Robot behavior planning method, controller and robot for adaptive grasping tasks

Abstract

The present invention belongs to the field of mobile machine technology, solves the problem that robots cannot make accurate inferences and decisions based on objects, obstacles and their own physical operation capabilities in the surrounding environment, and provides a robot behavior planning method, controller and robot for adaptive grasping tasks. Among them, the robot behavior planning method for adaptive grasping tasks includes generating navigation instructions based on the global scene graph, the real-time position of the robot and the grasping task; judging whether the robot has the ability to complete the grasping task; if so, controlling the movement of the robot according to the navigation instructions to perform the grasping task; otherwise, when there are candidate tools matching the grasping task in the global scene graph, calculating the scores of each candidate tool, selecting the one with the highest score as the grasping tool and updating the navigation instructions, and according to the updated navigation instructions, improving the flexibility and adaptability of the robot in complex environments.

Description

Technical Field

[0001] The present invention belongs to the technical field of mobile machines, and in particular relates to a robot behavior planning method for an adaptive grasping task, a controller and a robot. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Although the existing mobile grasping robots have added human-computer interaction modules to improve the robot's ability to understand high-level languages, accurately understand human intentions and instructions, and achieve the purpose of effective communication between the robot and humans when performing tasks, and flexibly adapt to changes in human needs. However, the existing robot system still has the following problems:

[0004] When existing mobile grasping robots handle complex tasks, they are unable to perform accurate reasoning and decision-making based on objects and obstacles in the surrounding environment and their own physical operation capabilities, which limits the robot's flexibility and adaptability in complex environments. Moreover, when performing tasks, the robot is unable to select the most appropriate tools and strategies based on the complexity of the task and its own capabilities, resulting in the robot's inefficiency in handling complex tasks or even failure to complete the task. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a robot behavior planning method, a controller and a robot for adaptive grasping tasks, which can autonomously perceive and identify the environment, plan navigation routes in complex environments, and help the robot meet the needs of intelligent operation in complex environments.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A first aspect of the present invention provides a robot behavior planning method for an adaptive grasping task.

[0008] A robot behavior planning method for an adaptive grasping task, comprising:

[0009] Get the global scene graph of the current scene and the capture tasks in the global scene graph;

[0010] Generate navigation instructions based on the global scene graph, the robot's real-time position and grasping tasks;

[0011] Determine whether the robot has the ability to complete the grasping task;

[0012] If yes, the robot is controlled to move according to the navigation instruction to perform the grasping task;

[0013] Otherwise, when there is a candidate tool matching the grasping task in the global scene graph, the scores of the candidate tools are calculated, the one with the highest score is selected as the grasping tool and the navigation instruction is updated, and according to the updated navigation instruction, the robot is controlled to obtain the grasping tool to perform the grasping task;

[0014] The process of calculating the scores of each candidate tool is as follows:

[0015] According to the preset large language model and the crawling task, a set of candidate tool solutions for executing the crawling task is obtained;

[0016] Obtaining the common sense usage frequency of each candidate tool in the candidate tool solution set in the common sense experience database, the scene usage frequency in the current scene type, and the appearance frequency in the global scene graph;

[0017] The common sense usage frequency and scene usage frequency of each candidate tool are respectively normalized to obtain the common sense weight and scene weight of each candidate tool; according to the frequency of occurrence, the distance from the current position of the robot to each candidate tool is standardized to obtain the standardized distance of each candidate tool;

[0018] The corresponding score is calculated based on the weighted sum of the occurrence frequency and the standardized distance of each candidate tool; the weight of the occurrence frequency is: the proportion of the common sense weight in the cumulative sum of the common sense weight and the scenario weight; the weight of the standardized distance is: the opposite of the proportion of the scenario weight in the cumulative sum of the common sense weight and the scenario weight.

[0019] As an implementation method, the standardized distance expression of each candidate tool is:

[0020]

[0021] in, Indicates Candidate tools; Indicates the robot moves from the current position to the The distance between the candidate tools; Indicates The frequency of occurrence of candidate tools in the global scene graph; is a positive integer greater than or equal to 1.

