Space intelligent reasoning method and system

Abstract

The invention relates to the technical field of data processing, in particular to a space intelligent reasoning method and system, and the method comprises the steps: generating a plurality of initial space reasoning tasks corresponding to a to-be-executed task instruction of a robot according to a pre-constructed space intelligent machine, and determining a first type of reasoning tasks and a second type of reasoning tasks from the initial space reasoning tasks, obtaining target space environment information corresponding to the first type of reasoning tasks and the second type of reasoning tasks, and obtaining structured data corresponding to the first type of reasoning tasks and the second type of reasoning tasks through a deep learning model, performing calculation according to the structured data corresponding to each first type of reasoning task by using a target calculation program corresponding to each first type of reasoning task in the sandbox simulator to obtain an accurate space prediction result; through the combination of the deep learning model and the sandbox simulator, the calculation accuracy can be improved, and the accuracy and reliability of overall space intelligent reasoning are enhanced.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a spatial intelligent reasoning method and system. Background Art

[0002] In the field of robotic navigation, spatial intelligence generally refers to robotic systems that achieve intelligent navigation through environmental perception, spatial modeling, and autonomous decision-making. Its core is to enable robots to understand physical space and plan optimal paths through the synergy of sensors, algorithms, and artificial intelligence. Existing spatial intelligence technologies typically rely on deep learning models to extract features and infer environmental data when understanding and reasoning about physical space. However, deep learning models often perform poorly for precise spatial computations, such as distance and position calculations. This leads to large errors in the calculation results, hindering the smooth execution of robotic tasks.

[0003] Therefore, how to overcome the insufficient performance of spatial intelligent machines in precise computing tasks and thus improve the accuracy and reliability of overall spatial intelligent reasoning has become an urgent problem to be solved. Summary of the Invention

[0004] In response to the above technical problems, the present invention provides a spatial intelligent reasoning method and system, which, through the combination of a deep learning model and a sandbox simulator, can improve calculation accuracy and enhance the accuracy and reliability of overall spatial intelligent reasoning.

[0005] According to a first aspect of the present invention, a spatial intelligent reasoning method is provided, comprising the following steps: S100, when receiving the robot's task instruction to be executed, generates several initial spatial reasoning tasks corresponding to the robot's task instruction to be executed according to the pre-built spatial intelligent machine; the several initial spatial reasoning tasks include spatial feature recognition tasks, entity state prediction tasks and semantic understanding tasks.

[0006] S200, determining a first category of reasoning tasks and a second category of reasoning tasks from a number of initial spatial reasoning tasks through a deep learning model embedded in the spatial intelligent machine; the first category of reasoning tasks refers to initial spatial reasoning tasks whose calculation accuracy must meet preset accuracy requirements; the second category of reasoning tasks refers to any initial spatial reasoning task other than the first category of reasoning tasks.

[0007] S300, obtains the target spatial environment information corresponding to each first-category reasoning task and each second-category reasoning task through a spatial model pre-built based on a spatial intelligent machine, and extracts features of each target spatial environment information through the deep learning model to obtain structured data corresponding to each first-category reasoning task and each second-category reasoning task.

[0008] S400, the structured data corresponding to each first-class reasoning task are sent to a pre-configured sandbox simulator respectively, so that the target computing program corresponding to each first-class reasoning task in the sandbox simulator is calculated according to the structured data corresponding to each first-class reasoning task respectively, and the calculated first prediction result is fed back to the spatial intelligent machine.

[0009] S500, generating a target space reasoning result corresponding to the task instruction to be executed by the robot based on the plurality of first prediction results and a second prediction result generated by the spatial intelligent machine based on the structured data corresponding to each second type of reasoning task.

[0010] According to a second aspect of the present invention, there is further provided a spatial intelligent reasoning system, the system comprising: The task generation module is used to generate several initial spatial reasoning tasks corresponding to the robot's task instructions when receiving the robot's task instructions to be executed based on a pre-built spatial intelligent machine; the several initial spatial reasoning tasks include spatial feature recognition tasks, entity state prediction tasks and semantic understanding tasks.

