173 results about "Behavior Trees" patented technology

The invention provides a legal affair intelligent auxiliary processing method and a multi-agent system. The method comprises the following steps: constructing a dynamically updated legal affair knowledge system; according to the dynamically updated law knowledge system, constructing a case shunting agent Agent, a law application Agent, a class case retrieval Agent and a risk early warning Agent, and realizing a multi-Agent collaboration mechanism through a collaboration structure based on message transmission and state sharing; according to a multi-Agent cooperation mechanism, a preference mapping relation model is constructed through case complexity evaluation and quality evaluation, and multi-case relation perception preference mapping is realized; constructing a planning tree PLANTOR framework by combining a large language model and a knowledge management technology based on Prolog based on multi-case relationship perception preference mapping, generating a robot-oriented knowledge base and a planning program, and converting the robot-oriented knowledge base and the planning program into a behavior tree; according to the PLANTOR framework, multi-Agent inference chain alignment is achieved through joint optimization of transmission optimization and embedded optimization, and law decision assistance is completed. According to the scheme, the legal affair processing accuracy and efficiency are effectively improved.

A system for controlling robot task decision-making on the basis of a semantic network and a knowledge base, the system comprising: a semantic network module, which is used for providing a semantic network for constructing a task decision-making behavior tree, wherein the semantic network is supplemented and expanded by means of question-answering results of a large language model and a knowledge base; a large language model module, which is used for using the large language model to perform knowledge question-answering and output the question-answering results; a knowledge base module, which is used for providing knowledge for the construction of a task decision-making behavior tree; a task decision-making module, which is used for constructing the task decision-making behavior tree on the basis of the semantic network module, the large language model module and the knowledge base module, and specifically, constructing nodes of the decision-making tree by means of querying the semantic network, performing node parameter filling by means of the question-answering results of the large language model, and performing node parameter filling by means of the knowledge base; and a variable storage module, which is used for storing temporary variables of when the task decision-making behavior tree is constructed. On the basis of the system, a decision-making behavior tree is constructed more accurately.

The invention provides a hydraulic generator inner bore inspection robot based on a dynamic behavior tree and an intelligent decision and operation method thereof, and the robot comprises an upper mechanical arm end executor, a six-axis mechanical arm, an intermediate control system, a mobile platform, a multi-physical quantity fusion detection system, and a central control platform. The method is specially designed for the complex dynamic shielding environment of the inner bore of the hydro-generator. According to the method, by constructing a three-level behavior tree architecture composed of a task decision-making layer, a scene adaptation layer and an execution control layer, inspection requirements are decomposed into environment perception, task execution and resource constraint events, and the events are dynamically mapped to preset behavior nodes. Weight factors are adjusted in real time based on a multi-dimensional evaluation system, and functions of autonomous obstacle avoidance, narrow space operation and anomaly detection are realized in combination with hardware characteristics of the robot. The problems that a traditional inspection method is poor in adaptability and low in efficiency in a dynamic shielding environment are solved, and the intelligent level and safety of hydraulic generator inner bore inspection are remarkably improved.

The embodiment of the invention provides an industrial robot autonomous assembly task planning method based on a multi-agent large model, and belongs to the technical field of automatic assembly. The method comprises the steps of obtaining assembly task data of an assembly object, wherein the assembly task data comprises a task instruction for assembling the assembly object and an assembly manual; analyzing the task instruction and the assembly manual to obtain task execution logic; performing real-time acquisition on an assembly environment when the assembly object is assembled to obtain a target image; multi-modal sensing is conducted on the target image, a sensing result is obtained, and the sensing result comprises the position of at least one assembly of the assembly object in the assembly environment; and converting the task execution logic and the perception result into a planning domain definition language format, and performing assembly task planning through the behavior tree to obtain a target assembly task. According to the embodiment of the invention, the assembly efficiency and precision of the industrial robot are improved.

