Path planning method and system, medium and robot

Abstract

The invention relates to a path planning method and system, a medium and a robot. The method comprises the following steps: acquiring a navigation instruction and a topological hierarchy semantic map corresponding to the navigation instruction; wherein the topological hierarchy semantic map takes each mark point as an area, and each communicated area is connected by a channel; inputting the navigation instruction and the topological hierarchy semantic map into a large language model, and identifying a starting region and a target region corresponding to a starting point and an end point in the navigation instruction in the topological hierarchy semantic map; based on an A * algorithm, planning a target path according to each channel passing from the starting area to the target area; detecting whether a non-passable area exists in the target path, and correspondingly updating the target path according to a detection result to obtain a final path; wherein the non-passable region is pre-generated through a large language model according to external information, and the external information is used for representing the corresponding channel state. The navigation path can be flexibly and dynamically adjusted in combination with external information.

Description

Technical Field

[0001] The present invention relates to the field of navigation technology, and in particular to a path planning method, system, medium and robot. Background Art

[0002] In the mobile Robotics Operating System (ROS), the default navigation framework uses a two-dimensional occupancy grid map for navigation. Although the robot navigation framework based on ROS has been widely used, it still has some limitations in path planning. On the one hand, it can only calculate the shortest path, and cannot intelligently determine the path based on real-time conditions (such as elevator maintenance notifications). On the other hand, there are a large number of dynamic objects in the occupancy grid map, so the map needs to be updated regularly, and reliable navigation throughout the life cycle cannot be achieved. Therefore, it is necessary to provide a path planning method, system, medium and robot. Summary of the invention

[0003] In view of the above-mentioned shortcomings of the prior art, an object of the present invention is to provide a path planning method, system, device and medium, which improve the problem that the path planning of the prior art cannot effectively cope with the dynamic changes of the environment.

[0004] To achieve the above-mentioned purpose and other related purposes, the present invention provides a path planning method, comprising: obtaining a navigation instruction and a topological level semantic map corresponding thereto; wherein the topological level semantic map uses each marking point as an area and uses channels to connect each interconnected area; inputting the navigation instruction and the topological level semantic map into a large language model, identifying the starting area and target area corresponding to the starting point and the end point in the navigation instruction in the topological level semantic map; planning a target path based on the A* algorithm according to each channel passing from the starting area to the target area; detecting whether there is an impassable area in the target path, and updating the target path accordingly according to the detection result to obtain a final path; wherein the impassable area is pre-generated by the large language model according to external information, and the external information is used to characterize the corresponding channel state.

[0005] In one embodiment of the present invention, the navigation instruction and the topological level semantic map are input into a large language model to identify the starting area and target area corresponding to the starting point and the end point in the navigation instruction in the topological level semantic map, including: if the navigation instruction is a voice instruction, based on a voice recognition algorithm, the navigation instruction is voice-to-text processed to generate a textual navigation instruction; if the navigation instruction is a text instruction, the navigation instruction is kept unchanged; the navigation instruction and the topological level semantic map are input into a large language model to identify the starting area corresponding to the starting point in the navigation instruction in the topological level semantic map, and the target area corresponding to the end point.

[0006] In one embodiment of the present invention, the target path is planned based on the A* algorithm according to each channel passed from the starting area to the target area, including: based on the A* algorithm, an expansion search is performed from the starting area to the target area, and the total cost of each channel of the path is calculated for each expansion search, and the channel with the smallest total cost for each expansion search is marked as a path channel for storage; and the saved paths are sequentially connected to obtain the target path from the starting area to the target area.

[0007] In one embodiment of the present invention, the process of calculating the total cost of a channel includes: taking a starting area as the starting point for expansion, performing an expansion search from the starting area to the target area, and taking the area at each expansion as the path area, obtaining all channels connected to the path area in the topological hierarchy semantic map; for each channel: calculating the cumulative cost of the current channel based on the cost of the current channel and the costs of each channel in the path; wherein the cost of the channel is preset; calculating the heuristic cost of the current channel based on the distance from the current channel to the target area; and obtaining the total cost of the current channel based on the cumulative cost and the heuristic cost.

[0008] In one embodiment of the present invention, the large language model is ChatGPT-4o.

