method to obtain the path planning result of the highway inspection robot for the inspection section of the highway inspection robot and the environmental grid map corresponding to the inspection section. Alternatively, the server 102 can obtain the inspection section of the highway inspection robot and the environmental grid map corresponding to the inspection section from the data storage system, and use an improved path planning method to obtain the path planning result of the highway inspection robot.

[0022] The terminal 101 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 102 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0023] In one exemplary embodiment, such as Figure 2 As shown, a path planning method for a highway inspection robot is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 102 as an example, the explanation includes the following steps 201 and 202.

[0024] Step 201: Determine the inspection section of the highway inspection robot and construct the corresponding environmental grid map of the inspection section. The inspection section includes an inspection start point and an inspection end point.

[0025] Step 202: Based on the environmental grid map and the inspection section, an improved path planning method is used to obtain the path planning result of the highway inspection robot.

[0026] The improved path planning method introduces a local action prior distribution into the node transfer mechanism of the path planning method to determine the transfer node of the current node. The local action prior distribution is a probability distribution function obtained by using a policy model based on the environmental state corresponding to the current node in the environmental grid map. The environmental state includes obstacle distribution and passable areas. The policy model is a probability model obtained in advance using an imitation learning method based on multiple sample environmental states and the sample local actions corresponding to each sample environmental state. Each element in the probability distribution function corresponds to the transfer probability of a transfer node.

[0027] In one example, the method for building the strategy model includes the following steps 301 to 306.

[0028] Step 301: Construct multiple sample environment raster maps and set multiple sample inspection road segments based on each sample environment raster map.

[0029] Step 302: For any sample inspection segment, based on the sample inspection segment and the sample environment grid map corresponding to the sample inspection segment, multiple path planning methods are used to obtain multiple demonstration paths corresponding to the sample inspection segment.

[0030] Step 303: For any demonstration path, perform time discretization processing on the demonstration path to obtain multiple discrete path points.

[0031] Step 304: For any discrete path point, extract the sample environment state corresponding to the discrete path point based on the sample environment raster map corresponding to the discrete path point.

[0032] Step 305: For any sample environment state, based on the demonstration path corresponding to the sample environment state, determine the action from the discrete path point corresponding to the sample environment state to the next discrete path point, so as to obtain the sample local action corresponding to the sample environment state.

[0033] Step 306: Based on the environmental states of all samples and the local actions of each sample corresponding to the environmental state, the policy model is obtained by imitation learning.

[0034] In one example, after step 302, the method for building the strategy model further includes: modifying each demonstration path based on preset correction rules.

[0035] In another example, such as Figure 3 As shown, in the offline prior knowledge construction stage, the strategy model construction method includes the following steps S1 to S3.

[0036] Step S1: Highway scene setup and task configuration.

[0037] First, the model training environment is set up. In typical highway engineering scenarios, various representative road structures are selected as sample inspection sections, and highway inspection tasks are configured for these sections, such as mainline lane inspection, ramp connection section inspection, service area / parking lane entry and exit, and inspection of road sections near bridges and slopes. The road environment of the sample inspection sections is discretized into a sample environment raster map or a sample environment topology map. The inspection start point, inspection end point, and passable area are defined for each sample inspection section.

[0038] Step S2: Demonstrate path generation and data collection for manual teaching.

[0039] Based on the highway engineering scenario and inspection task obtained in step S1, a demonstration path is generated using traditional and classic deterministic path planning algorithms (such as the shortest path first algorithm and the A* algorithm) and uncertain path planning algorithms (such as the fast exploratory random tree star algorithm and the ant colony optimization algorithm).

[0040] In the model training environment built in step S1, a high-performance global planner (such as the A* algorithm based on a global map, ant colony optimization algorithm, etc.) is invoked for path planning. Expert inspection trajectories are obtained through remote manual control to generate a large number of safe and reasonable demonstration paths. These demonstration paths cover different road segment types, obstacle distributions, and traffic control conditions, and can be regarded as an "expert experience base" in the robot's highway inspection task.

[0041] Step S3, Environmental State - Construction of Local Action Samples and Prior Modeling.