According to the agent construction method and the multi-agent cooperation method provided by the invention, the agent parameter configuration file is acquired, and the agent basic information, the ECA rule, the behavior tree path index and the performance function parameter configuration are loaded from the agent parameter configuration file; the knowledge hypergraph is adopted to aggregate agent basic information, ECA rules, behavior tree path indexes and performance function parameter configuration, and a dynamic knowledge network with semantic association is obtained; and after a prompt word used for calling the large language model is dynamically generated based on the dynamic knowledge network, the large language model is driven to generate an executable code, and when the executable code passes verification, an agent corresponding to the agent parameter configuration file is obtained. And then, aiming at a received user instruction, performing task processing by adopting the constructed multi-agent. Therefore, the autonomous decision-making capability and the dynamic adaptability of the intelligent agent are enhanced, and higher flexibility and expansibility are achieved.

The invention provides a tunnel drilling and blasting slag removal vehicle autonomous path planning method and device based on a Naligation 2 behavior tree scheduling framework, and the method comprises the following steps: S1, carrying out the data collection and map construction of a tunnel scene after drilling and blasting; s2, identifying a blasting residue area and generating a target point; s3, constructing a behavior tree scheduling structure and planning a path; and S4, performing path search, execution and feedback evaluation based on a path planning algorithm. According to the method, multi-source sensor mapping and deep learning semantic segmentation technologies are combined, high-precision recognition and traffic area extraction of residual obstacles in an unstructured tunnel environment after drilling and blasting are performed are performed, complex scenes faced by slag removal path planning are fully considered, a behavior tree scheduling mechanism is adopted to fuse path planning, target scheduling and an anomaly recovery strategy, and the path planning efficiency is improved. And a path execution system of real-time response and fault-tolerant capability is further constructed, and the operation continuity and robustness of the robot in a complex operation environment are improved.

The invention provides a multi-agent construction method and device and related equipment, and relates to the technical field of artificial intelligence, and the method comprises the steps: according to a first target instruction input by a first user, obtaining a plurality of sub-tasks corresponding to the first target instruction and a logic relation between the sub-tasks; obtaining at least one target agent from an agent capability library based on the first target instruction, wherein the target agent is an agent matched with the subtask in the agent capability library; determining a behavior tree structure according to the plurality of subtasks and the logical relationship, wherein the behavior tree structure is used for mapping a connection relationship between the at least one target agent; and adjusting a connection relationship between the at least one target agent based on the behavior tree structure to obtain a target multi-agent. The ability to understand complex business contexts and user intentions is enhanced, and accurate construction of a multi-agent system is realized.

Methods and apparatuses for conversation dialogue orchestration in virtual assistant communication sessions include a server that establishes a chat session between a virtual assistant (VA) application and a client device. The VA application captures an utterance generated by a user and processes the utterance to instantiate a dialogue behavior tree comprising workflow agents each associated with executable code for completing a corresponding workflow action. The VA application traverses the behavior tree to generate a response to the utterance, including evaluating one or more conditions associated with a workflow agent to determine whether to execute the code in the workflow agent, and when the conditions associated with the workflow agent are met, executing the code to complete the workflow action and storing a sub-response in a dialogue memory. The VA application coalesces the sub-responses to generate a final response and transmits the final response to the client device.

A method for generating a behaviour tree for controlling a robot device. The method includes: combining a plurality of predetermined behaviour trees and background knowledge into a behaviour tree knowledge graph, representing the behaviour tree knowledge graph in a latent space; extracting, from a prompt describing a desired behaviour of the robot device, a prompt representation graph; supplementing the prompt representation graph according to relations specified by the behaviour tree knowledge graph; selecting a sub graphs of the behaviour tree knowledge graph depending on a similarity to the supplemented prompt representation graph; and generating the behaviour tree for controlling the robot device by adjusting the selected sub graph according to knowledge from the prompt and the behaviour tree knowledge graph.