[0022] As an implementation mode, when the difference between the common sense usage frequency and the scene usage frequency of each candidate tool is greater than a set multiple of the occurrence frequency in the global scene graph, the common sense usage frequency and the scene usage frequency of each candidate tool are updated.

[0023] As an implementation method, when the difference between the common sense usage frequency and the scene usage frequency of each candidate tool is greater than three times the appearance frequency in the global scene graph, the common sense usage frequency of the corresponding candidate tool is updated to twice the appearance frequency in the global scene graph, and the scene usage frequency of the corresponding candidate tool is updated to be equal to the appearance frequency in the global scene graph.

[0024] As an implementation mode, when there is no candidate tool matching the grasping task in the global scene graph, the robot is controlled to interact with the user to obtain a candidate tool matching the grasping task.

[0025] As an implementation method, the process of obtaining the grabbing task in the global scene graph is:

[0026] Get the grasping task and determine whether there is a conflict between the grasping task and the global scene graph. If there is no conflict, the grasping task remains unchanged.

[0027] Otherwise, the grasping task is updated by interacting with the user to ensure that the grasping task does not conflict with the global scene graph.

[0028] As an implementation mode, before determining whether the robot has the ability to complete the grasping task, the following steps are further included:

[0029] It is necessary to retrieve from the preset knowledge database whether there is a global scene graph and / or a grasping task for the current scene. If so, the corresponding global scene graph and / or the navigation instructions corresponding to the grasping task are directly retrieved from the preset knowledge database; wherein, the preset knowledge database stores the associated storage information of the global scene graph, the grasping task and the navigation instructions.

[0030] As an implementation method, the global scene graph, grabbing tasks and navigation instructions of the current scene are fed back and stored in a preset knowledge database to update the preset knowledge database.

[0031] A second aspect of the present invention provides a robot behavior planning controller for an adaptive grasping task.

[0032] A robot behavior planning controller for an adaptive grasping task, comprising:

[0033] An information acquisition module, which is used to obtain the global scene graph of the current scene and the capture tasks in the global scene graph;

[0034] The command generation module is used to generate navigation commands based on the global scene graph, the robot's real-time position and the grasping task;

[0035] A capability judgment module, which is used to judge whether the robot has the ability to complete the grasping task;

[0036] A grasping task execution module, which is used to control the movement of the robot according to the navigation instruction to execute the grasping task when the robot has the ability to complete the grasping task;

[0037] A candidate tool determination module is used to calculate the scores of various candidate tools when the robot does not have the ability to complete the grasping task and there are candidate tools matching the grasping task in the global scene graph, select the one with the highest score as the grasping tool and update the navigation instructions, and control the robot to obtain the grasping tool to perform the grasping task according to the updated navigation instructions;

[0038] Wherein, in the candidate tool determination module, the process of calculating the score of each candidate tool is as follows:

[0039] According to the preset large language model and the crawling task, a set of candidate tool solutions for executing the crawling task is obtained;

[0040] Obtaining the common sense usage frequency of each candidate tool in the candidate tool solution set in the common sense experience database, the scene usage frequency in the current scene type, and the appearance frequency in the global scene graph;

[0041] The common sense usage frequency and scene usage frequency of each candidate tool are respectively normalized to obtain the common sense weight and scene weight of each candidate tool; according to the frequency of occurrence, the distance from the current position of the robot to each candidate tool is standardized to obtain the standardized distance of each candidate tool;

[0042] The corresponding score is calculated based on the weighted sum of the occurrence frequency and the standardized distance of each candidate tool; the weight of the occurrence frequency is: the proportion of the common sense weight in the cumulative sum of the common sense weight and the scenario weight; the weight of the standardized distance is: the opposite of the proportion of the scenario weight in the cumulative sum of the common sense weight and the scenario weight.

[0043] A third aspect of the present invention provides a robot.