[0011] The task classification module is used to determine the first category of reasoning tasks and the second category of reasoning tasks from several initial spatial reasoning tasks through the deep learning model embedded in the spatial intelligent machine; the first category of reasoning tasks refers to the initial spatial reasoning tasks whose calculation accuracy must meet the preset accuracy requirements; the second category of reasoning tasks refers to any initial spatial reasoning tasks other than the first category of reasoning tasks.

[0012] The data acquisition module is used to obtain the target spatial environment information corresponding to each first-category reasoning task and each second-category reasoning task through a spatial model pre-built based on a spatial intelligent machine, and to extract features of each target spatial environment information through the deep learning model to obtain structured data corresponding to each first-category reasoning task and each second-category reasoning task.

[0013] The first computing module is used to send the structured data corresponding to each first-class reasoning task to a pre-configured sandbox simulator, so that the target computing program corresponding to each first-class reasoning task in the sandbox simulator can perform calculations based on the structured data corresponding to each first-class reasoning task, and feed back the calculated first prediction results to the spatial intelligent machine.

[0014] The reasoning result generation module is used to generate a target space reasoning result corresponding to the task instruction to be executed by the robot based on several first prediction results and a second prediction result generated by the spatial intelligent machine according to the structured data corresponding to each second type of reasoning task.

[0015] The present invention has at least the following beneficial effects: The present invention discloses a spatial intelligent reasoning method. First, a number of initial spatial reasoning tasks corresponding to the robot's task instructions to be executed are generated according to a pre-built spatial intelligent machine, and first-category reasoning tasks and second-category reasoning tasks are determined therefrom. The tasks that require precise calculation are determined and subsequently sent to a sandbox simulator for processing to improve the calculation accuracy. Then, the target spatial environment information corresponding to the first-category reasoning tasks and the second-category reasoning tasks is obtained, and feature extraction is performed through a deep learning model to obtain structured data corresponding to the first-category reasoning tasks and the second-category reasoning tasks, respectively. The target calculation program corresponding to each first-category reasoning task in the sandbox simulator is used to perform calculations according to the structured data corresponding to each first-category reasoning task, to obtain accurate spatial prediction results. The present invention introduces a sandbox simulator and, through the reasonable combination of the deep learning model and the sandbox simulator, can improve the calculation accuracy and enhance the accuracy and reliability of the overall spatial intelligent reasoning. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 A flowchart of a spatial intelligent reasoning method provided in Example 1 of the present invention; Figure 2 This is a flowchart of step S100 provided in Example 1 of the present invention; Figure 3 A flowchart of a method for determining the first type of reasoning task provided in the first embodiment of the present invention; Figure 4 A flowchart of another determination method for the first type of reasoning task provided in the first embodiment of the present invention; Figure 5 A flowchart of a method for obtaining a first calculation result provided in the first embodiment of the present invention; Figure 6 Flowchart of step S400 provided in Example 1 of the present invention; Figure 7 This is a structural diagram of a spatial intelligent reasoning system provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] Example 1 The first embodiment of the present invention provides a spatial intelligent reasoning method, such as Figure 1 As shown, the method includes the following steps: S100: Upon receiving a task instruction to be performed by the robot, the robot generates several initial spatial reasoning tasks corresponding to the task instruction based on a pre-built spatial intelligence machine. These initial spatial reasoning tasks include spatial feature recognition tasks, entity state prediction tasks, and semantic understanding tasks. For example, the spatial feature recognition task may involve identifying the features of several objects within the spatial intelligence machine's monitoring range; the entity state prediction task may involve predicting the motion state of an entity within the current time period based on its historical motion state; and the semantic understanding task may involve performing semantic understanding based on a combination of signs and recognized entities within the monitoring range.

[0020] In a specific embodiment, Figure 2 As shown, step S100 includes the following steps: S101: Upon receiving a robot's task instruction, each spatial intelligent machine acquires environmental data corresponding to the spatial model constructed by each spatial intelligent machine. In a specific implementation, the spatial intelligent machine is capable of constructing a local spatial model and acquiring environmental data within the local space. These constructing and acquiring functions are inherent to the spatial intelligent machine and will not be further described here.