The invention discloses a behavior tree code automatic generation method based on a large language model. The method comprises the following steps: 1, acquiring a behavior tree model of an XML file format defined by a user and prompt information input by the user; 2, analyzing the obtained XML file, and providing a generation rule of a behavior tree code; step 3; the analyzed XML file and prompt information input by a user are combined with a behavior tree code generation rule, and a task specific prompt is generated; 4, combining the task specific prompt with the general code generation prompt to generate a complete prompt; 5, a proper LLM is selected, complete prompt information is input, and the LLM generates codes according to prompts; 6, running the generated code in a safe environment to ensure the safety of the code and the correctness of an output format; and 7, displaying an execution result on a visual interface, auditing the result by a user, and modifying a generated code. The method has the characteristics of high universality, high flexibility and high interpretability.

The invention provides a multi-agent black-box strategy analysis method and system based on a behavior tree, and relates to the technical field of agents, and the method comprises the steps: obtaining an expert black-box strategy model based on an observation data set through a multi-agent reinforcement learning algorithm and a rule mapping relation between learning environment observation data and execution actions; performing feature distillation on the environmental observation data, establishing a trajectory data set consisting of feature vectors after distillation and corresponding execution actions, and fitting a decision tree; recursively traversing nodes of the decision tree, generating a feature condition and an execution state of each node of the behavior tree, and obtaining a condition mapping set; performing black box strategy analysis on currently input environment observation data to obtain a final mapping path, a state sequence and an execution action, and introducing a decision tree fitting degree reward item into a reward function of reinforcement learning; according to the method, the advantages of the neural network, the decision tree and the behavior tree are integrated, and interpretable analysis of the black box model decision process is realized on the premise of keeping the model performance.

Disclosed are a target object persona construction method and system, and a storage medium, which are applied to the technical field of information, and can dynamically update a target object persona in real time and effectively improve the precision of the target object persona. The method comprises: analyzing target object usage behavior events, and constructing behavior event data, the behavior event data comprising a plurality of target object behavior events; on the basis of the behavior event data, creating a first behavior tree, the first behavior tree comprising a plurality of behavior nodes, and the behavior nodes corresponding to the target object behavior events; acquiring target object behavior data by means of a preset event-driven model, so as to update the first behavior tree on the basis of the target object behavior data and generate a second behavior tree; and constructing and obtaining a target object persona by means of analyzing the second behavior tree.

Various embodiments include a workflow generation method. An example includes: receiving a behavior tree construction operation entered on a graphical user interface (GUI) by a user, including an addition operation and a line connection operation performed on behavior tree nodes comprising function block nodes, one behavior tree is used for representing one workflow of service operations to be executed by resources in one workcell, and the function block nodes are used for implementing the service operations in the workflow; instantiating the behavior tree nodes and establishing connection relationships among the nodes in response to the construction operation, wherein some or all of the instantiated function block nodes are associated with resources for executing corresponding service operations; and generating a behavior tree corresponding to one workflow using the connection relationships among the tree nodes.

Various embodiments of the teachings herein include a dynamic workflow implementation method. An example includes: constructing a behavior tree comprising: a plurality of function block nodes, and connection relationships between the function block nodes; analyzing the behavior tree to obtain a workflow representing service operations to be performed by resources comprising the dynamic resource in one workcell; and deploying the workflow to a runtime of the corresponding workcell so the runtime determines a resource instance of the dynamic resource based on on-site resource data and controls resources comprising the resource instance to perform the operations based on the workflow. At least one of the plurality of function block nodes is bound to a dynamic resource including a placeholder for reference to a plurality of resource instances of a same type.