[0044] A robot, comprising:

[0045] A robot body, a grasping actuator, and a robot behavior planning controller for the adaptive grasping task as described above;

[0046] The robot behavior planning controller and the grasping actuator of the adaptive grasping task are both installed on the robot body.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] (1) The present invention generates navigation instructions based on the global scene graph, the real-time position of the robot and the grasping task, and then determines whether to directly execute the navigation instructions or to search for the candidate tool with the highest score to update the navigation instructions according to whether the robot has the ability to complete the corresponding grasping task. This realizes autonomous perception and recognition of the environment, plans navigation routes in complex environments, and improves the flexibility and adaptability of the robot in complex environments as well as the efficiency of the robot in handling complex grasping tasks.

[0049] (2) In the process of calculating the score of each candidate tool, the present invention calculates the corresponding score based on the weighted sum of the appearance frequency and the standardized distance of each candidate tool, taking into account the common sense usage frequency of each candidate tool in the common sense experience database, the scene usage frequency in the current scene type, and the appearance frequency in the global scene graph, so that the robot can efficiently utilize the tools in the environment, optimize the execution process of the grasping task, and improve the success rate and efficiency of completing the grasping task. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0051] Figure 1 is a flow chart of a robot behavior planning method for an adaptive grasping task according to an embodiment of the present invention;

[0052] Figure 2 is a flowchart of calculating the scores of each candidate tool according to an embodiment of the present invention;

[0053] Figure 3 Schematic diagram of the robot behavior planning principle for the adaptive grasping task according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0055] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0056] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0057] In this embodiment, the robot is a multi-degree-of-freedom robot equipped with a variety of sensors such as a three-dimensional laser camera, a depth camera, an inertial measurement unit and a microphone array.

[0058] For example, the robot uses a SegWay chassis for navigation and movement in the scene; the robot arm uses a Realman robot arm, and the gripper is a Dahuan RGD-35-30 gripper for grabbing the target object; the laser radar is deployed on the top of the mobile robot to capture external point cloud information. The visual sensor is placed near the laser radar and will cooperate with the laser radar to perceive and model the external environment. The inertial measurement unit is located in the center of the robot body and is used to collect the robot's posture information. The camera can use a Realsense D435 camera to collect image data of the object to be grabbed in the scene.

[0059] Figure 1 FIG. 1 is a flow chart of a robot behavior planning method for an adaptive grasping task according to an embodiment of the present invention. Figure 1 As shown, a robot behavior planning method for an adaptive grasping task is provided, which includes the following steps:

[0060] S100: Obtaining a global scene graph of the current scene and a capture task in the global scene graph;

[0061] S200: Generate navigation instructions based on the global scene graph, the robot's real-time position and the grasping task;

[0062] S300: Determine whether the robot has the ability to complete the grasping task;

[0063] S400: If yes, control the robot to move according to the navigation instruction to perform the grasping task;

[0064] S500: Otherwise, when there are candidate tools matching the grasping task in the global scene graph, the scores of the candidate tools are calculated, the one with the highest score is selected as the grasping tool and the navigation instructions are updated. According to the updated navigation instructions, the robot is controlled to obtain the grasping tool to perform the grasping task.

[0065] This embodiment generates navigation instructions based on the global scene graph, the robot's real-time position and the grasping task, and then determines whether to directly execute the navigation instructions or to find the candidate tool with the highest score to update the navigation instructions based on whether the robot has the ability to complete the corresponding grasping task. This achieves autonomous perception and recognition of the environment, plans navigation routes in complex environments, and improves the robot's flexibility and adaptability in complex environments as well as the robot's efficiency in handling complex grasping tasks.

[0066] In step S100, relevant information of the grasping task includes the grasped target object, the grasped quantity and the position of the target object.

[0067] Specifically, the crawling task can be obtained by parsing the natural language instructions.

[0068] For example, the natural language is to grab four apples on the table at a set distance from the current robot;

[0069] The parsed relevant information of the grasping task includes: the target object to be grasped is apples, the number of grasps is four, and the position of the target object is on a table at a set distance from the current robot.