[0042] The demonstration path obtained in step S2 is subjected to time discretization. Based on the Markov decision idea, the continuous demonstration path is decomposed into a set of sample pairs of "sample environment state - sample local action" in the time dimension. Based on this set of sample pairs, a deep learning method is used to train the policy model to obtain the policy model.

[0043] In this example, the demonstration path is discretized in time. For each path point, a corresponding sample environment state is constructed based on the local environmental state perception that the highway inspection robot can actually obtain (such as the distribution of local obstacles, passable areas, approximate direction of the target or next path point, simplified historical movement trajectory, etc.). For example, environmental state information is modeled based on the local perception information of the highway inspection robot, and the sample environment state of the highway inspection robot is represented by a high-dimensional matrix. The size of the matrix corresponds to the size of the local observation area. A binary occupancy representation is used to construct matrices representing information such as the distribution of local obstacles, passable areas, approximate direction of the target or next inspection point, and simplified historical movement trajectory, thus forming the sample environment state matrix. Furthermore, the movement direction or discrete action taken in the demonstration path under this sample environment state is obtained, resulting in a set of "sample environment state - sample local action" sample pairs.

[0044] like Figure 3 As shown, a local action prior module is constructed using statistical or learning methods (such as imitation learning, lightweight neural networks, lookup table strategies, etc.). In another example, the trained local action prior module is deployed in hardware and software form on the host computer or onboard computing platform of the highway inspection robot, and a unified application programming interface (API) structure is defined. When the local environment state is input, the API structure outputs the local action prior distribution, providing unified prior guidance information for subsequent different types of path planning algorithms. In the prior guidance planning stage, the environmental state in the environmental grid map is obtained by sensing environmental information, and the local action prior distribution is further fused to output the optimal inspection path for the highway inspection robot.

[0045] In one example, the imitation learning method is a stochastic policy behavior cloning method, a maximum entropy inverse reinforcement learning method, or a generative adversarial imitation learning method. The policy model is ultimately obtained using the imitation learning method. ,in, For the strategy model, For environmental conditions, for The corresponding probability distribution function. This represents the environmental state of the current node in the environmental grid map. By substituting the strategy model, we can obtain... Corresponding probability distribution function (a). Through calculation The local action with the highest probability can be used as the action of the inspection robot in the next moment. To make the function Local actions when taking the maximum value As an action variable, for The action space within.

[0046] In one example, the improved path planning method is an improved A* algorithm.

[0047] For any given node, the improved A* algorithm uses the evaluation function shown in the following formula to calculate the evaluation value of each neighboring node of the current node, and updates the OPEN list based on all neighboring nodes and the evaluation value of each neighboring node.

[0048] .

[0049] in, n The adjacent nodes of the current node. for n The corresponding actual cost function, for n The corresponding local action prior probability, for n Based on the corresponding evaluation value, the improved A* algorithm uses the node with the smallest evaluation value in the OPEN list as the transition node for the current node.

[0050] In this example, based on the actual highway inspection task, according to the inspection section issued by the current inspection task, combined with the current position of the highway inspection robot and existing map / perception information, an environmental grid map or graph structure of the workspace is constructed, and data structures such as the OPEN list and CLOSED list of deterministic programming algorithms such as A* algorithm are initialized.

[0051] During the A* algorithm search process, each time a node to be expanded is taken from the OPEN list, a local environmental state (such as the distribution of road boundaries, guardrails, obstacles and the relative direction of the inspection endpoint within a certain distance) is constructed within the local range that the highway inspection robot can obtain, with the current node as the center. This environmental state is then input into the local action prior module to obtain the local action prior distribution of each feasible path around the current node.

[0052] By treating the local action prior distribution as a reflection of "direction priority" or "implicit residual cost," a local action prior distribution is introduced for each neighboring node of the current node to form an evaluation function. .

[0053] In this example, actions with higher prior probabilities are assigned lower additional costs, making the search more inclined towards directions deemed better by expert experience. Furthermore, in the presence of traditional geometric heuristics... At that time, the weighting coefficient can be used to... and Perform linear or nonlinear combinations.