[0021] S102, through the deep learning model embedded in each spatial intelligent machine, semantic analysis is performed on the robot's task instructions to be executed and feature extraction is performed on each environmental data information, thereby generating several spatial reasoning tasks corresponding to the robot's task instructions to be executed.

[0022] Based on the semantic analysis results and extracted environmental features, the deep learning model generates several spatial reasoning tasks, including but not limited to distance calculation, position relationship judgment, angle calculation, buffer generation, spatial overlap analysis, spatial union, spatial difference, spatial network analysis, spatial accessibility analysis, and spatial view analysis.

[0023] S200, determines the first category of reasoning tasks and the second category of reasoning tasks from a number of initial spatial reasoning tasks through the deep learning model embedded in the spatial intelligent machine; the first category of reasoning tasks refers to the initial spatial reasoning tasks whose calculation accuracy must meet the preset accuracy requirements; the second category of reasoning tasks refers to any initial spatial reasoning tasks other than the first category of reasoning tasks; it can be understood that: the first category of reasoning tasks refers to tasks that require precise calculation to obtain precise results.

[0024] In a specific embodiment, Figure 3 As shown, the initial spatial reasoning task whose accuracy must meet the preset accuracy requirements is determined by the following steps: S201, according to the reasoning task description text corresponding to each initial spatial reasoning task, perform semantic analysis on each reasoning task description text based on the deep learning model to obtain the semantic analysis text and the corresponding task scenario label corresponding to each initial spatial reasoning task.

[0025] At step S202, for any task scenario tag, if the task scenario tag matches any preset scenario tag in a pre-built scenario tag library, the initial spatial reasoning task corresponding to the task scenario tag is determined to be an initial spatial reasoning task whose accuracy must meet the preset accuracy requirements. For example, if the task scenario tag is for calculating the distance between entities, the initial spatial reasoning task is determined to be a first-class reasoning task.

[0026] S203: Keyword extraction is performed on each semantically analyzed text to obtain a number of task keywords. If any of the extracted task keywords matches any of the pre-set keywords in the pre-built keyword library, the initial spatial reasoning task corresponding to the semantically analyzed text is determined to be an initial spatial reasoning task whose accuracy must meet the pre-set accuracy requirements. For example, if the task keywords are distance calculation, position determination, or posture prediction, the initial spatial reasoning task is determined to be a first-category reasoning task.

[0027] Furthermore, if Figure 4 As shown, the method further includes the following steps: S210, for any initial spatial reasoning task, when the task scene label corresponding to the initial spatial reasoning task does not hit any preset scene label in the scene label library, and the task keyword in the semantic analysis text corresponding to the initial spatial reasoning task does not hit any preset keyword in the keyword library, obtain the first calculation result of the sandbox simulator on the structured data corresponding to the initial spatial reasoning task and the second calculation result of the deep learning model on the structured data corresponding to the initial spatial reasoning task.

[0028] In a specific embodiment, Figure 5 As shown, the first calculation result of the sandbox simulator on the structured data corresponding to the initial spatial reasoning task is obtained through the following steps: S211, sending the initial spatial reasoning task to the background server.

[0029] S212 , when receiving the target computing program fed back by the background server for the initial spatial reasoning task, sending the received target computing program to the sandbox simulator.

[0030] S213 , in the sandbox simulator environment, using the received target computing program to compute the structured data corresponding to the initial spatial reasoning task to obtain a first computing result.

[0031] As mentioned above, when the sandbox simulator does not store the target computing program corresponding to a certain initial spatial reasoning task, the computing program in the sandbox simulator is expanded through the background server to improve the comprehensiveness of the computing program. At the same time, the computing results of the sandbox simulator can be compared with the computing results of the deep learning model to achieve the optimization effect of the model parameters.