[0070] Figure 3 Schematic diagram of the robot behavior planning principle of the adaptive grasping task of the embodiment of the present invention; Figure 3 In the specific implementation process of step S100, a language understanding module can be used to parse natural language instructions. The language understanding module is implemented using a pre-trained large language model (such as GPT-4), and the process is: parsing the input natural language instructions, extracting key grasping task-related information, and converting it into instructions executable by the robot.

[0071] In one or more embodiments, before determining whether the robot has the ability to complete the grasping task, the process further includes:

[0072] It is necessary to retrieve from the preset knowledge database whether there is a global scene graph and / or a grasping task for the current scene. If so, the corresponding global scene graph and / or the navigation instructions corresponding to the grasping task are directly retrieved from the preset knowledge database; wherein, the preset knowledge database stores the associated storage information of the global scene graph, the grasping task and the navigation instructions.

[0073] In this embodiment, the process of constructing the global scene graph can be implemented using existing methods, which will not be described in detail here.

[0074] In step S100, the process of obtaining the grabbing task in the global scene graph is:

[0075] Get the grasping task and determine whether there is a conflict between the grasping task and the global scene graph. If there is no conflict, the grasping task remains unchanged.

[0076] Otherwise, the grasping task is updated by interacting with the user to ensure that the grasping task does not conflict with the global scene graph.

[0077] Among them, the global scene graph is a hierarchical scene graph, which can be used Indicates that represents a node set, Represents an edge set. By verifying the position of the target object and the global scene graph Whether the nodes in the scene graph match, and then determine whether there is a conflict between the parsed grasping task and the global scene graph.

[0078] In step S200, the real-time position of the robot is used as the starting point, the position of the target object is used as the end point, and then combined with the global scene graph to form a corresponding navigation instruction.

[0079] The navigation instructions here are determined by the path planning algorithm to obtain the optimal path for the robot from the current position to the tool position and the task target position, and the path planning algorithm can be implemented using an existing algorithm, which will not be described in detail here.

[0080] In step S300, according to the target object to be grasped and the required grasping quantity, the robot's own capability database is queried to determine whether it has the capability to complete the grasping task.

[0081] The robot's own capability database contains multiple operation skills, each of which is set to , Using the Large Language Model (LLM) based on the current global scene graph To analyze the task Generate and operate skills Can be broken down into specific operational actions :

[0082] .

[0083] In the specific implementation process of step S500, if Figure 2 As shown, the process of calculating the scores of each candidate tool is:

[0084] S501: Obtaining a set of candidate tool solutions for executing the crawling task according to a preset large language model and the crawling task;

[0085] S502: Obtaining the common sense usage frequency of each candidate tool in the candidate tool solution set in the common sense experience database, the scene usage frequency in the current scene type, and the appearance frequency in the global scene graph;

[0086] S503: Normalize the common sense usage frequency and scene usage frequency of each candidate tool to obtain the common sense weight and scene weight of each candidate tool; standardize the distance from the current position of the robot to each candidate tool according to the appearance frequency to obtain the standardized distance of each candidate tool;

[0087] S504: Calculate the corresponding score based on the weighted sum of the occurrence frequency and the standardized distance of each candidate tool; wherein the weight of the occurrence frequency is: the proportion of the common sense weight in the cumulative sum of the common sense weight and the scenario weight; the weight of the standardized distance is: the opposite of the proportion of the scenario weight in the cumulative sum of the common sense weight and the scenario weight.

[0088] In the process of calculating the score of each candidate tool, this embodiment calculates the corresponding score based on the weighted sum of the appearance frequency and the standardized distance of each candidate tool, taking into account the common sense usage frequency of each candidate tool in the common sense experience database, the scene usage frequency in the current scene type, and the appearance frequency in the global scene graph, so that the robot can efficiently utilize the tools in the environment, optimize the execution process of the grasping task, and improve the success rate and efficiency of completing the grasping task.