[0054] After calculating the evaluation value of each neighboring node of the current node, the OPEN list is sorted and expanded based on all neighboring nodes and the evaluation value of each neighboring node until a feasible path to the inspection endpoint is found or the termination condition is met. Since local action priors are referenced in each expansion step, the path planning process will automatically avoid a large number of low-value branches, reduce the search space, and improve the efficiency of path planning.

[0055] In another example, the path planning method is an improved ant colony optimization algorithm.

[0056] Targeting ants k For any current node, the improved ant colony optimization algorithm uses the state transition probability function shown in the following formula to calculate the transition probability corresponding to each neighboring node of the current node, and uses a probability selection method to determine the transition node of the current node based on the transition probabilities corresponding to all neighboring nodes.

[0057] .

[0058] in, i For the current node, j for i The adjacent nodes, For ants k corresponding i The set of adjacent nodes, From i Transferred to j The prior probability of local actions, For path ( i ,j The concentration of pheromones on the surface. For path ( i , j Heuristic information on ) As a factor of pheromone importance, As a factor of the importance of heuristic information, l for Adjacent nodes in the array From i Transferred to l The prior probability of local actions, For path ( i , l The concentration of pheromones on the surface. For path ( i , l Heuristic information on ) For ants k from i arrive j The transition probability.

[0059] When using the ACO algorithm for path planning, a set of nodes and edges is constructed based on the road environment, the pheromone matrix and heuristic factor are initialized (such as based on distance or road level), and parameters such as the number of iterations and the number of ants are set according to task requirements.

[0060] In each state transition during the ant colony search, when the ants k Located at the current node i When constructing nodes i The corresponding local environment state is input into the local action prior module to obtain the local action prior probability of transitioning from the current node to each neighboring node. Then, the pheromone concentration of the traditional ACO, heuristic information, and local action prior probabilities are combined to construct a new state transition probability function, making actions with higher prior probabilities more likely to be selected during sampling, thereby focusing the search on regions that are considered more promising in the prior.

[0061] In another example, the initial pheromone matrix is ​​non-uniformly initialized using the local action prior distribution, or the pheromone update strategy is dynamically adjusted during the iteration process, making it easier for actions with high prior probabilities to accumulate and retain pheromones, thereby further accelerating the ant colony convergence speed.

[0062] After completing a predetermined number of iterations, the path with the lowest cost or the best overall performance is selected from the path set generated by the ant colony as the execution path for the inspection robot's current task. Since the entire search process is guided by prior knowledge, it can balance path length, safety redundancy, and planning time while ensuring completion rate.

[0063] Based on the same inventive concept, this application also provides a road inspection robot path planning device for implementing the road inspection robot path planning method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more road inspection robot path planning device embodiments provided below can be found in the limitations of the road inspection robot path planning method described above, and will not be repeated here.

[0064] In one exemplary embodiment, such as Figure 4 As shown, a path planning device for a highway inspection robot is provided, including a task construction module 001 and a path planning module 002.

[0065] The task construction module 001 is used to determine the inspection sections of the highway inspection robot and construct the corresponding environmental grid map of the inspection sections. The inspection sections include an inspection start point and an inspection end point. The path planning module 002 is used to obtain the path planning result of the highway inspection robot based on the environmental grid map and the inspection sections, using an improved path planning method.

[0066] The improved path planning method introduces a local action prior distribution into the node expansion mechanism of the path planning method to determine the transition node of the current node. The local action prior distribution is a probability distribution function obtained by using a policy model based on the environmental state corresponding to the current node in the environmental grid map. The environmental state includes obstacle distribution and passable areas. The policy model is a probability model obtained in advance using an imitation learning method based on multiple sample environmental states and the sample local actions corresponding to each sample environmental state; each element in the probability distribution function corresponds to the transition probability of a transition node.

[0067] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 5As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores the inspection sections of the highway inspection robot and the corresponding environmental grid maps. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a path planning method for a highway inspection robot.

[0068] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0069] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0070] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0071] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0072] Compared with traditional path planning methods, this application has the following significant advantages in the intelligent safety inspection scenario of highway engineering.