[0009] In one embodiment of the present invention, the detecting whether there is an impassable area in the target path, updating the target path accordingly according to the detection result, and obtaining the final path includes: detecting whether there is an impassable area in the target path: if there is, replanning the target path based on the impassable area and the A* algorithm; if there is no impassable area, taking the target path as the final path.

[0010] In one embodiment of the present invention, the target path is updated based on the impassable area based on the A* algorithm, including: deleting the channel corresponding to the impassable area from each channel passing from the starting area to the target area, and updating the channel from the starting area to the target area; based on the A* algorithm, planning a new target path from each updated channel.

[0011] In one embodiment of the present invention, a path planning system is also provided, the system comprising: an instruction acquisition module, used to acquire navigation instructions and a topological level semantic map corresponding thereto; wherein the topological level semantic map uses each marking point as an area and uses channels to connect each interconnected area; an area determination module, used to input the navigation instructions and the topological level semantic map into a large language model, and identify the starting area and target area corresponding to the starting point and the end point in the navigation instructions in the topological level semantic map; an initial planning module, used to plan a target path based on the A* algorithm according to each channel passing from the starting area to the target area; a verification module, used to detect whether there is an impassable area in the target path, and update the target path accordingly according to the detection result to obtain a final path; wherein the impassable area is pre-generated by the large language model according to external information, and the external information is used to characterize the corresponding channel state.

[0012] In one embodiment of the present invention, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor of a computer, the computer executes any one of the above-mentioned path planning methods.

[0013] In one embodiment of the present invention, a robot is further provided, wherein the robot uses any of the above-mentioned path planning methods to perform path navigation.

[0014] As described above, a path planning method, system, medium and robot of the present invention have the following beneficial effects: the navigation instructions and the corresponding topological hierarchical semantic map are input into the large language model so as to accurately identify the starting area and target area of ​​the starting point and end point in the navigation instructions in the map. The optimal path from the starting area to the target area is planned based on the A* algorithm, and at the same time, whether there is an impassable area in the path is detected, and the path is dynamically updated to generate a final feasible path. Among them, since the information of the impassable area is pre-generated by the large language model through external data, the navigation system can adapt to environmental changes in a timely manner, improve the accuracy and robustness of navigation, and make the final generated navigation path can be flexibly and dynamically adjusted in combination with external information, so as to generate the optimal navigation path. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1A schematic diagram of a flow chart of a path planning method provided by an embodiment of the present invention;

[0016] Figure 2 Shown is a schematic diagram of the overall process of a path planning system provided by an embodiment of the present invention;

[0017] Figure 3a The displayed route information is based on distance only.

[0018] Figure 3b Displays the route information planned based on distance and cost;

[0019] Figure 3c Displaying the route information planned for the method of the present invention;

[0020] Figure 3d Displaying information provided to the robot by means of the present invention;

[0021] Figure 4 It is a structural block diagram of the path planning system of the present invention. DETAILED DESCRIPTION

[0022] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0023] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0024] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0025] The inventors found that robots face significant challenges navigating large, dynamic environments because they cannot always perceive or update their surroundings in real time. A vivid real-world example is a campus delivery robot encountering a path obstruction caused by an unexpected pipe maintenance closure. Despite advance notice on a public website, the robot's navigation algorithm was unaware of the obstacle. Robots face unique challenges navigating complex indoor environments, such as large campus buildings, due to frequent changes in the environment, such as doors opening and closing, furniture moving, and occasional external temporary events, such as fire drills or wireless network updates, which the robot needs to avoid. Given the scale and dynamic nature of these spaces, it is impossible for robots to generate new maps for each task. Instead, they should use a permanent map that only contains the infrastructure and adjust their behavior based on real-time conditions. When humans navigate familiar buildings, they use experiential knowledge - such as avoiding doors that are usually closed or choosing automatic doors when carrying items - as well as external information, such as public notices or campus news.

[0026] To improve the above problems, the present invention provides a path planning method that utilizes the capabilities of large language models that can process general text-based information and incorporate human-like contextual understanding into the robot system. By using a large language model as a co-pilot, the present invention enhances lidar-based navigation by integrating socially relevant information such as public notices, ensuring the adaptability and robustness of robot navigation to the human environment. In addition, the topological hierarchical semantic map proposed by the present invention is in text format, which is convenient for users to edit and incorporate preference information, thereby ensuring compatibility with large language models. This map is easy to integrate seamlessly into existing robot systems, ensuring a smooth navigation process, and can also minimize the need for frequent updates due to environmental changes, thereby improving the stability and maintenance efficiency of the map.

