System for controlling robot task decisions based on semantic networks and knowledge bases

Abstract

The application discloses a system for controlling robot task decision based on a semantic network and a knowledge base, comprising a semantic network module for providing a semantic network for constructing a task decision behavior tree, the semantic network being supplemented and expanded through the question and answer results of a large language model and a knowledge base; a large language model module for outputting question and answer results through knowledge question and answer using the large language model; a knowledge base module for providing knowledge for constructing the task decision behavior tree; a task decision module for constructing the task decision behavior tree according to the semantic network module, the large language model module and the knowledge base module, specifically constructing nodes of a decision tree by querying the semantic network, filling node parameters through the question and answer results of the large language model, and filling node parameters through the knowledge base; and a variable storage module for storing temporary variables during the construction of the task decision behavior tree. In this way, the decision behavior tree is more accurate.

Description

Technical Field

[0001] This invention belongs to the field of robotics and artificial intelligence technology, and in particular relates to a system for controlling robot task decision-making based on semantic networks and knowledge bases. Background Technology

[0002] In existing technologies, robot control relies more on fixed programmed scripts, knowledge symbols, or newly developed large models. From a technical application perspective, programmed scripts and knowledge symbols cannot adapt to changes in the environment. Large language models can handle robot task control in dynamic scenarios, but they cannot make decisions quickly and efficiently, and the training cost is huge.

[0003] Traditional knowledge bases only store knowledge about objects, people, common sense, and actions. Pre-defined robot planning methods are then used for planning. In this process, knowledge forms isolated nodes, and parameters and attributes are only synchronized and injected when the robot performs its tasks. Furthermore, the limited size of the knowledge base restricts the robot's autonomous decision-making and control capabilities, making its operational behavior overly programmed and rigid.

[0004] Meanwhile, the robot task planning output through the pre-built model is a series of behavior sequences. This results in a large number of redundant actions and behaviors when constructing complex robot tasks, and these behaviors and tasks are not fully covered during runtime.

[0005] Therefore, there is an urgent need to improve and innovate on the basis of knowledge base, graph computing, and large language model, and to design a new scheme based on the big data semantic network foundation of robot operation to apply semantic network to robot behavior decision-making.

[0006] Patent document CN114153943A discloses a system and method for constructing a robot behavior tree based on a knowledge graph. The system includes: a knowledge graph and a behavior tree. The knowledge graph is constructed according to the AOG format through a set of directed action nodes. Each action node includes an action category and its corresponding set of action features and the object to which the action applies. The directed connections connect the action nodes based on the action features. The behavior tree, based on the subject-verb-object structure of the task, takes the subject and object as the object of the action, classifies the predicate into a set of action features corresponding to the action category, constructs the corresponding behavior based on the action features, constructs behavior nodes based on the behavior, and rearranges the order of the behavior nodes in the behavior tree according to the directed connections of the sub-action nodes. The method includes: S1, constructing an object knowledge base; S2, constructing an action knowledge base; S3, constructing a knowledge graph; S4, constructing a behavior tree.

[0007] The above technical solution is based on knowledge graphs to build behavior trees. However, many behaviors do not exist in knowledge graphs, which will cause the behavior tree construction to fail or be inaccurate. Summary of the Invention

[0008] In view of the above, the purpose of this invention is to provide a system for controlling robot task decision-making based on semantic networks and knowledge bases, which combines semantic networks, large language models and knowledge bases to generate decision behavior trees for robot tasks.

[0009] To achieve the above-mentioned objectives, this invention provides a system for controlling robot task decision-making based on semantic networks and knowledge bases, including a semantic network module, a large language model module, a knowledge base module, a task decision-making module, and a variable storage module;

[0010] The semantic network module is used to provide a semantic network for constructing a task decision behavior tree, which is supplemented and expanded by the question-answering results and knowledge base of the large language model;

[0011] The large language model module is used to perform knowledge-based question answering and output question answering results using the large language model.

[0012] The knowledge base module is used to provide knowledge for constructing task decision-making behavior trees;

[0013] The task decision module is used to construct a task decision behavior tree based on the semantic network module, the large language model module, and the knowledge base module. Specifically, it constructs the nodes of the decision tree by querying the semantic network, fills the node parameters by the question-and-answer results of the large language model, and fills the node parameters by the knowledge base.

