Dynamic scene-oriented task intention scheduling execution method, device and equipment

Abstract

The invention relates to a dynamic scene-oriented task intention scheduling execution method, device and equipment. The method comprises the steps that an intention block library is constructed, a plurality of intention blocks are stored in the intention block library, function types of the intention blocks are predefined, and three-dimensional scheduling attributes are bound; receiving a target task of the industrial robot, and calling a task model matched with the target task; determining the type of a scene where the industrial robot is located according to information collected by a sensor, selecting a corresponding scheduling strategy according to the type of the scene, and screening to obtain an intention block to be executed in the round according to the scheduling strategy and the three-dimensional scheduling attribute of each node intention block in the directed acyclic graph; conflict resolution is carried out on the to-be-executed intention blocks, a to-be-executed queue is generated according to a conflict resolution result, and the industrial robot calls corresponding hardware and software resources to execute corresponding functions according to the three-dimensional scheduling attributes of the intention blocks in the to-be-executed queue. By adopting the method, the industrial robot can efficiently, safely and autonomously complete the target task in a dynamic industrial scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial robots, in particular to a task intention scheduling execution method, device and equipment for dynamic scenarios. BACKGROUND

[0002] With the continuous development of the technical field of industrial robots, there appears an industrial robot control technology based on a preset program or a fixed process. The core feature of this technology is that it can enable an industrial robot to execute a basic industrial task according to a pre-written instruction sequence. In a production scenario with single task and structured environment, a standardized operation process can be constructed to effectively replace manual repetitive and mechanical operations, thereby providing technical support for the preliminary improvement of early industrial production efficiency. As a result, most devices in the current industrial robot field still generally use this preset program or fixed process-based control mode to carry out daily task execution work.

[0003] However, the current preset program or fixed process-based control mode has exposed many problems such as low efficiency, low safety, and resource waste when dealing with complex and variable modern industrial scenarios. First, the low task reusability makes it difficult to abstract general operation logic into reusable modules, and a large amount of manpower and time cost needs to be repeatedly invested when developing new task programs. Not only is the development efficiency low, but the maintenance process also needs to be adjusted for customized programs one by one, causing serious waste of human and time resources. Second, the scene adaptability is poor, and the execution strategy cannot be dynamically adjusted according to real-time environmental changes (such as workpiece position offset), task urgency, or robot's own cognitive state (such as device load change). When facing complex or unexpected situations, it is easy to have slow response and improper operation, which directly affects production efficiency. Third, the human-machine interaction efficiency is not high. In human-robot collaboration or remote monitoring scenarios, the robot has difficulty in intuitively and efficiently conveying its own intention and decision-making process to the operator, increasing the understanding difficulty and communication cost of personnel, and indirectly reducing the collaboration efficiency. Fourth, the conflict processing capability is weak. When there are multiple concurrent instructions (such as receiving assembly and carrying instructions at the same time) or abnormal events (such as device fault warning), there is a lack of effective priority judgment and conflict resolution mechanism, which may cause safety risks such as device collision and task interruption, threatening production safety and task continuity. SUMMARY

[0004] Therefore, it is necessary to provide a task intention scheduling execution method, device and equipment for dynamic scenarios to solve the above technical problems.

[0005] A task intention scheduling execution method for dynamic scenarios, the method comprising: construct an intent block library, the intent block library stores a plurality of atomized and reusable intent blocks, each intent block predefines a function type and binds a three-dimensional scheduling attribute; the function type corresponds to a basic operation or a high-level behavior of an industrial robot; the three-dimensional scheduling attribute includes a capability level, a task abstraction level, and a main input / output mode; receive a target task of the industrial robot, and call a task model matched with the target task; the task model includes a directed acyclic graph with nodes being the intent blocks, and a connection relationship between the nodes represents an execution dependency relationship of the intent blocks; determine a scene type in which the industrial robot is located according to real-time received sensor collection information, select a corresponding scheduling strategy according to the scene type, and filter to obtain an intent block to be executed in this round according to the scheduling strategy and the three-dimensional scheduling attribute of each node intent block in the directed acyclic graph; perform conflict resolution on the intent block to be executed, generate an execution queue according to a conflict resolution result, and call corresponding hardware and software resources to execute corresponding functions according to the three-dimensional scheduling attribute of the intent block in the execution queue.