[0032] At step S220, when the difference between the first calculation result and the second calculation result is less than a preset value, the initial spatial reasoning task is determined to be a second-category reasoning task. Those skilled in the art will set the corresponding preset value based on different tasks and actual needs, and will not be further described here.

[0033] At step S230, when the difference between the first calculation result and the second calculation result is not less than a preset value, the initial spatial reasoning task is determined to be a first-category reasoning task, and the first calculation result is determined to be a target calculation result, thereby optimizing the model parameters of the deep learning model using the target calculation result as the actual result. Those skilled in the art will select model parameters for optimization based on actual needs. Since the optimization process is not a novel feature of the present invention, it will not be further described here.

[0034] In the above, the initial spatial reasoning tasks are first classified according to the two-dimensional analysis of task scenario labels and task keywords to obtain classification results. For the initial spatial reasoning tasks that cannot obtain classification results, spatial reasoning is further performed on the initial spatial reasoning tasks according to the deep learning model and the sandbox simulator respectively. The task types are comprehensively analyzed based on the two prediction results obtained, which can improve the accuracy and reliability of the division of reasoning tasks of different categories, and can also optimize the deep learning model according to the calculation results of the sandbox simulator to improve the performance of the model and subsequent prediction accuracy.

[0035] Specifically, the first category of reasoning tasks includes but is not limited to entity distance calculation, entity position determination, path planning simulation, motion collision risk assessment, personnel trajectory prediction, stacking stability assessment, robot grasping position prediction and lighting and visual occlusion relationship simulation; it can be understood that: these reasoning tasks are the refinement of the initial spatial reasoning task in step S100.

[0036] S300, obtains the target spatial environment information corresponding to each first-category reasoning task and each second-category reasoning task through a spatial model pre-built based on a spatial intelligent machine, and extracts features of each target spatial environment information through the deep learning model to obtain structured data corresponding to each first-category reasoning task and each second-category reasoning task.

[0037] S400, the structured data corresponding to each first-class reasoning task are sent to a pre-configured sandbox simulator respectively, so that the target computing program corresponding to each first-class reasoning task in the sandbox simulator is calculated according to the structured data corresponding to each first-class reasoning task respectively, and the calculated first prediction result is fed back to the spatial intelligent machine.

[0038] In a specific embodiment, Figure 6 As shown, step S400 also includes the following steps: S401, for any first-category reasoning task, when the sandbox simulator receives structured data corresponding to the first-category reasoning task, it builds a physical simulation world corresponding to the structured data in real time according to the pre-configured structured scene information; wherein, the sandbox simulator obtains the changing state of the entity in the physical simulation world in real time, and updates the structured data in real time according to the changing state of the entity, so as to optimize the construction of the physical model world in real time.

[0039] S402: extracting the spatial relationship of a plurality of entities in the physical simulation world according to the first type of reasoning task target calculation program, and obtaining a first prediction result according to the spatial relationship of the plurality of entities.

[0040] For example, in a computing scenario, when it is necessary to calculate the distance between two people on both sides of a corner, the spatial intelligent machine can only calculate the straight-line distance between the two people based on the monitored image, resulting in errors in the distance calculation. For such tasks that require precise calculation, the corresponding spatial data is sent to the sandbox simulator. The spatial data includes the angle information, person location information and other data monitored by the spatial intelligent machine based on the constructed spatial model. Based on the spatial data, a physical simulation world is built in the sandbox simulator to calculate the accurate person distance as the first prediction result.

[0041] Furthermore, the deep learning model in the spatial intelligent machine and the sandbox simulator use structured entity description code and API interface to exchange information; wherein, the interactive information includes the location and attribute information of the object, the entity information of preset specific categories of items, spatial structure data and historical trajectory data.

[0042] As mentioned above, since spatial intelligence technology usually relies on deep learning models to extract features and reason about environmental data, and deep learning models cannot obtain calculation results with sufficient accuracy for tasks that require precise calculations, this application introduces a sandbox simulator, which can send the environmental data monitored by the spatial model constructed by the spatial intelligence machine and the environmental data information extracted by the deep learning model to the sandbox simulator and perform real-time simulation, making the calculation results more accurate and enhancing the accuracy and reliability of the overall spatial intelligence reasoning.