[0014] The variable storage module is used to store temporary variables when constructing the task decision behavior tree.

[0015] Preferably, the semantic network is a directed graph structure that sequentially connects related semantic nodes. In-degree and directed connections are used to represent the weights and probabilities between semantic nodes. Each semantic node includes an identifier and a parameter list. The directed connection relationship includes a relationship type and relationship attributes. The relationship attributes include routing probability, execution order, and condition description. Semantic nodes include semantic nodes with single semantic meaning, semantic nodes with decorative information, semantic nodes with conditions, and semantic nodes that allow the target to be a set, to adapt to the construction of the task decision behavior tree.

[0016] Preferably, the large language model has pre-stored knowledge that can perform question-and-answer search based on the input question and output the question-and-answer results when constructing the extended semantic network and the task decision behavior tree. At the same time, it can perform a search based on the task decomposition question and provide multiple possible results for task planning. The question-and-answer results and possible results are connected to the semantic network as extended knowledge or added to the nodes of the task decision behavior tree as supplementary knowledge.

[0017] Preferably, the knowledge base is refined knowledge, including an action knowledge base and an object knowledge base. The action knowledge base includes actions, the action type to which the action belongs, action characteristics, and the object to which the action acts. The object knowledge base includes objects and object attributes.

[0018] Preferably, the construction of the task decision behavior tree includes:

[0019] Based on the task understanding command and task classification results, determine the predicate and target object corresponding to the task;

[0020] Based on the target object, search the object knowledge base for the corresponding object instance of the target object as the target object instance for the action;

[0021] The predicates are categorized into action categories in the action knowledge base, and a set of behaviors is constructed based on a set of task semantics corresponding to the action categories.

[0022] Search for the set of semantic nodes corresponding to the set of behaviors in the semantic network, then construct a task decision behavior tree according to the set of semantic nodes, and add node parameters to the action nodes in the task decision behavior tree. The node parameters come from at least one of the target object instance searched from the object knowledge base, the dynamically perceived scene instance, and the question answering result obtained through multiple rounds of dialogue based on the large language model.

[0023] Preferably, the step of searching for a set of semantic nodes corresponding to a set of behaviors in a semantic network, and then constructing a task decision behavior tree according to the set of semantic nodes, includes:

[0024] The set of semantic nodes corresponding to a behavior includes a single semantic node, a semantic node with decorative information, a semantic node with conditions, and a semantic node whose target is a set.

[0025] When constructing a decision behavior tree based on the topology corresponding to the set of semantic nodes, if the semantic node is a single semantic node, an action node is generated and executed sequentially. If the semantic node is a semantic node with conditions, a branch is generated for the semantic node with conditions, specifically a selection node, and a condition node and the action node of the branch are connected under the selection node. If the semantic node is a semantic node with decoration information, a decoration node is generated for the semantic node with decoration information, and a sequence node and an action node are generated under the decoration node. If the semantic node is a semantic node that allows the target to be a set, multiple decoration nodes are generated for the semantic node with the target to be a set, and each decoration node is expanded and generated in the above manner.

[0026] The action nodes, sequence nodes, selection nodes, condition nodes, and decoration nodes generated based on the topology corresponding to the semantic node set constitute the task decision behavior tree.

[0027] Preferably, during the specific construction process, when a semantic node is accompanied by conditions, rule knowledge is queried, selection nodes are constructed based on the rule knowledge, and condition nodes are generated under the selection nodes based on the rule knowledge.

[0028] Preferably, when constructing the task decision behavior tree, the execution order of each sub-behavior in the task decision behavior tree is changed according to the execution order of the sub-action nodes.

[0029] Preferably, the process of constructing the task decision behavior tree also includes a task decision behavior tree verification process, with the verification conditions being:

[0030] Each sub-behavior under each behavior must correspond to either sequential nodes or selection nodes. Parallel nodes can coexist with sequential nodes. Selection nodes must contain condition nodes and Boolean parameters for condition node judgment. Any behavior must contain an action identifier parameter to complete the action.