[0089] In step S501, the large language model is preset as an existing structural model, such as GPT-4.

[0090] The input of the preset large language model is the relevant information of the crawling task, and the output is the candidate tool solution set T={T 1, T2,...,T m}.

[0091] In step S502, let the first The common sense usage frequency of the candidate tools in the common sense experience database, the scene usage frequency in the current scene type, and the appearance frequency in the global scene graph are , , .

[0092] In order to avoid falling into the problem of over-reliance on certain tools, when the difference between the common sense usage frequency and the scene usage frequency of each candidate tool is greater than the set multiple of the appearance frequency in the global scene graph, the common sense usage frequency and the scene usage frequency of each candidate tool are updated.

[0093] For example, when the difference between the common sense usage frequency and the scene usage frequency of each candidate tool is greater than three times the appearance frequency in the global scene graph, the common sense usage frequency of the corresponding candidate tool is updated to twice the appearance frequency in the global scene graph, and the scene usage frequency of the corresponding candidate tool is updated to be equal to the appearance frequency in the global scene graph.

[0094] In step S503, the existing method can be used to Common sense usage frequency of candidate tools And scene usage frequency Normalize the frequencies respectively and get Common sense weights of candidate tools and scene weight .

[0095] Among them, The standardized distance expression of candidate tools is:

[0096]

[0097] in, Indicates Candidate tools; Indicates the robot moves from the current position to the The distance between the candidate tools; Indicates The frequency of occurrence of candidate tools in the global scene graph; is a positive integer greater than or equal to 1. This ensures that tools that are closer to the current robot can be used first.

[0098] In step S504, The candidate tools scored :

[0099]

[0100] In step S500, when there is no candidate tool matching the grasping task in the global scene graph, the robot is controlled to interact with the user to obtain a candidate tool matching the grasping task.

[0101] In the process of executing the grasping task, at the position of the target object, a work scene diagram is constructed according to the camera field of view carried by the robot, and then the position and posture of the target object are determined; according to the position and posture of the target object, the position and posture of the robot's grasping end are adjusted in real time to grasp the target object.

[0102] In the process of building the work scene graph, a preliminary incremental grid map is constructed through sensors. , and calculate the Grid cells The probability of occupation Next, we calculate the information entropy of the map based on these probabilities , to reflect the unknown extent of the map:

[0103]

[0104] Then, select a set of locations to be observed , simulate the robot to observe at these positions and calculate the updated information entropy Information Gain Represents the change in information entropy before and after observation:

[0105]

[0106] By comparing the information gain of each proposed observation position, the position with the largest information gain is selected for observation, thereby maximizing the amount of new information obtained in each observation, and gradually building and updating a more accurate environmental map. In addition, by extracting parts of the environment with larger information gain, the robot is driven to observe these areas, further improving the robot's observation accuracy.

[0107] After the target object is captured, the global scene graph, tasks to be executed and navigation instructions corresponding to the current scene information are fed back and stored in the preset knowledge database to update the preset knowledge database.

[0108] In one or more embodiments, after the execution of grabbing the target object is completed, the global scene graph of the current scene, the grabbing task and the navigation instruction are also fed back and stored in the preset knowledge database to update the preset knowledge database.

[0109] In one or more embodiments, a robot behavior planning controller for an adaptive grasping task is provided, comprising:

[0110] An information acquisition module, which is used to obtain the global scene graph of the current scene and the capture tasks in the global scene graph;

[0111] The command generation module is used to generate navigation commands based on the global scene graph, the robot's real-time position and the grasping task;

[0112] A capability judgment module, which is used to judge whether the robot has the ability to complete the grasping task;

[0113] A grasping task execution module, which is used to control the movement of the robot according to the navigation instruction to execute the grasping task when the robot has the ability to complete the grasping task;

[0114] A candidate tool determination module is used to calculate the scores of various candidate tools when the robot does not have the ability to complete the grasping task and there are candidate tools matching the grasping task in the global scene graph, select the one with the highest score as the grasping tool and update the navigation instructions, and control the robot to obtain the grasping tool to perform the grasping task according to the updated navigation instructions;