The invention relates to the technical field of artificial intelligence, and discloses a multi-agent training method and device based on a large language model, equipment and a medium, and the method comprises the steps: obtaining role behavior tendency information of agents in a multi-agent cluster; constructing a role behavior tree according to the role behavior tendency information through the first large language model; pre-training the intelligent agent according to the role behavior tree so as to initialize a role behavior strategy and obtain a pre-trained intelligent agent; obtaining a first soft role embedding vector according to the situation abstract through a first large language model, and generating a reward function according to the situation abstract and the first soft role embedding vector; acquiring first local observation information; and performing reinforcement learning training on the pre-trained agent according to the first local observation information, the first soft role embedding vector and the reward function so as to perform differentiation on the role behavior strategy. According to the method, the training efficiency can be improved, and the role division effect, the dynamic cooperation capability and the strategy interpretability of the multi-agent are enhanced.

Various embodiments of the teachings herein include workflow creation methods. An example includes: storing preset behavior tree nodes in a node library, the nodes including main control nodes for controlling a running state of a workflow and function block nodes operable for implementing operations of a resource in a workcell; displaying the library on a GUI; receiving construction of behavior a tree and an instantiation configuration operation performed by a user on the GUI based on the node library, the behavior tree representing a workflow used for defining operations to be executed by a workcell; and generating a behavior tree instance corresponding to a workflow in response to the construction of the behavior tree. Construction of the behavior tree based on the node library comprises: connecting the start node, the function block nodes, and the end node or the start node, the function block nodes, and the loop node sequentially.

The invention discloses a software performance test system, method and device in a high-concurrency scene, and relates to the field of software testing, the system comprises a scene intelligent construction module used for generating a concurrency scene through a behavior tree modeling and time sequence prediction algorithm according to service demand parameters and historical traffic data; the distributed test environment scheduling module is used for constructing a distributed test environment by adopting a container arrangement technology; the virtual user load generation module is used for simulating concurrent user requests by adopting a coroutine pool technology; the real-time performance data acquisition module is used for acquiring multi-source performance data of the tested software in real time; the intelligent performance analysis module is used for calculating a core performance index according to the multi-source performance data and carrying out anomaly detection and trend prediction; and the performance bottleneck positioning module is used for mining a dependency relationship among the core performance indexes and generating a bottleneck positioning report and an optimization suggestion document. According to the method, the accuracy, the efficiency and the stability of software performance testing in a high-concurrency scene are improved.

The invention relates to the technical field of robot autonomous decision-making, and discloses an underwater robot behavior decision-making method based on a behavior tree, which describes and controls the specific behavior of an underwater robot based on the behavior tree, organizes and manages complex behavior logic through a tree structure, and analyzes a specific task to determine the behavior of the underwater robot. And decomposing into behavior sub-trees, and combining control nodes and action nodes in the behavior sub-trees to complete a specific logic switching function. Compared with a traditional finite-state machine, the method has the advantages of modularization, expandability and high readability, the coupling degree between modules in the sub-tree is low, the sub-tree can be transplanted to multiple positions of the behavior tree, the problem that the workload of the reuse state of the finite-state machine is large is effectively solved, and the method has good practical application value.

The invention provides a robot control method in combination with a behavior tree framework. The method comprises the steps that S1, an instruction is received; s2, analyzing an instruction through a pre-trained large language model used for executing a behavior tree framework generation task, and then selecting a preset sub-task to generate a corresponding behavior tree framework; s3, the large language model sends key information obtained through analysis from the instruction to the corresponding subtasks, and the key information is used for guiding specific execution of the subtasks; s4, controlling the robot to execute subtasks in sequence according to the behavior tree framework and the key information; the framework of the behavior tree realizes the convenience of man-machine collaboration, and tasks can be adjusted and executed more naturally according to the requirements of human beings; the life support task is intelligently scheduled and executed through the behavior tree framework, manual intervention is reduced, the automation of the task is improved, and the execution efficiency and the task completion precision of the robot are improved.