[0043] S500, based on several first prediction results and the second prediction result generated by the spatial intelligent machine according to the structured data corresponding to each second type of reasoning task, generates the target space reasoning result corresponding to the task instruction to be executed by the robot; it can be understood as: combining the first prediction result and the second prediction result to obtain the target space reasoning result of the total task.

[0044] Furthermore, the method further comprises the following steps: S10, sending the target space reasoning result to the target robot, and obtaining the execution data of the target robot when performing the task according to each first reasoning task; S20, sending the acquired execution data to the sandbox simulator, so that the sandbox simulator guides the target robot to perform subsequent execution tasks according to the execution data.

[0045] In summary, the present invention discloses a spatial intelligent reasoning method. First, a number of initial spatial reasoning tasks corresponding to the robot's task instructions to be executed are generated according to a pre-built spatial intelligent machine, and the first type of reasoning tasks and the second type of reasoning tasks are determined therefrom. By determining the tasks that require precise calculation, they are subsequently sent to a sandbox simulator for processing to improve the calculation accuracy. Then, the target spatial environment information corresponding to the first type of reasoning tasks and the second type of reasoning tasks is obtained, and feature extraction is performed through a deep learning model to obtain structured data corresponding to the first type of reasoning tasks and the second type of reasoning tasks, respectively. The target calculation program corresponding to each first type of reasoning task in the sandbox simulator is used to perform calculations according to the structured data corresponding to each first type of reasoning task, and accurate spatial prediction results are obtained. The present invention introduces a sandbox simulator and through the reasonable combination of a deep learning model and a sandbox simulator, it can improve the calculation accuracy and enhance the accuracy and reliability of the overall spatial intelligent reasoning.

[0046] Example 2 The second embodiment of the present invention provides a spatial intelligent reasoning system, such as Figure 7 As shown, the system includes: The task generation module 100 is used to generate several initial spatial reasoning tasks corresponding to the robot's task instructions when receiving the robot's task instructions to be performed based on a pre-built spatial intelligent machine; the several initial spatial reasoning tasks include spatial feature recognition tasks, entity state prediction tasks and semantic understanding tasks.

[0047] In a specific embodiment, the task generation module 100 includes: The first acquisition module is used to acquire the environmental data information corresponding to each spatial intelligent machine through the spatial model constructed by each spatial intelligent machine when receiving the task instruction to be executed by the robot.

[0048] The generation module is used to perform semantic analysis on the robot's task instructions and feature extraction on each piece of environmental data information through the deep learning model embedded in each spatial intelligent machine, and generate several spatial reasoning tasks corresponding to the robot's task instructions.

[0049] The task classification module 200 is used to determine the first category of reasoning tasks and the second category of reasoning tasks from a number of initial spatial reasoning tasks through the deep learning model embedded in the spatial intelligent machine; the first category of reasoning tasks refers to the initial spatial reasoning tasks whose calculation accuracy must meet the preset accuracy requirements; the second category of reasoning tasks refers to any initial spatial reasoning tasks other than the first category of reasoning tasks.

[0050] In a specific embodiment, the task classification module 200 includes: The analysis module is used to perform semantic analysis on the reasoning task description text corresponding to each initial spatial reasoning task based on the deep learning model, and obtain the semantic analysis text and corresponding task scenario label corresponding to each initial spatial reasoning task.

[0051] The first judgment module is used to judge the initial spatial reasoning task corresponding to any task scene label as an initial spatial reasoning task whose accuracy must meet the preset accuracy requirements when the task scene label hits any preset scene label in the pre-built scene label library.

[0052] The second judgment module is used to extract keywords from each semantic analysis text to obtain several task keywords. When any extracted task keyword hits any preset keyword in the pre-built keyword library, the initial spatial reasoning task corresponding to the semantic analysis text itself is judged as an initial spatial reasoning task whose accuracy must meet the preset accuracy requirements.