[0027] See also Figure 1 , the path planning method includes the following steps:

[0028] S1. Obtain navigation instructions and a corresponding topological level semantic map; wherein the topological level semantic map uses each marked point as an area and uses channels to connect each interconnected area.

[0029] The present invention uses osmAG map as a topological semantic map, which is based on polygons and defines physical spaces such as corridors and rooms as regions, and polygonal segments connecting two adjacent regions are defined as channels. As an example, the corridors and rooms in a certain teaching building are used as regions, and the doors of the rooms are used as channels, which constitute the topological semantic map of the teaching building. osmAG can be easily edited by humans to add semantic information, such as marking an area as a "robot training laboratory", making it very suitable for adapting to human environments.

[0030] S2. Input the navigation instruction and the topological level semantic map into a large language model, and identify the starting area and the target area corresponding to the starting point and the end point in the navigation instruction in the topological level semantic map.

[0031] The user inputs navigation instructions including the starting position and the target position through a relevant interface (such as a mobile phone application or a robot control panel, etc.), wherein the navigation instructions are in the form of natural language. The navigation instructions and the topological hierarchical semantic map are input into the large language model together. The large language model can accurately identify the starting point and end point information in the navigation instructions, and match the corresponding areas in the topological hierarchical semantic map, so as to determine the starting area and target area corresponding to the starting point and the end point in the map. The large language model can be any model that provides natural language processing tasks, including but not limited to RoBERTa, GPT-3, ChatGPT-4o, etc. In order to improve the intelligence and adaptability of the navigation system, in one embodiment of the present invention, the large language model is ChatGPT-4o.

[0032] In one embodiment of the present invention, the step of inputting the navigation instruction and the topological hierarchical semantic map into a large language model and identifying the starting area and the target area corresponding to the starting point and the end point in the navigation instruction in the topological hierarchical semantic map includes:

[0033] If the navigation instruction is a voice instruction, then based on a voice recognition algorithm, the navigation instruction is processed by voice-to-text conversion to generate a textual navigation instruction;

[0034] If the navigation instruction is a text instruction, keep the navigation instruction unchanged;

[0035] The navigation instruction and the topological level semantic map are input into a large language model to identify the starting area corresponding to the starting point in the navigation instruction in the topological level semantic map and the target area corresponding to the end point.

[0036] After receiving the navigation instruction, the navigation instruction will be detected. If the navigation instruction is a voice instruction, a speech recognition algorithm is used to convert the instruction into a text format, wherein the speech recognition algorithm includes but is not limited to a hidden Markov model, a deep neural network, and the like. Furthermore, in order to improve the accuracy of subsequent recognition, the present invention also performs text cleaning on the converted textual navigation instruction to obtain a final textual instruction, wherein text cleaning includes but is not limited to correcting grammatical errors, removing noise, and performing vocabulary correction according to the context, so that the generated textual navigation instruction is more accurate. If the navigation instruction is already a textual instruction when it is received, it does not need to be processed. The textual instruction and the corresponding topological level semantic map are input into the large language model together. The model extracts the starting point and end point information from the navigation instruction, and matches it in combination with the topological level semantic map to identify the starting area and target area corresponding to the area in the map.

[0037] S3. Based on the A* algorithm, a target path is planned according to each channel from the starting area to the target area.

[0038] In one embodiment of the present invention, based on the A* algorithm, an expansion search is performed from the starting area to the target area, the total cost of each channel of the path during each expansion search is calculated, and the channel with the smallest total cost of each expansion search is marked as a path channel for storage;

[0039] The saved pathways are connected in sequence to obtain a target path from the starting area to the target area.

[0040] Based on the A* algorithm, an extended search is performed with the starting area as the starting point. The starting area is marked as the first path area, and the open list and closed list are initialized, where the open list contains the area to be evaluated and the closed list contains the evaluated area. The initialized open list only contains the starting area, and the initialized closed list is empty. It should be noted that the open list will change dynamically as the algorithm progresses, and each neighbor area connected to the path area and its corresponding total cost will be added. Each time an extended search is performed, the path area with the lowest total cost is selected from the open list for processing. Specifically, for each path area, all channels connected to the area are obtained, and the total cost of each channel is calculated, and the total cost of the channel is used as the total cost of the corresponding neighbor area connected to it. After all channels in the path area have calculated the total cost, the channel with the smallest total cost is identified as the path channel and saved, the corresponding neighbor area is identified as the new path area, and the old path area is moved from the open list to the closed list. Repeat the above process until the target area is found or the open list is empty. Connect the saved path channels in sequence to obtain the target path from the starting area to the target area. The target path includes each channel from the starting area to the target area.