[0031] Compared with the prior art, the beneficial effects of the present invention include at least the following:

[0032] This invention, based on action knowledge, introduces the design thinking of complex semantic networks, enabling dependencies between action knowledge points. Robot behavior can be constructed solely using semantic networks, ultimately generating a task decision behavior tree. Compared to language and script control sequences, behavior trees are superior in control logic structure. Behavior trees can represent both sequence and hierarchy between action behaviors, and identical behavior plans will not be repeated. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a schematic diagram of the system for controlling robot task decisions based on semantic networks and knowledge bases, provided in the embodiment.

[0035] Figure 2 This is a schematic diagram of the generation of action nodes from end semantic nodes provided in the embodiment;

[0036] Figure 3 This is a schematic diagram of the semantic node generation behavior tree based on the provided end-conditions;

[0037] Figure 4 This is a schematic diagram of the semantic node generation behavior tree with attached decoration information provided in the embodiment;

[0038] Figure 5 This is a schematic diagram illustrating the generation of a behavior tree from semantic nodes of a set, as provided in the embodiment.

[0039] Figure 6 This is the overall flowchart of the task decision behavior tree generation provided in the embodiment. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of this invention.

[0041] like Figure 1 As shown, the system 100 for controlling robot task decision based on semantic network and knowledge base provided in the embodiment includes a semantic network module 110, a large language model module 120, a knowledge base module 130, a task decision module, and a variable storage module 140.

[0042] Semantic network model 110 is used to provide a semantic network for constructing a task decision behavior tree. This semantic network is supplemented and expanded by the question-answering results and knowledge base of a large language model. Specifically, the semantic network corresponds to the task decomposition structure, establishing action nodes M based on the minimum controllable actions of the robot. Using task decomposition methods, a task is decomposed into multiple necessary sub-tasks, some of which involve conditional decisions and selection values.

[0043] The semantic network is a directed graph structure that sequentially connects semantic nodes. In-degree and directed connection relationships represent the weights and probabilities between semantic nodes. Each semantic node, when it is an action node, includes a unique action identifier and a parameter list `vars` for its target object. Directed connection relationships include relationship type and relationship attributes. Relationship attributes include routing probability (`prop`), execution order (`seq`), and condition description (`condition`). For the routing probability `prop` attribute, if a semantic node's child nodes lack execution order parameters, then `prop` will be used as the sorting criterion. When it is a condition relationship, `prop` determines the probability of the node being placed in the behavior tree, a method of random number falling within a certain range. The relationship type is the association attribute between semantic nodes, including dependency (`is`), action (`action`), state transition (`transform`), and spatial relationship (`space`). Regardless of the relationship type, all connections are constrained by ascending execution order (`seq`). The condition description (`condition`) indicates the existence of conditions in the decomposition relationship, requiring the creation of decorator nodes and condition nodes in the corresponding decision behavior tree to address this issue.

[0044] Directed links include two distinct states: feasible connections and necessary connections. When a possible connection state occurs, the routing probability (prop) represents the strength of the action node's choice. Semantic nodes include semantic nodes with a single semantic meaning, semantic nodes with decorative information, semantic nodes with conditions, and semantic nodes that allow the target to be a set, to adapt to the construction of the task decision behavior tree. A single semantic meaning contains various action semantics.

[0045] The large language model module 120 is used to perform knowledge-based question answering and output question-answering results using a large language model. The large language model stores knowledge that enables it to perform question-answering searches based on the input question and output question-answering results during the construction of the extended semantic network and task decision behavior tree. Simultaneously, it searches based on the task decomposition question, providing multiple possible results for task planning. The question-answering results and multiple possible results are connected to the semantic network as extended knowledge or added to the nodes of the task decision behavior tree as supplementary knowledge.

[0046] The knowledge base module 130 provides the knowledge for constructing the task decision behavior tree. This is refined knowledge, used to store knowledge of the real-time dynamic scene, including an action knowledge base and an object knowledge base. The object knowledge base includes objects in the scene and their attributes. These objects include person information {People} and environmental object information {Env}, both containing attribute information such as person name, title, coordinates, and characteristics; and object name, coordinates, characteristics, historical trajectory, and social posture. When constructing the decision behavior tree, based on robot information and task classification results, the corresponding target object is retrieved from the object knowledge base as the basis for filling the action action object parameter vars.