[0073] (1) Make full use of historical tasks and expert experience to reduce reliance on artificial inspiration and parameter tuning.

[0074] This application explicitly transforms manual inspection experience and high-quality planner output into a queryable action prior distribution through the "demonstration path - local environment state - local action" approach. This avoids repeatedly designing the A* algorithm heuristic function manually or repeatedly adjusting the hyperparameters of the ant colony optimization algorithm for different road environments, thus significantly reducing engineering deployment and maintenance costs.

[0075] (2) It significantly improves planning efficiency and planning path quality in large-scale road environments.

[0076] By introducing local action prior distribution into the node evaluation function of the A* algorithm or the state transition probability function of the ant colony optimization algorithm, this application effectively reduces the search space, reduces invalid branch expansion and blind sampling, and can achieve planning path quality close to or even better than some traditional algorithms under various map scales and complex obstacle distributions. At the same time, it significantly reduces the average planning time and adapts to the real-time planning needs of long-distance highway inspection.

[0077] (3) Adapt to local observations and incomplete maps, and reduce the reliance on accurate global maps.

[0078] This application outputs action suggestions based on local environmental state. The underlying planner promotes global path generation through sliding window or local replanning. Therefore, even if the global map is incomplete or has local update lag, robust navigation can be achieved by relying on real-time perception and prior guidance, which is more in line with the actual situation of "maps are difficult to maintain completely" in highway engineering sites.

[0079] (4) It provides unified services for deterministic and stochastic planning algorithms and has good algorithm and platform versatility.

[0080] The local action prior module of this application is decoupled from the specific planning algorithm. It provides "candidate action probability distribution" to different algorithms through a unified interface. The planner only needs to introduce prior factors into the original cost function or transition probability function to complete the integration. It is applicable to A* algorithm and its improved algorithms, as well as ant colony optimization algorithm and other swarm intelligence algorithms. At the same time, it can be reused among different types of highway inspection robot platforms such as wheeled and tracked robots.

[0081] (5) Improve the safety and robustness of highway inspection tasks.

[0082] By encoding the expert behavior pattern of "safe detour, appropriate conservatism, and avoidance of high-risk road sections" into action priors, this application can reduce the probability of searching for and adopting dangerous routes in complex working conditions such as construction zones, temporary obstacles, and traffic congestion, reduce the risk of robots getting stuck in local dead ends or clashing with dynamic and static obstacles, and improve the safety and robustness of the overall inspection task.

[0083] (6) The computational overhead is controllable and it is suitable for real-time deployment on embedded platforms.

[0084] In a preferred embodiment, the local action prior module adopts a lightweight structure with a small number of parameters and low inference overhead. The amount of floating-point operations required for a single forward computation is much lower than that of common visual backbone networks. Combined with the prior fusion method of this application, which only requires a small amount of additional computation for the A* algorithm or ant colony optimization algorithm, near real-time path planning can be achieved on ordinary industrial computers or vehicle-mounted embedded processors, providing computational support for the practical application of intelligent safety inspection robots in highway engineering.

[0085] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0086] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0087] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, etc., and are not limited to these.

[0088] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0089] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A path planning method for a highway inspection robot, characterized in that, The path planning method for the highway inspection robot includes: The inspection route of the highway inspection robot is determined, and an environmental grid map corresponding to the inspection route is constructed; the inspection route includes an inspection start point and an inspection end point; Based on the environmental grid map and the inspection section, an improved path planning method is used to obtain the path planning result of the highway inspection robot. The improved path planning method introduces a local action prior distribution into the node transfer mechanism of the path planning method to determine the transfer node of the current node. The local action prior distribution is a probability distribution function obtained by using a policy model based on the environmental state corresponding to the current node in the environmental grid map. The environmental state includes obstacle distribution and passable area. The policy model is a probability model obtained in advance by using an imitation learning method based on multiple sample environmental states and sample local actions corresponding to each sample environmental state. Each element in the probability distribution function corresponds to the transfer probability of a transfer node. The sample environmental state is constructed based on the local environmental state perception of the highway inspection robot. The local environmental state perception of the highway inspection robot includes the local obstacle distribution, passable area, and the direction of the target or the next path point.