[0053] In an extended embodiment, the task classification module 200 further includes: The second acquisition module is used to obtain, for any initial spatial reasoning task, the first calculation result of the sandbox simulator on the structured data corresponding to the initial spatial reasoning task and the second calculation result of the deep learning model on the structured data corresponding to the initial spatial reasoning task when the task scene label corresponding to the initial spatial reasoning task does not hit any preset scene label in the scene label library, and the task keyword in the semantic analysis text corresponding to the initial spatial reasoning task does not hit any preset keyword in the keyword library.

[0054] Specifically, the first calculation result of the sandbox simulator on the structured data corresponding to the initial spatial reasoning task is obtained through the following steps: The first sending module is used to send the initial spatial reasoning task to the background server.

[0055] The second sending module is configured to send the received target computing program to the sandbox simulator when receiving the target computing program fed back by the background server for the initial spatial reasoning task.

[0056] The second computing module is used to calculate the structured data corresponding to the initial spatial reasoning task using the received target computing program in the sandbox simulator environment to obtain a first computing result.

[0057] The third determination module is configured to determine the initial spatial reasoning task as a second type of reasoning task when the difference between the first calculation result and the second calculation result is less than a preset value.

[0058] The fourth judgment module is used to judge the initial spatial reasoning task as a first-category reasoning task when the difference between the first calculation result and the second calculation result is not less than a preset value, and to determine the first calculation result as the target calculation result, so as to optimize the model parameters of the deep learning model using the target calculation result as the true result.

[0059] Furthermore, the first type of reasoning tasks includes but is not limited to entity distance calculation, entity position determination, path planning simulation, motion collision risk assessment, personnel trajectory prediction, stacking stability assessment, robot grasping position prediction and lighting and visual occlusion relationship simulation.

[0060] The data acquisition module 300 is used to obtain the target spatial environment information corresponding to each first-class reasoning task and each second-class reasoning task through a spatial model pre-built based on a spatial intelligent machine, and to extract features of each target spatial environment information through the deep learning model to obtain structured data corresponding to each first-class reasoning task and each second-class reasoning task.

[0061] The first computing module 400 is used to send the structured data corresponding to each first-class reasoning task to a pre-configured sandbox simulator, so that the target computing program corresponding to each first-class reasoning task in the sandbox simulator performs calculations based on the structured data corresponding to each first-class reasoning task, and feeds back the calculated first prediction results to the spatial intelligent machine.

[0062] In a specific embodiment, the first calculation module 400 includes: A construction module is used for any first-class reasoning task. When the sandbox simulator receives structured data corresponding to the first-class reasoning task, it builds a physical simulation world corresponding to the structured data in real time according to pre-configured structured scene information; wherein, the sandbox simulator obtains the changing state of the entity in the physical simulation world in real time, and updates the structured data in real time according to the changing state of the entity, so as to optimize the construction of the physical model world in real time.

[0063] The third calculation module is used to extract the spatial relationship of multiple entities in the physical simulation world according to the first type of reasoning task target calculation program, and obtain a first prediction result based on the spatial relationship of the multiple entities.

[0064] Furthermore, the deep learning model in the spatial intelligent machine and the sandbox simulator use structured entity description code and API interface to exchange information; wherein, the interactive information includes the location and attribute information of the object, the entity information of preset specific categories of items, spatial structure data and historical trajectory data.

[0065] The reasoning result generation module 500 is used to generate a target space reasoning result corresponding to the task instruction to be executed by the robot based on several first prediction results and a second prediction result generated by the spatial intelligent machine based on the structured data corresponding to each second type of reasoning task.

[0066] Furthermore, the system further comprises: The third acquisition module is used to send the target space reasoning result to the target robot and obtain the execution data of the target robot when performing the task according to each first reasoning task.

[0067] The third sending module is used to send the acquired execution data to the sandbox simulator, so that the sandbox simulator guides the target robot to perform subsequent execution tasks according to the execution data.