The invention discloses a puzzle solution game adaptive design method and system based on ecological balance, and the method specifically comprises the steps: receiving ecological scene data selected by a player, carrying out the evaluation processing of the capability level of the player through combining with a player capability evaluation model, and generating an evaluation result; dynamically generating hierarchical ecological problems according to an evaluation result, and synchronously activating Agent behavior tree loaders of corresponding hierarchies to form an Agent modeling system; simulating a biological interaction network through an Agent modeling system, inputting real-time operation data of the biological interaction network into an LSTM double-time-sequence prediction engine for prediction processing, and generating an ecological evolution graph; and generating a dynamic visual interface according to the ecological evolution map, and triggering a difficulty regulator when detecting that the ecological imbalance index exceeds a difficulty threshold value of the current player. Complex characteristics of an ecological system are effectively simulated, more challenging and interesting puzzle solving game experience is provided for players, and meanwhile understanding and protection awareness of the players to the ecological system can be improved.

Various embodiments of the teachings herein include a workflow creation method. Some examples include: receiving an operation performed by a user on a GUI constructing a functional block model; receiving an operation performed by the user including nesting a behavior tree into a second functional block, wherein the behavior tree defines an operation for a first workcell, and the functional block model defines an operation for a second workcell; in response to a first functional block in the functional block model triggering an input event of the second functional block, then the second functional block triggers an execution signal to a root node in the behavior tree at a preset frequency, so the behavior tree starts execution from the root node and outputs an execution result; the second functional block triggers a corresponding output event according to the execution result; and generating an instance of the functional block model.

The invention relates to an automatic driving collaborative decision-making method and device based on big language model scene analysis and a medium, and the method comprises the steps: S1, carrying out the pruning operation of a future behavior tree of a current participant, which is obtained through a neural network model based on an OPM process, through employing a big language model according to the current scene information and the intention of the participant, generating a future behavior tree of the vehicle in the current scene, and giving an efficiency constraint and a safety constraint required by the next decision; s2, according to the efficiency constraint and the safety constraint, screening a future trajectory generated according to the current state of the own vehicle, sending an optimal control instruction corresponding to an optimal own vehicle trajectory obtained after screening to the vehicle-mounted control terminal, executing the optimal control instruction, and updating the state space of the OPM process after executing the optimal control instruction; and S3, circularly executing the steps S1 to S2 until the current conflict scene is passed. Compared with the prior art, the method introduces the intention of the participant as an important factor of the automatic driving decision, and improves the passing efficiency, safety and reliability of automatic driving in a complex scene.

The invention relates to a CGF-oriented typical normal form behavior model conversion method and device. The method comprises the following steps: receiving a conversion instruction input by a user; when a first conversion instruction is received, adopting a preset first conversion algorithm to respectively map each task flow of the finite-state machine and a state and transfer relationship in each task flow into a sub-tree of a behavior tree and a hierarchical node structure of each sub-tree, and outputting a behavior tree description file; when a second conversion instruction is received, disassembling the tree structure of the behavior tree into a state transition network of a finite-state machine by adopting a preset second conversion algorithm, and outputting a finite-state machine description file; and according to the description file output after receiving the conversion instruction, generating a corresponding code framework and integrating the code framework to a CGF simulation platform to output a deployable CGF entity model file. By adopting the method, the efficiency, reusability and application flexibility of CGF modeling can be improved.

The invention discloses a multi-satellite cooperative task planning method based on a behavior tree. S101, the ground planning layer carries out modeling description of satellite tasks by combing satellite task requirements and sends a behavior tree task template to the high-orbit satellite layer; s102, after receiving the task, the high-orbit satellite layer plans the task according to the behavior tree task template; task decomposition is carried out to obtain meta-task sub-trees (tasks capable of being executed by a single satellite), specific satellite planning is carried out, satellites conforming to task constraints are planned to bear task execution, and the task sub-behavior trees are issued to a low-orbit execution layer; and S103, after receiving the task sub-behavior tree, the low-orbit execution layer binds a function according to a mechanism of the behavior tree and constructs code execution so as to realize execution of a specific task. According to the method, high flexibility, expandability and execution efficiency of task planning are remarkably improved, and the defects of a traditional method in the aspects of complex task decomposition, multi-satellite coordination and real-time response capability are effectively overcome.