[0041] In one embodiment of the present invention, the process of calculating the total channel cost includes:

[0042] Taking the starting area as the expansion starting point, performing expansion search from the starting area to the target area, and taking the area at each expansion as the path area, obtaining all channels connected to the path area in the topological hierarchical semantic map;

[0043] For each channel:

[0044] Calculate the cumulative cost of the current channel according to the cost of the current channel and the costs of each channel in the path; wherein the cost of the channel is preset;

[0045] Calculate the heuristic cost of the current channel based on the distance from the current channel to the target area;

[0046] According to the cumulative cost and the heuristic cost, the total cost of the current channel is obtained.

[0047] For each pathway area, it is necessary to calculate the total cost of all channels connected to it, and select the channel with the smallest total cost from it, and use the neighboring area corresponding to the channel as the new pathway area to perform the above process. For each pathway area, when calculating the total cost of the channels connected to it, for each channel: calculate the cumulative cost of the current channel based on the preset cost of the channel and the cost of all channels previously passed by the pathway area. Among them, the cumulative cost represents the comprehensive cost of all channels from the starting area to the current pathway area, which is used to characterize the actual cost of the path. According to the actual distance from the current channel to the target area, calculate the heuristic cost of the current channel. Among them, the heuristic cost can effectively estimate the cost from the current channel to the target area. Add the cumulative cost and the heuristic cost to get the total cost of the current channel.

[0048] S4. Detect whether there is an impassable area in the target path, and update the target path accordingly according to the detection result to obtain a final path; wherein the impassable area is pre-generated by the large language model according to external information, and the external information is used to characterize the corresponding channel state.

[0049] The large language model will continuously collect external information, input this external information and the corresponding topological level semantic map into the large language model, and the model will identify the areas in the map affected by the external information, and save these affected areas and the channels connected to them. These affected areas are recorded as impassable areas, and the channels connected to them are marked as impassable channels. Among them, the external information includes but is not limited to email notifications, public announcements, and social media updates, as long as the information can represent the current state of the environment. Specifically, the external information and the corresponding topological level semantic map are input into the large language model, and the model can identify the influencing factors that affect the channel state or regional access, wherein the influencing factors include but are not limited to temporary closure of the area, regional construction, or suspension of wireless networks in the area. These influencing factors will affect the walking path of the robot. Therefore, it is necessary to match each channel passed through in the generated target path with the saved affected channels to determine whether there is an impassable area in the target path. If not, the target path is navigated as the final path. On the contrary, if there is an impassable area in the target path, it is necessary to further process these impassable areas so that there is no impassable area in the final path.

[0050] In an embodiment of the present invention, the detecting whether there is an impassable area in the target path, and updating the target path accordingly according to the detection result to obtain the final path includes:

[0051] Check whether there is an impassable area in the target path:

[0052] If it exists, replan the target path based on the impassable area and the A* algorithm;

[0053] If it does not exist, the target path is used as the final path.

[0054] When the large language model generates an impassable area, it will simultaneously save the channels connected to the area and mark these channels as impassable channels. For the target path generated for the first time, first check whether there is an impassable channel in the target path. If so, it means that there is an impassable area in the target path. At this time, the A* algorithm is required to recalculate the cost of each channel and plan a new target path. This new target path does not contain the previously marked impassable area. For the new target path, since it may contain new impassable areas, it is necessary to follow the above process again to check whether each channel passed through it contains an impassable channel. Repeat the above process until there is no impassable area in the final generated target path. The path at this time is recorded as the final path. The robot can navigate from the starting area to the target area along this path. In this way, combined with changes in external information, it can effectively respond to changes in dynamic environments.

[0055] In one embodiment of the present invention, updating the target path based on the impassable area and the A* algorithm includes:

[0056] Deleting the channel corresponding to the impassable area from each channel from the starting area to the target area, and updating the channel from the starting area to the target area;

[0057] Based on the A* algorithm, a new target path is planned from the updated channels.