[0047] The motion knowledge base includes actions, their corresponding motion types (motion), action features, and the objects to which the actions are applied. Motion features refer to the robot hardware components required to execute the action and the necessary parameter list for performing the action. A type identifier is constructed for each action category ID, semantic associations are established for action features, and target objects are associated with the objects to which the actions are applied. A target object is an instance of an object created in the computer memory according to its data format in the object knowledge base, possessing a unique object identifier. A unique action ID is used to apply to the target object and the object parameter list (vars). The object parameter list (vars) includes location, target object, target person, language content, grasping spatial pose information, interaction targets, preprocessing functions, etc., as shown in Table 1.

[0048] Table 1

[0049]

[0050] The task decision module 140 is used to construct a task decision behavior tree based on the semantic network module, the large language model module, and the knowledge base module. Specifically, it constructs the nodes of the decision tree by querying the semantic network, fills the node parameters by the question-and-answer results of the large language model, and fills the node parameters by the knowledge base.

[0051] Specifically, combining the task and using a semantic network, a decision behavior tree is constructed through selected conditional rules. The calculation of these conditional rules is related to the robot's perceptual information and is performed in real-time. When a conditional rule is deemed executable, a multi-way branch behavior is constructed, obtaining a real-time conditional judgment result of boolean type obtained through synchronous condition-rule-execution features. When generating targets for the decision behavior tree, a target object is queried based on the subject corresponding to the behavior. Location information, motion category, description information, language information, task type, and target list are obtained and injected into the object parameter list of the action node. The motion category is injected as content into the behavior. The task type and a random number are used as a unique identifier ID for the decision behavior tree and cached in the database. For actions where the target parameter object cannot be determined at all, the behavior tree construction is deemed to have failed.

[0052] More specifically, such as Figure 6 As shown, constructing a task decision behavior tree includes: determining the predicate and target object corresponding to the task based on the task understanding command and task classification results; searching for the corresponding object instance in the object knowledge base as the target object instance for the action; classifying the predicate into action categories in the action knowledge base, and constructing a set of behaviors based on a set of task semantics corresponding to the action categories; searching for the set of semantic nodes corresponding to the set of behaviors in the semantic network, and then constructing a task decision behavior tree according to the set of semantic nodes, and adding node parameters to the action nodes in the task decision behavior tree, wherein the node parameters come from at least one of the target object instance searched from the object knowledge base, dynamically perceived scene instance, and question-answering results obtained through multiple rounds of dialogue based on a large language model.

[0053] When constructing a decision behavior tree based on the topology corresponding to the set of semantic nodes, such as Figure 2 As shown, when a semantic node is a single semantic node, corresponding action nodes are generated and executed sequentially; for example... Figure 3 As shown, when a semantic node is a semantic node with attached conditions, a branch is generated for the semantic node with attached conditions, specifically a selection node, and a condition node and the action node of the branch are connected under the selection node; for example... Figure 4 As shown, when a semantic node is a semantic node with attached decoration information, a decoration node is generated for the semantic node with attached decoration information, and a sequence node and an action node are generated under the decoration node; for example... Figure 5As shown, when a semantic node is a semantic node that allows a set of targets, multiple decoration nodes are generated for the semantic node that allows a set of targets. Each decoration node is expanded and generated in the manner described above. The action nodes, sequence nodes, selection nodes, condition nodes, and decoration nodes generated based on the topology structure corresponding to the set of semantic nodes constitute the task decision behavior tree.

[0054] Specifically, the semantic network searches for a set of semantic nodes corresponding to a set of behaviors. This set of semantic nodes represents the formation of action nodes and sub-behavior sets, where sub-behaviors are composed of action nodes. For the set of semantic nodes, based on whether there are attached conditions in the relationship type of the semantic nodes, the action is split into action nodes of different behaviors, and selection nodes are added above the action nodes. When a task has decoration information, a decoration node is constructed and a decoration attribute is added. A selection node is added to the decoration node, and conditional nodes are placed below the selection nodes in sequence. Sequential nodes with conditions are also expanded according to this pattern. Both sequential nodes and conditional nodes must be processed according to this pattern until a sub-behavior tree is constructed. During construction, the execution order of each sub-behavior in the task decision behavior tree is rearranged according to the execution order of the sub-action nodes. If, during the construction of the robot behavior tree, there is a failure to acquire feature data in the scene, a node type that is not within the range, or an empty node when generating sequential nodes and selection nodes, then the behavior tree construction process is considered to have failed.