[0068] It should be noted that the information interaction, execution process and other contents between the above modules are based on the same concept as the embodiment of the method of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0069] Although some specific embodiments of the present invention have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present invention. It should also be understood by those skilled in the art that various modifications may be made to the embodiments without departing from the scope and spirit of the present invention. The scope of the present invention is defined by the appended claims.

Claims

1. A spatial intelligent reasoning method, characterized in that: The method comprises the following steps: S100, upon receiving a task instruction to be performed by the robot, generating, based on a pre-built spatial intelligence machine, a number of initial spatial reasoning tasks corresponding to the task instruction to be performed by the robot; the number of initial spatial reasoning tasks includes a spatial feature recognition task, an entity state prediction task, and a semantic understanding task; S200, determining a first category of reasoning tasks and a second category of reasoning tasks from a plurality of initial spatial reasoning tasks using a deep learning model embedded in the spatial intelligent machine; the first category of reasoning tasks being initial spatial reasoning tasks whose computational accuracy must meet preset accuracy requirements; the second category of reasoning tasks being any initial spatial reasoning tasks other than the first category of reasoning tasks; S300, obtaining target spatial environment information corresponding to each first-category reasoning task and each second-category reasoning task using a spatial model pre-built based on the spatial intelligent machine, and performing feature extraction on each target spatial environment information using the deep learning model to obtain structured data corresponding to each first-category reasoning task and each second-category reasoning task, respectively; S400, sending the structured data corresponding to each first-category reasoning task to a pre-configured sandbox simulator, so that a target computing program corresponding to each first-category reasoning task in the sandbox simulator performs calculations based on the structured data corresponding to each first-category reasoning task, and feeding back the calculated first prediction results to the spatial intelligent machine; S500, generating a target space reasoning result corresponding to the task instruction to be executed by the robot based on the plurality of first prediction results and a second prediction result generated by the spatial intelligent machine based on the structured data corresponding to each second type of reasoning task.

2. The spatial intelligent reasoning method according to claim 1, characterized in that: Step S100 includes the following steps: S101, when receiving a task instruction to be executed by the robot, obtaining environmental data information corresponding to each spatial intelligent machine through the spatial model constructed by each spatial intelligent machine; S102, through the deep learning model embedded in each spatial intelligent machine, semantic analysis is performed on the robot's task instructions to be executed and feature extraction is performed on each environmental data information, thereby generating several spatial reasoning tasks corresponding to the robot's task instructions to be executed.

3. The spatial intelligent reasoning method according to claim 1, characterized in that: The initial spatial reasoning task whose accuracy must meet the preset accuracy requirements is determined by the following steps: S201, performing semantic analysis on each reasoning task description text corresponding to each initial spatial reasoning task based on a deep learning model to obtain a semantic analysis text and a corresponding task scenario label corresponding to each initial spatial reasoning task; S202: For any task scene label, when the task scene label hits any preset scene label in the pre-built scene label library, the initial spatial reasoning task corresponding to the task scene label itself is determined as an initial spatial reasoning task whose accuracy must meet the preset accuracy requirements; S203, keyword extraction is performed on each semantic analysis text to obtain several task keywords, and when any extracted task keyword hits any preset keyword in the pre-built keyword library, the initial spatial reasoning task corresponding to the semantic analysis text itself is determined as an initial spatial reasoning task whose accuracy must meet the preset accuracy requirements.

4. The spatial intelligent reasoning method according to claim 3, characterized in that: The method further comprises the steps of: S210, for any initial spatial reasoning task, when the task scene label corresponding to the initial spatial reasoning task does not hit any preset scene label in the scene label library, and the task keyword in the semantic analysis text corresponding to the initial spatial reasoning task does not hit any preset keyword in the keyword library, obtaining a first calculation result of the sandbox simulator on the structured data corresponding to the initial spatial reasoning task and a second calculation result of the deep learning model on the structured data corresponding to the initial spatial reasoning task; S220, when the difference between the first calculation result and the second calculation result is less than a preset value, determining the initial spatial reasoning task as a second type of reasoning task; S230: When the difference between the first calculation result and the second calculation result is not less than a preset value, the initial spatial reasoning task is determined to be a first-class reasoning task, and the first calculation result is determined to be a target calculation result, so as to optimize the model parameters of the deep learning model using the target calculation result as the true result.