[0058] If there is an impassable area in the target path, the impassable channels corresponding to the impassable area are identified and deleted from all channels from the starting area to the target area in the backup list of the topological hierarchy semantic map, and the backup list is updated so that only the passable channels and corresponding areas are included in the backup list. Then, based on the A* algorithm, the total cost of each channel is calculated from the updated backup list according to the channel cost and heuristic cost, and the channel with the lowest total cost at each expansion is gradually expanded from the starting area to the target area, and a new target path is obtained. Each channel passed by the new target path is matched with the impassable channels obtained by the pre-saved large language model to detect whether there is an impassable area in the target path. If not, the target path is used as the final path, and the backup list is restored to the initial state. Among them, the backup list in the initial state contains all channels and corresponding area information, which can provide complete topological information and ensure that all possible channels and areas can be accessed without being affected by the previous impassable areas or channels. On the contrary, if there is still an impassable area, the above process needs to be repeated until there is no impassable area in the target path.

[0059] like Figure 2As shown, when the method described in the present invention is used for path planning, it involves the NavigationEventMonitor module, the osmAGPathPlanner module and the PassageCostEvaluator module, wherein the NavigationEventMonitor module receives external information in real time and determines its impact on the robot navigation task to ensure that the system can adapt to the dynamic environment; the PassageCostEvaluator module scores each channel according to the large language model and historical experience, and outputs the target area and its passing cost to provide a basis for path planning; the osmAGPathPlanner module uses the A* algorithm to calculate the shortest path from the starting area to the target area and ensure the feasibility of the path. Specifically, the NavigationEventMonitor module continuously receives external information, such as time, weather, elevator maintenance notifications, and wireless network upgrade notifications. The module calls the large language model to determine whether this information will affect the robot's future navigation tasks. If it is determined that there is an impact, the module will generate a JSON string and save the relevant information to the document. When receiving a human navigation command, the PassageCostEvaluator module will call the large language model to identify the target room in the navigation command based on the information of the osmAG map. At the same time, the module also needs to score each door based on historical experience and the type of door. For example, automatic doors have lower passing costs because they do not need to be opened manually. Finally, the module outputs the target area and the cost of each channel to the osmAGPathPlanner module. After receiving the target area and the cost of each channel, the osmAGPathPlanner module uses the A* algorithm to calculate the shortest path P (the path is a list of all the areas to be passed). Next, the calculated shortest path will be input into the PassageCostEvaluator module to determine whether the path is feasible. Specifically, the module will confirm the feasibility of the path based on external information. For example, the path cannot include elevators under maintenance, and for wheeled robots, the path cannot pass through areas that cannot be driven (such as stairs). Once the path is reviewed and approved by the PassageCostEvaluator module, the final path will be sent to the Move_base Planner module of ROS and packaged into a function package for execution.

[0060] like Figure 2As shown, as a specific example, the navigation instruction is "Please take this document to the robot laboratory". The PassageCostEvaluator module uses the large language model (LLM) to parse the target area in the instruction and evaluates the cost of each channel based on the osmAG map. For example, after identifying the failure record of the channel "E1c-F1-COR-02_to_outside" reaching the preset number of times, the channel is given a higher cost so that the osmAGPathPlanner module avoids using this channel and outputs the channel cost evaluation value in JSON format. The osmAGPathPlanner module uses the osmAG map and the channel cost information to plan the shortest path as the target path from the starting area to the target area through the A* algorithm. Then, the various areas passed in the target path are compared with the impassable areas generated by the NavigationEventMonitor module to detect whether there are impassable areas in the current target path. If so, these impassable areas and corresponding channels are deleted from the pre-stored backup map, and the osmAGPathPlanner module is reused to generate a new target path using the A* algorithm again. Among them, the NavigationEventMonitor module continuously receives external information (such as general information such as elevator maintenance notifications and wireless network upgrade notifications), and calls the large language model combined with historical information to identify areas affected by external information, and converts this information into JSON format for use by other modules. When the system needs to perform a navigation task, the osmAG map is converted into an occupancy grid map and provided to the Move_base Planner module as a global map to support the Move_base Planner module of ROS to implement low-level control. After the path is successfully planned, the Move_base Planner module performs the navigation task and updates the feasible channels and inaccessible areas through the recorded channel accessibility information. In this way, the system can flexibly adjust the navigation path in a changing environment, effectively deal with dynamic obstacles and ensure that the robot can reach the target location smoothly.