Various embodiments of the teachings herein include a workflow construction method. An example includes: receiving function block node addition and connection operations from a GUI on the basis of function block type graphs, wherein each function block node implements a service operation; and constructing a behavior tree corresponding to one workflow in response to the function block node addition and connection operations. Each function block type graph comprises: a function block name, a function block header, a link input port and a link output port, an input data chunk and an output data chunk, a data input port and a data output port, and a function block body. The function block node addition operation comprises a dragging operation.

The invention provides a motion path planning method for a desktop wheeled robot, which comprises the following steps: when a target task is received, loading a behavior tree corresponding to the target task, the behavior tree comprising a tree logic structure composed of a control flow node, a condition node and an action node; periodically executing the behavior tree through a behavior tree engine so as to determine an execution flow according to a logic structure of the behavior tree; when the execution process triggers a specific path planning action node, a task context corresponding to a current task stage is generated, and the task context is multi-dimensional structured data containing target description and planning constraints; dynamically selecting at least one target motion planning algorithm from a plurality of preset motion planning algorithms based on the task context; and generating a current motion path instruction by using the target motion planning algorithm, and controlling the desktop wheeled robot to execute. The method has the technical effect that the planning strategy can be dynamically adjusted according to the task context.

The present application provides a method for intelligent legal assistance and a multi-agent system, the method including: constructing a dynamically updated legal knowledge system; constructing a case diversion agent, a legal application agent, a similar case retrieval agent and a risk warning agent based on the dynamically updated legal knowledge system, and realizing a multi-agent collaborative mechanism through a collaborative structure based on message passing and state sharing; constructing a preference mapping relationship model through case complexity assessment and quality assessment based on the multi-agent collaborative mechanism, and realizing multi-case relationship perception preference mapping; based on multi-case relationship perception preference mapping, combining a large language model with Prolog-based knowledge management technology to construct a planning tree PLANTOR framework, generate a robot-oriented knowledge base and planning program, and convert them into a behavior tree; according to the PLANTOR framework, realize multi-agent reasoning chain alignment through the joint optimization of transmission optimization and embedding optimization, and complete legal decision-making assistance. The above scheme effectively improves the accuracy and efficiency of legal processing.

The present invention discloses a behavior tree-based unmanned vehicle action modeling and control method. The method constructs an unmanned vehicle action model according to a visual behavior tree. When the unmanned vehicle action model is used for simulation, when a target enters the sensor detection range, a target detection node transmits brief intelligence. A target identification node identifies the target model and payload to obtain intelligence information. A target threat analysis node analyzes the threat level of the target based on the identification results of the target identification node and provides treatment recommendations. A target tracking node is used to implement mobile tracking or non-mobile tracking based on environmental conditions, concealment capabilities, and target detection capabilities. When mobile tracking is implemented, the target tracking node controls the speed and direction of the unmanned vehicle to achieve target tracking. A cluster collaborative control node is used to achieve collaborative control between the individual unmanned vehicle entities in the unmanned vehicle cluster. The present invention implements the application of behavior trees in an unmanned vehicle simulation system.

The present application belongs to the technical field of decision-making behavior modeling and intelligent simulation, and relates to a multi-agent cooperative confrontation deduction method and system based on a hierarchical behavior tree control framework. The method comprises: constructing an editable behavior tree, wherein the parameters of the action nodes can be edited; constructing a multi-level intelligent decision-making model based on the editable behavior tree; and using the multi-level intelligent decision-making model to perform cooperative confrontation deduction of multi-agents. The present application can solve the difference between human and machine thinking, significantly improve the interpretability and artificial intervention ability of strategies, realize parameter configuration combining dynamic and static, greatly enhance the adaptability to dynamic unstructured environment, break through the bottleneck of large-scale agent cooperation, and realize multi-scale efficient management and control of "high layer-middle layer-low layer".