[0055] In this example, the process of constructing a behavior tree based on the semantic network is as follows: Robot data is found in the robot database based on task trigger data; object entities are found in the object or dynamic scene library as the action objects; construction fails when no object entity exists in the scene. The task is classified into a specific action by a classifier. This action corresponds to the feature list of action features in the action knowledge base, i.e., the action features in action node M. The parameter list vars of the action object is populated with specific parameters obtained from the scene information and the dest information in the supplementary information. The language content is obtained from the corpus of the corresponding target in the knowledge base. The corresponding behavior is constructed according to the action features. Based on the behavior features, the use of motion, action, or language dialogue is selected. The completed behavior is recorded as an action node in the behavior tree.

[0056] For an already constructed action node, a subgraph corresponding to the subtask is found in the semantic network, centered on the subtask. This subgraph is represented by all next-order nodes containing the action node. Based on whether the edges in the semantic network description contain conditional information, decoration information, or prop variables, they are mapped to three types of behavior tree logical nodes: selection nodes, sequential nodes, and parallel nodes (selection nodes are those where the semantic network relationship has conditional information, and sequential nodes are those otherwise. Nodes with the same sequence number within sequential nodes are grouped together as parallel nodes). For cases where a node is already a leaf node, a sequential node is constructed. The first node under the sequential node is recorded as the corresponding conditional node, and the rest are the corresponding action nodes. Nodes with the same sequence are designated as parallel nodes. For all {Behavior}sel nodes, selection behavior nodes are constructed, and during construction, a conditional judgment `cond(env)` is performed to obtain a boolean real-time conditional judgment result. For all {Behavior}seq nodes, sequential behavior nodes are constructed, and during construction, they are sorted, and the order of the child nodes of the behavior tree is rearranged from largest to smallest according to the seq sorting method.

[0057] When injecting parameters into the decision behavior tree, the following method is used for any node: The subject always uses the subject s from the task understanding. Based on the subject s, the environment and target are queried to obtain location information, the target object's ID (i.e., the action category motion), and description information, which are then injected into the target parameter list of the action node. The action ID is used as content injection into the behavior, and the predicate v from the task understanding information is used as the behavior tree name. For behaviors whose parameters cannot be determined, a knowledge base search is performed to find actions with similar features for processing. For actions whose object parameters cannot be determined at all, the behavior cannot be constructed, and the entire behavior tree construction is deemed to have failed.

[0058] In this embodiment, a task decision behavior tree verification process is also performed, and the verification conditions are:

[0059] Each sub-behavior under each behavior must correspond to either sequential nodes or selection nodes. Parallel nodes can coexist with sequential nodes. Selection nodes must contain condition nodes and Boolean parameters for condition node judgment. Any behavior must contain an action identifier parameter to complete the action.

[0060] The variable storage module 150 is used to store temporary variables when constructing the task decision behavior tree. These temporary variables include scene information read during dynamic perception, question-and-answer results from multi-turn dialogue solving by the large language model, and knowledge read from the knowledge base.