[0061] When a robot navigates within a large building, path determination relies heavily on choosing which corridors to take as movement between different areas requires traversing these corridors. The cost of traversing an area depends on the specific entry and exit corridors used. However, when humans talk about locations, we typically refer to general areas (e.g., rooms, hallways, etc.). For example, we might say that a fire drill will take place in Area B of a certain building, rather than specifying which doors will be affected. This difference necessitates different classification and treatment of corridors and areas when using a large language model (LLM) as a co-pilot for robot navigation. Planning a path based solely on path distance yields something like Figure 3aThe path diagram shown in the figure shows that the green lines represent the paths from one area to the adjacent areas, and the purple lines represent the final target path after taking into account the path distance. When the door on the left is often not opened, the door on the left is given a greater cost. At this time, the A* algorithm will get the following Figure 3b The purple line shows the target path. Further, combined with external information, if Figure 3c The yellow area is an impassable area. The replanned target path is shown by the purple line. Since there is no impassable area in the target path, the final path is this path. Figure 3d As shown, each channel passed through the path is sent to the robot as a target in sequence to achieve robot navigation.

[0062] As shown in Table 1, the following four cases respectively represent the robot's walking paths in a certain teaching building when there is a graduation party in the hall, classroom decoration in area D, fire drills in the hall and area B, and wireless network upgrades in the hall and area B, which are impassable areas. In 37 experiments in the four experimental cases, the method of the present invention enables the robot to always avoid restricted areas and successfully reach the designated destination. Compared with the conventional move_base that clears the obstacle layer between each experiment, the total path length is reduced by 58%. And the average path length of navigation in the method of the present invention is always kept at a low level.

[0063] Table 1 Average path lengths for different navigation modes

[0064]

[0065] See also Figure 4 , the path planning system 100 includes: an instruction acquisition module 110, an area determination module 120, an initial planning module 130 and a verification module 140. The instruction acquisition module 110 is used to obtain navigation instructions and the corresponding topological level semantic map; wherein the topological level semantic map uses each marked point as an area and uses channels to connect each interconnected area. The area determination module 120 is used to input the navigation instructions and the topological level semantic map into the large language model, and identify the starting area and target area corresponding to the starting point and the end point in the navigation instructions in the topological level semantic map. The initial planning module 130 is used to plan the target path based on the A* algorithm according to each channel from the starting area to the target area. The verification module 140 is used to detect whether there is an impassable area in the target path, and update the target path according to the detection result to obtain the final path; wherein the impassable area is pre-generated by the large language model according to external information, and the external information is used to characterize the corresponding channel state.

[0066] For the specific definition of the path planning system, please refer to the definition of the path planning method above, which will not be repeated here. Each module in the above path planning system can be implemented in whole or in part by software, hardware and a combination thereof. The above modules can be embedded in or independent of the processor in the computer device in hardware format, or stored in the memory of the computer device in software format, so that the processor can call the operations corresponding to the above modules.

[0067] It should be noted that, in order to highlight the innovative part of the present invention, the present embodiment does not introduce modules that are not closely related to solving the technical problem proposed by the present invention, but this does not mean that there are no other modules in the present embodiment.

[0068] The present invention also provides a robot, which uses any of the above-mentioned path planning methods for navigation.

[0069] The above-mentioned integrated unit implemented in the form of a software function module can be stored in a computer-readable storage medium, and the computer-readable storage medium can be non-volatile or volatile. The above-mentioned software function module is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a computer device, or a network device, etc.) or a processor to perform part of the functions of the path planning method described in each embodiment of the present application.

[0070] In summary, the present invention discloses a path planning method, system, medium and robot, which inputs navigation instructions and corresponding topological hierarchical semantic maps into a large language model so as to accurately identify the starting area and target area of ​​the starting point and end point in the navigation instructions in the map. The optimal path from the starting area to the target area is planned based on the A* algorithm, and at the same time, it is detected whether there is an impassable area in the path, and the path is dynamically updated to generate a final feasible path. Among them, since the information of the impassable area is pre-generated by the large language model through external data, the navigation system can adapt to environmental changes in a timely manner, improves the accuracy and robustness of navigation, and enables the final generated navigation path to be combined with external information, and can be flexibly adjusted dynamically according to changes in the external environment, thereby generating the optimal navigation path. Therefore, the present invention effectively overcomes the various shortcomings in the prior art and has a high industrial utilization value.