[0115] Wherein, in the candidate tool determination module, the process of calculating the score of each candidate tool is as follows:

[0116] According to the preset large language model and the crawling task, a set of candidate tool solutions for executing the crawling task is obtained;

[0117] Obtaining the common sense usage frequency of each candidate tool in the candidate tool solution set in the common sense experience database, the scene usage frequency in the current scene type, and the appearance frequency in the global scene graph;

[0118] The common sense usage frequency and scene usage frequency of each candidate tool are respectively normalized to obtain the common sense weight and scene weight of each candidate tool; according to the frequency of occurrence, the distance from the current position of the robot to each candidate tool is standardized to obtain the standardized distance of each candidate tool;

[0119] The corresponding score is calculated based on the weighted sum of the occurrence frequency and the standardized distance of each candidate tool; the weight of the occurrence frequency is: the proportion of the common sense weight in the cumulative sum of the common sense weight and the scenario weight; the weight of the standardized distance is: the opposite of the proportion of the scenario weight in the cumulative sum of the common sense weight and the scenario weight.

[0120] It should be noted here that each module in the robot behavior planning controller for the adaptive grasping task corresponds one-to-one to each step in the robot behavior planning method for the adaptive grasping task described above, and the specific implementation process is the same and will not be described in detail here.

[0121] In one or more embodiments, a robot is also provided, comprising:

[0122] A robot body, a grasping actuator, and a robot behavior planning controller for the adaptive grasping task as described above;

[0123] The robot behavior planning controller and the grasping actuator of the adaptive grasping task are both installed on the robot body.

[0124] It should be noted here that the robot body and the grasping actuator in this embodiment can be implemented using existing structures, which will not be described in detail here.

[0125] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A robot behavior planning method for adaptive grasping tasks, characterized in that: include: Get the global scene graph of the current scene and the capture tasks in the global scene graph; Generate navigation instructions based on the global scene graph, the robot's real-time position and grasping tasks; Determine whether the robot has the ability to complete the grasping task; If yes, the robot is controlled to move according to the navigation instruction to perform the grasping task; Otherwise, when there is a candidate tool matching the grasping task in the global scene graph, the scores of the candidate tools are calculated, the one with the highest score is selected as the grasping tool and the navigation instruction is updated, and according to the updated navigation instruction, the robot is controlled to obtain the grasping tool to perform the grasping task; The process of calculating the scores of each candidate tool is as follows: According to the preset large language model and the crawling task, a set of candidate tool solutions for executing the crawling task is obtained; Obtaining the common sense usage frequency of each candidate tool in the candidate tool solution set in the common sense experience database, the scene usage frequency in the current scene type, and the appearance frequency in the global scene graph; The common sense usage frequency and scene usage frequency of each candidate tool are respectively normalized to obtain the common sense weight and scene weight of each candidate tool; according to the frequency of occurrence, the distance from the current position of the robot to each candidate tool is standardized to obtain the standardized distance of each candidate tool; The corresponding score is calculated based on the weighted sum of the occurrence frequency and the standardized distance of each candidate tool; the weight of the occurrence frequency is: the proportion of the common sense weight in the cumulative sum of the common sense weight and the scenario weight; the weight of the standardized distance is: the opposite of the proportion of the scenario weight in the cumulative sum of the common sense weight and the scenario weight.

2. The robot behavior planning method for adaptive grasping tasks according to claim 1, characterized in that: The standardized distance expression of each candidate tool is: ； in, Indicates Candidate tools; Indicates the robot moves from the current position to the The distance between the candidate tools; Indicates The frequency of occurrence of candidate tools in the global scene graph; is a positive integer greater than or equal to 1.

3. The robot behavior planning method for adaptive grasping tasks according to claim 1, characterized in that: When the difference between the common sense usage frequency and the scene usage frequency of each candidate tool is greater than a set multiple of the appearance frequency in the global scene graph, the common sense usage frequency and the scene usage frequency of each candidate tool are updated.