[0006] A task intent scheduling and execution device for a dynamic scene, the device comprising: an intent block definition module configured to construct an intent block library, the intent block library storing a plurality of atomized and reusable intent blocks, each intent block predefining a function type and binding a three-dimensional scheduling attribute; the function type corresponding to a basic operation or a high-level behavior of an industrial robot; the three-dimensional scheduling attribute including a capability level, a task abstraction level, and a main input / output mode; a task model calling module configured to receive a target task of the industrial robot, and call a task model matched with the target task; the task model including a directed acyclic graph with nodes being the intent blocks, and a connection relationship between the nodes representing an execution dependency relationship of the intent blocks; a task intent scheduling module configured to determine a scene type in which the industrial robot is located according to real-time received sensor collection information, select a corresponding scheduling strategy according to the scene type, and filter to obtain an intent block to be executed in this round according to the scheduling strategy and the three-dimensional scheduling attribute of each node intent block in the directed acyclic graph; a task intent execution module configured to perform conflict resolution on the intent block to be executed, generate an execution queue according to a conflict resolution result, and call corresponding hardware and software resources to execute corresponding functions according to the three-dimensional scheduling attribute of the intent block in the execution queue.

[0007] A computer device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the following steps when executing the computer program: construct an intent block library storing a plurality of atomized and reusable intent blocks, each intent block pre-defining a function type and binding a three-dimensional scheduling attribute; the function type corresponding to a basic operation or a high-level behavior of an industrial robot; the three-dimensional scheduling attribute including a capability level, a task abstraction level, and a main input / output mode; receive a target task of the industrial robot, and call a task model matched with the target task; the task model including a directed acyclic graph with nodes being the intent blocks, and a connection relationship between the nodes representing an execution dependency relationship of the intent blocks; determine a scene type in which the industrial robot is located according to real-time received sensor collection information, select a corresponding scheduling strategy according to the scene type, and screen an intent block to be executed in this round according to the scheduling strategy and the three-dimensional scheduling attribute of each node intent block in the directed acyclic graph; perform conflict resolution on the intent block to be executed, generate an execution queue according to a conflict resolution result, and call corresponding hardware and software resources to execute corresponding functions according to the three-dimensional scheduling attribute of the intent block in the execution queue.

[0008] The above task intent scheduling and execution method, device and equipment oriented to a dynamic scene, by constructing an intent block library storing atomized and reusable intent blocks, a general industrial robot operation logic can be highly modularized and reused, greatly shortening a new task development cycle and cost, effectively improving task reusability and development efficiency, by calling a directed acyclic graph containing intent blocks after receiving a target task, relying on an execution logic of a clear dependency relationship between nodes, determining a scene type according to real-time sensor information, matching a scheduling strategy, and screening an intent block to be executed according to the strategy and three-dimensional attribute, the robot can break away from the limitation of a fixed execution flow, dynamically adapt to different scene requirements, significantly enhance scene adaptability and robustness, by performing conflict resolution on the intent block to be executed to generate an execution queue, and calling corresponding software and hardware resources according to the three-dimensional scheduling attribute of the intent block, system confusion caused by execution conflicts can be avoided, stable system operation is ensured, accurate resource matching is realized, and efficient task execution is supported. The embodiment of the present application can enable the industrial robot to efficiently, safely and autonomously complete a target task in a dynamic industrial scene. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 a flowchart of a task intent scheduling and execution method oriented to a dynamic scene in an embodiment; Figure 2 a flowchart of a task intent scheduling and execution method oriented to a dynamic scene in a specific embodiment; Figure 3 a DAG construction flowchart in an embodiment; Figure 4 a dynamic scheduling strategy selection flowchart in an embodiment; Figure 5 a structural block diagram of a task intention scheduling execution system for a dynamic scene in an embodiment; Figure 6 an internal structural diagram of a computer device in an embodiment. DETAILED DESCRIPTION

[0010] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0011] In one embodiment, as shown in Figure 1 a task intention scheduling execution method for a dynamic scene is provided, comprising the following steps: Step 102, constructing an intention block library, the intention block library storing a plurality of atomized and reusable intention blocks, each intention block predefining a function type and binding a three-dimensional scheduling attribute.

[0012] An intention block is to decompose the intention of an industrial robot to perform a task into the smallest, independent and reusable functional unit. Each intention block encapsulates specific operation logic, has atomization characteristics (non-divisible) and three-dimensional attribute binding. The three-dimensional scheduling attribute includes the capability hierarchy (SAT hierarchy), the task abstraction level and the main input and output modalities. The function type is a core capability label predefined by the intention block, and the function type corresponds to the basic operation or high-level behavior of the industrial robot. The basic operation includes basic actions such as clamping and moving, and the high-level behavior corresponds to complex logic (such as path planning) behavior.

[0013] It can be understood that by atomization encapsulation and reusable design, the general operation logic is abstracted into a standardized module, without writing code from zero for each new task, only need to call or combine existing intention blocks, greatly shorten the development cycle, and the binding of the three-dimensional scheduling attribute provides accurate scheduling labels for subsequent dynamic scene adaptation and resource calling.