[0071] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by a person of ordinary skill in the art without departing from the spirit and technical concept disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A path planning method, characterized in that: The method comprises: Obtaining navigation instructions and a corresponding topological hierarchical semantic map; wherein the topological hierarchical semantic map uses each marked point as an area and uses channels to connect each interconnected area; Inputting the navigation instruction and the topological level semantic map into a large language model, identifying the starting area and the target area corresponding to the starting point and the end point in the navigation instruction in the topological level semantic map; Based on the A* algorithm, a target path is planned according to each channel from the starting area to the target area; Detect whether there is an impassable area in the target path, and update the target path accordingly according to the detection result to obtain a final path; wherein the impassable area is pre-generated by the large language model according to external information, and the external information is used to characterize the corresponding channel state.

2. The path planning method according to claim 1, characterized in that: The step of inputting the navigation instruction and the topological level semantic map into the large language model and identifying the starting area and the target area corresponding to the starting point and the end point in the navigation instruction in the topological level semantic map includes: If the navigation instruction is a voice instruction, then based on a voice recognition algorithm, the navigation instruction is processed by voice-to-text conversion to generate a textual navigation instruction; If the navigation instruction is a text instruction, keep the navigation instruction unchanged; The navigation instruction and the topological level semantic map are input into a large language model to identify the starting area corresponding to the starting point in the navigation instruction in the topological level semantic map and the target area corresponding to the end point.

3. The path planning method according to claim 1, characterized in that: The A* algorithm is used to plan a target path according to each channel from the starting area to the target area, including: Based on the A* algorithm, an expansion search is performed from the starting area to the target area, the total cost of each channel of the path is calculated for each expansion search, and the channel with the smallest total cost of each expansion search is marked as a path channel for storage; The saved pathways are connected in sequence to obtain a target path from the starting area to the target area.

4. The path planning method according to claim 3, characterized in that: The process of calculating the total channel cost includes: Taking the starting area as the expansion starting point, performing expansion search from the starting area to the target area, and taking the area at each expansion as the path area, obtaining all channels connected to the path area in the topological hierarchical semantic map; For each channel: Calculate the cumulative cost of the current channel according to the cost of the current channel and the costs of each channel in the path; wherein the cost of the channel is preset; Calculate the heuristic cost of the current channel based on the distance from the current channel to the target area; According to the cumulative cost and the heuristic cost, the total cost of the current channel is obtained.

5. The path planning method according to claim 1, characterized in that: The large language model is ChatGPT-4o.

6. The path planning method according to claim 1, characterized in that: The detecting whether there is an impassable area in the target path, and updating the target path accordingly according to the detection result to obtain a final path includes: Check whether there is an impassable area in the target path: If it exists, replan the target path based on the impassable area and the A* algorithm; If it does not exist, the target path is used as the final path.

7. The path planning method according to claim 6, characterized in that: The updating of the target path according to the impassable area based on the A* algorithm includes: Deleting the channel corresponding to the impassable area from each channel from the starting area to the target area, and updating the channel from the starting area to the target area; Based on the A* algorithm, a new target path is planned from the updated channels.

8. A path planning system, characterized in that: The system comprises: An instruction acquisition module is used to acquire navigation instructions and a topological hierarchical semantic map corresponding thereto; wherein the topological hierarchical semantic map uses each marked point as an area and uses channels to connect each interconnected area; An area determination module, used for inputting the navigation instruction and the topological level semantic map into a large language model, and identifying the starting area and the target area corresponding to the starting point and the end point in the navigation instruction in the topological level semantic map; An initial planning module, for planning a target path based on the A* algorithm according to each channel from the starting area to the target area; A verification module is used to detect whether there is an impassable area in the target path, and update the target path accordingly according to the detection result to obtain a final path; wherein the impassable area is pre-generated by the large language model according to external information, and the external information is used to characterize the corresponding channel state.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor of a computer, the computer is enabled to execute the path planning method according to any one of claims 1 to 7.

10. A robot, characterized in that: The robot performs path navigation using the path planning method described in any one of claims 1 to 7.

Images