2. The path planning method for highway inspection robots according to claim 1, characterized in that, The method for constructing the strategy model includes: Construct multiple sample environment raster maps, and set multiple sample inspection road segments based on each sample environment raster map; For any sample inspection road segment, based on the sample inspection road segment and the sample environment grid map corresponding to the sample inspection road segment, multiple path planning methods are used to obtain multiple demonstration paths corresponding to the sample inspection road segment. For any given demonstration path, the demonstration path is discretized in time to obtain multiple discrete path points; For any discrete path point, the sample environment state corresponding to the discrete path point is extracted based on the sample environment raster map corresponding to the discrete path point. For any sample environment state, based on the demonstration path corresponding to the sample environment state, determine the action from the discrete path point corresponding to the sample environment state to the next discrete path point, so as to obtain the sample local action corresponding to the sample environment state. The policy model is obtained by imitation learning based on the environmental states of all samples and the local actions of each sample corresponding to the environmental state.

3. The path planning method for highway inspection robots according to claim 2, characterized in that, For any sample inspection road segment, after obtaining multiple demonstration paths corresponding to the sample inspection road segment by employing multiple path planning methods based on the sample inspection road segment and the corresponding sample environment grid map, the method for building the strategy model further includes: Each demonstration path is corrected based on preset correction rules.

4. The path planning method for highway inspection robots according to claim 1, characterized in that, The imitation learning method is a stochastic policy behavior cloning method, a maximum entropy inverse reinforcement learning method, or a generative adversarial imitation learning method.

5. The path planning method for a highway inspection robot according to claim 1, characterized in that, The improved path planning method is an improved A* algorithm; For any given node, the improved A* algorithm calculates the evaluation value of each neighboring node using the evaluation function shown below, and updates the OPEN list based on all neighboring nodes and the evaluation value of each neighboring node: ； in, n The adjacent nodes of the current node. for n The corresponding actual cost function, for n The corresponding local action prior probability, for n Based on the corresponding evaluation value, the improved A* algorithm uses the node with the smallest evaluation value in the OPEN list as the transition node for the current node.

6. The path planning method for a highway inspection robot according to claim 1, characterized in that, The path planning method is an improved ant colony optimization algorithm; Targeting ants k For any current node, the improved ant colony optimization algorithm calculates the transition probability corresponding to each neighboring node of the current node using the state transition probability function shown in the following formula, and determines the transition node of the current node using a probability selection method based on the transition probabilities corresponding to all neighboring nodes: ； in, i For the current node, j for i The adjacent nodes, For ants k corresponding i The set of adjacent nodes, From i Transferred to j The prior probability of local actions, For path ( i , j The concentration of pheromones on the surface. For path ( i , j Heuristic information on ) As a factor of pheromone importance, As a factor of the importance of heuristic information, l for Adjacent nodes in the array From i Transferred to l The prior probability of local actions, For path ( i , l The concentration of pheromones on the surface. For path ( i , l Heuristic information on ) For ants k from i arrive j The transition probability.

7. A path planning device for a highway inspection robot, used to implement the path planning method for a highway inspection robot as described in any one of claims 1-6, characterized in that, The highway inspection robot path planning device includes: The task construction module is used to determine the inspection section of the highway inspection robot and construct the environmental grid map corresponding to the inspection section; the inspection section includes an inspection start point and an inspection end point; The path planning module is used to obtain the path planning result of the highway inspection robot by adopting an improved path planning method based on the environmental grid map and the inspection road section. The improved path planning method introduces a local action prior distribution into the node expansion mechanism of the path planning method to determine the transition node of the current node. The local action prior distribution is a probability distribution function obtained by using a policy model based on the environmental state corresponding to the current node in the environmental grid map. The environmental state includes obstacle distribution and passable area. The policy model is a probability model obtained in advance by using an imitation learning method based on multiple sample environmental states and sample local actions corresponding to each sample environmental state. Each element in the probability distribution function corresponds to the transition probability of a transition node.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the road inspection robot path planning method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the path planning method for a highway inspection robot as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the path planning method for the highway inspection robot as described in any one of claims 1-6.

Images