4. The robot behavior planning method for adaptive grasping tasks as claimed in claim 3, characterized in that: When the difference between the common sense usage frequency and the scene usage frequency of each candidate tool is greater than three times the appearance frequency in the global scene graph, the common sense usage frequency of the corresponding candidate tool is updated to twice the appearance frequency in the global scene graph, and the scene usage frequency of the corresponding candidate tool is updated to be equal to the appearance frequency in the global scene graph.

5. The robot behavior planning method for adaptive grasping tasks according to claim 1, characterized in that: When there is no candidate tool matching the grasping task in the global scene graph, the robot is controlled to interact with the user to obtain the candidate tool matching the grasping task.

6. The robot behavior planning method for adaptive grasping tasks according to claim 1, characterized in that: The process of obtaining the grasping task in the global scene graph is: Get the grasping task and determine whether there is a conflict between the grasping task and the global scene graph. If there is no conflict, the grasping task remains unchanged. Otherwise, the grasping task is updated by interacting with the user to ensure that the grasping task does not conflict with the global scene graph.

7. The robot behavior planning method for adaptive grasping tasks according to claim 1, characterized in that: Before judging whether the robot has the ability to complete the grasping task, the following steps are also included: It is necessary to retrieve from the preset knowledge database whether there is a global scene graph and / or a grasping task for the current scene. If so, the corresponding global scene graph and / or the navigation instructions corresponding to the grasping task are directly retrieved from the preset knowledge database; wherein, the preset knowledge database stores the associated storage information of the global scene graph, the grasping task and the navigation instructions.

8. The robot behavior planning method for adaptive grasping tasks according to claim 7, characterized in that: The global scene graph, grasping tasks and navigation instructions of the current scene are fed back and stored in the preset knowledge database to update the preset knowledge database.

9. A robot behavior planning controller for adaptive grasping tasks, characterized in that: include: An information acquisition module, which is used to obtain the global scene graph of the current scene and the capture tasks in the global scene graph; The command generation module is used to generate navigation commands based on the global scene graph, the robot's real-time position and the grasping task; A capability judgment module, which is used to judge whether the robot has the ability to complete the grasping task; A grasping task execution module, which is used to control the movement of the robot according to the navigation instruction to execute the grasping task when the robot has the ability to complete the grasping task; A candidate tool determination module is used to calculate the scores of various candidate tools when the robot does not have the ability to complete the grasping task and there are candidate tools matching the grasping task in the global scene graph, select the one with the highest score as the grasping tool and update the navigation instructions, and control the robot to obtain the grasping tool to perform the grasping task according to the updated navigation instructions; Wherein, in the candidate tool determination module, the process of calculating the score of each candidate tool is as follows: According to the preset large language model and the crawling task, a set of candidate tool solutions for executing the crawling task is obtained; Obtaining the common sense usage frequency of each candidate tool in the candidate tool solution set in the common sense experience database, the scene usage frequency in the current scene type, and the appearance frequency in the global scene graph; The common sense usage frequency and scene usage frequency of each candidate tool are respectively normalized to obtain the common sense weight and scene weight of each candidate tool; according to the frequency of occurrence, the distance from the current position of the robot to each candidate tool is standardized to obtain the standardized distance of each candidate tool; The corresponding score is calculated based on the weighted sum of the occurrence frequency and the standardized distance of each candidate tool; the weight of the occurrence frequency is: the proportion of the common sense weight in the cumulative sum of the common sense weight and the scenario weight; the weight of the standardized distance is: the opposite of the proportion of the scenario weight in the cumulative sum of the common sense weight and the scenario weight.

10. A robot, characterized in that: include: A robot body, a grasping actuator, and a robot behavior planning controller for an adaptive grasping task as claimed in claim 9; The robot behavior planning controller and the grasping actuator of the adaptive grasping task are both installed on the robot body.

Images