5. The spatial intelligent reasoning method according to claim 4, characterized in that: Obtain the first calculation result of the sandbox simulator on the structured data corresponding to the initial spatial reasoning task through the following steps: S211, sending the initial spatial reasoning task to the backend server; S212, when receiving the target computing program fed back by the backend server for the initial spatial reasoning task, sending the received target computing program to the sandbox simulator; S213 , in the sandbox simulator environment, using the received target computing program to compute the structured data corresponding to the initial spatial reasoning task to obtain a first computing result.

6. The spatial intelligent reasoning method according to claim 1, characterized in that: The first category of reasoning tasks includes distance calculation between entities, entity position determination, path planning simulation, motion collision risk assessment, personnel trajectory prediction, stacking stability assessment, robot grasping position prediction and simulation of the relationship between lighting and visual occlusion.

7. The spatial intelligent reasoning method according to claim 1, characterized in that: Step S400 also includes the following steps: S401: For any first-category reasoning task, when the sandbox simulator receives structured data corresponding to the first-category reasoning task, it builds a physical simulation world corresponding to the structured data in real time based on pre-configured structured scenario information; wherein the sandbox simulator obtains the changing state of entities in the physical simulation world in real time, and updates the structured data in real time based on the changing state of the entities, so as to optimize the construction of the physical model world in real time; S402: extracting the spatial relationship of a plurality of entities in the physical simulation world according to the first type of reasoning task target calculation program, and obtaining a first prediction result according to the spatial relationship of the plurality of entities.

8. The spatial intelligent reasoning method according to claim 1, characterized in that: The method further comprises the steps of: S10, sending the target space reasoning result to the target robot, and obtaining the execution data of the target robot when performing the task according to each first reasoning task; S20, sending the acquired execution data to the sandbox simulator, so that the sandbox simulator guides the target robot to perform subsequent execution tasks according to the execution data.

9. The spatial intelligent reasoning method according to claim 1, characterized in that: The deep learning model in the spatial intelligent machine and the sandbox simulator use structured entity description code and API interface to exchange information; The interactive information includes the location and attribute information of the object, the entity information of preset specific categories of items, spatial structure data and historical trajectory data.

10. A spatial intelligent reasoning system, characterized in that: The system comprises: A task generation module is configured to, upon receiving a task instruction to be performed by the robot, generate a number of initial spatial reasoning tasks corresponding to the task instruction to be performed by the robot based on a pre-built spatial intelligence machine; the initial spatial reasoning tasks include a spatial feature recognition task, an entity state prediction task, and a semantic understanding task; A task classification module is used to determine, from a number of initial spatial reasoning tasks, first-category reasoning tasks and second-category reasoning tasks using a deep learning model embedded in the spatial intelligence machine; the first-category reasoning tasks refer to initial spatial reasoning tasks whose computational accuracy must meet preset accuracy requirements; the second-category reasoning tasks refer to any initial spatial reasoning tasks other than the first-category reasoning tasks; a data acquisition module, configured to obtain target spatial environment information corresponding to each first-category reasoning task and each second-category reasoning task using a spatial model pre-built based on the spatial intelligent machine, and to perform feature extraction on each target spatial environment information using the deep learning model to obtain structured data corresponding to each first-category reasoning task and each second-category reasoning task; a first computing module, configured to send the structured data corresponding to each first-category reasoning task to a pre-configured sandbox simulator, so that a target computing program corresponding to each first-category reasoning task in the sandbox simulator performs calculations based on the structured data corresponding to each first-category reasoning task, and feeds back a calculated first prediction result to the spatial intelligent machine; The reasoning result generation module is used to generate a target space reasoning result corresponding to the task instruction to be executed by the robot based on several first prediction results and a second prediction result generated by the spatial intelligent machine according to the structured data corresponding to each second type of reasoning task.

Images