[0061] The specific embodiments described above illustrate the technical solution and beneficial effects of the present invention in detail. It should be understood that the above description is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, additions, and equivalent substitutions made within the scope of the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A system for controlling robot task decision-making based on semantic networks and knowledge bases, characterized in that, It includes a semantic network module, a large language model module, a knowledge base module, a task decision module, and a variable storage module; The semantic network module is used to provide a semantic network for constructing a task decision behavior tree, which is supplemented and expanded by the question-answering results and knowledge base of the large language model; The large language model module is used to perform knowledge-based question answering and output question answering results using the large language model. The knowledge base module is used to provide knowledge for constructing task decision-making behavior trees; The task decision module is used to construct a task decision behavior tree based on the semantic network module, the large language model module, and the knowledge base module. Specifically, it constructs the nodes of the decision tree by querying the semantic network, fills the node parameters by the question-and-answer results of the large language model, and fills the node parameters by the knowledge base. The variable storage module is used to store temporary variables when constructing the task decision behavior tree; The construction of the task decision behavior tree includes: Based on the task understanding command and task classification results, determine the predicate and target object corresponding to the task; based on the target object, search for the corresponding object instance in the object knowledge base as the target object instance for the action; The predicate is classified into action categories in the action knowledge base, and a set of actions is constructed based on a set of task semantics corresponding to the action categories. The set of semantic nodes corresponding to the action set is searched in the semantic network, and then a task decision action tree is constructed according to the set of semantic nodes. Node parameters are added to the action nodes in the task decision action tree, where the node parameters come from at least one of the target object instance searched from the object knowledge base, the dynamically perceived scene instance, and the question-and-answer result obtained through multiple rounds of dialogue based on the large language model. This involves searching for a set of semantic nodes corresponding to a set of behaviors in the semantic network, and then constructing a task decision behavior tree based on the set of semantic nodes, including: The set of semantic nodes corresponding to a behavior includes a single semantic node, a semantic node with decorative information, a semantic node with conditions, and a semantic node whose target is a set. When constructing a decision behavior tree based on the topology corresponding to the set of semantic nodes, if the semantic node is a single semantic node, an action node is generated and executed sequentially. If the semantic node is a semantic node with conditions, a branch is generated for the semantic node with conditions, specifically a selection node, and a condition node and the action node of the branch are connected under the selection node. If the semantic node is a semantic node with decoration information, a decoration node is generated for the semantic node with decoration information, and a sequence node and an action node are generated under the decoration node. If the semantic node is a semantic node that allows the target to be a set, multiple decoration nodes are generated for the semantic node with the target to be a set, and a sequence node and an action node are generated for each decoration node. The action nodes, sequence nodes, selection nodes, condition nodes, and decoration nodes generated based on the topology corresponding to the semantic node set constitute the task decision behavior tree.

2. The system for controlling robot task decision-making based on semantic networks and knowledge bases according to claim 1, characterized in that, The semantic network is a directed graph structure that sequentially connects related semantic nodes. In-degree and directed connections are used to represent the weights and probabilities between semantic nodes. Each semantic node includes an identifier and a parameter list. The directed connection relationship includes the relationship type and relationship attributes. The relationship attributes include routing probability, execution order, and condition description. Semantic nodes include semantic nodes with single semantic meaning, semantic nodes with decoration information, semantic nodes with conditions, and semantic nodes that allow the target to be a set, to adapt to the construction of the task decision behavior tree.

3. The system for controlling robot task decision-making based on semantic networks and knowledge bases according to claim 2, characterized in that, Directed links include two different states: feasible links and necessary links. When a feasible link state occurs, the routing probability prop represents the selection strength of the action node.

4. The system for controlling robot task decision-making based on semantic networks and knowledge bases according to claim 1, characterized in that, The large language model's pre-stored knowledge enables it to perform question-and-answer search based on the input question and output the question-and-answer results when constructing the extended semantic network and task decision behavior tree. At the same time, it performs a search based on the task decomposition question and provides multiple possible results for task planning. The question-and-answer results and multiple possible results are connected to the semantic network as extended knowledge or added to the nodes of the task decision behavior tree as supplementary knowledge.

5. The system for controlling robot task decision-making based on semantic networks and knowledge bases according to claim 1, characterized in that, The knowledge base is refined knowledge, including an action knowledge base and an object knowledge base. The action knowledge base includes actions, the action type to which the action belongs, action characteristics, and the object on which the action acts. The object knowledge base includes objects and object attributes.

6. The system for controlling robot task decision-making based on semantic networks and knowledge bases according to claim 1, characterized in that, In the specific construction process, when a semantic node is accompanied by conditions, the rule knowledge is queried, a selection node is constructed based on the rule knowledge, and a condition node is generated under the selection node based on the rule knowledge.

7. The system for controlling robot task decision-making based on semantic networks and knowledge bases according to claim 1, characterized in that, When constructing the task decision behavior tree, the execution order of each sub-behavior in the task decision behavior tree is changed according to the execution order of the sub-action nodes.

8. The system for controlling robot task decision-making based on semantic networks and knowledge bases according to claim 1, characterized in that, It also includes a task decision behavior tree validation process, with the validation conditions being: Each sub-behavior under each behavior must correspond to either sequential nodes or selection nodes. Parallel nodes can coexist with sequential nodes. Selection nodes must contain condition nodes and Boolean parameters for condition node judgment. Any behavior must contain an action identifier parameter to complete the action.

Images