[0014] Step 104, receiving a target task of an industrial robot, calling a task model matched with the target task. The task model includes a directed acyclic graph with nodes being intention blocks, and the connection relationship between nodes represents the execution dependency relationship of each intention block.

[0015] The task model is a predefined execution scheme for the target task, and its essence is a structured execution logic composed of intention blocks. This embodiment adopts a directed acyclic graph (DAG), the nodes of the DAG correspond to the intention blocks, the directed edges represent the execution dependency relationship between the intention blocks (such as moving above the workpiece needs to be executed before grasping the workpiece), and there is no circular dependency in the graph, ensuring that the execution logic has no contradictions.

[0016] By calling the task model based on the intention block DAG, the complex task is disassembled into a traceable and manageable combination of intention blocks. The task structure is visualized through the DAG, facilitating debugging and modification. Meanwhile, the clear definition of the dependency relationship ensures the orderliness of the task execution, providing a clear logical framework for subsequent dynamic scheduling, which indirectly improves the stability of the system.

[0017] In step 106, the scene type in which the industrial robot is located is determined according to the real-time sensor acquisition information. According to the scene type, the corresponding scheduling strategy is selected. According to the scheduling strategy and the three-dimensional scheduling attributes of the intention blocks in the directed acyclic graph, the intention blocks to be executed in this round are screened.

[0018] The sensor acquisition information refers to the environmental data (such as obstacle distance, workpiece position) and ontology state data obtained by the robot through laser radar, camera and other devices. The scene type is the environment and task state category in which the robot is located according to real-time information, including high safety scene (such as close-range human-robot interaction, high-speed motion), low cognitive load scene (such as the need to explain the decision to the operator) and regular scene. The scheduling strategy is the intention block screening rule formulated for different scene types, which determines how to select the current intention block to be executed from the DAG.

[0019] It can be understood that step 106 makes the robot no longer rely on fixed execution flow, but can dynamically adjust the block selection logic according to environmental changes (such as sudden appearance of obstacles). At the same time, combined with the three-dimensional scheduling attributes of the intention blocks, the selected intention blocks are ensured to accurately match the scene requirements, improving the robustness and human-robot interaction efficiency of the system.

[0020] In step 108, the conflict resolution of the intention blocks to be executed is performed, and the to-be-executed queue is generated according to the conflict resolution result. The industrial robot executes the corresponding functions by calling the corresponding hardware and software resources according to the three-dimensional scheduling attributes of the intention blocks in the to-be-executed queue.

[0021] Conflict resolution is a mechanism for handling contradictions through priority rules when multiple to-be-executed intention blocks have logical mutual exclusion (such as emergency stop and continue work) or resource competition (such as simultaneous calling of the same sensor). The to-be-executed queue is a list of intention blocks arranged in execution order after conflict resolution, which is the actual execution instruction sequence of the robot.

[0022] In the above task intention scheduling execution method for dynamic scenarios, by constructing an intention block library that stores atomized and reusable intention blocks, the operation logic of a general industrial robot can be highly modularized and reused, greatly shortening the development cycle and cost of new tasks, effectively improving task reusability and development efficiency. By receiving a target task and calling a directed acyclic graph containing intention blocks, relying on the clear execution logic between nodes, combining real-time sensor information to determine the scene type, matching the scheduling strategy, and selecting the intention blocks to be executed according to the strategy and three-dimensional attributes, the robot can break free from the restrictions of fixed execution flow and dynamically adapt to different scene requirements, significantly enhancing scene adaptability and robustness. By resolving conflicts among the intention blocks to be executed to generate an execution queue, and calling corresponding hardware and software resources according to the three-dimensional scheduling attributes of the intention blocks, the system can avoid chaos caused by execution conflicts, ensure stable operation, achieve precise resource matching, support efficient task execution, and provide structured support for the robot to autonomously complete the target task, promoting more orderly autonomous execution.

[0023] In one embodiment, the method further comprises: after the execution of the intention blocks in the execution queue is completed, determining whether all the intention blocks in the directed acyclic graph have been executed; if not, selecting the intention blocks to be executed in the next round until all the intention blocks in the directed acyclic graph have been executed, and determining that the target task is completed.

[0024] In one embodiment, the method further comprises: if there is no task model matching the target task, selecting or dynamically generating intention blocks matching the requirements of the target task from the intention block library, binding three-dimensional scheduling attributes to the selected or generated intention blocks, and constructing a directed acyclic graph according to the execution dependency relationship of each intention block, taking the directed acyclic graph as the new task model corresponding to the target task.

[0025] In one embodiment, in the three-dimensional scheduling attributes, the capability level includes the perception layer, the action layer, and the reasoning and prediction layer, and the task abstraction level includes the target layer, the planning layer, and the operation layer; the main input and output modalities include vision, hearing, and touch.

[0026] In this embodiment, the design of three-dimensional scheduling attributes allows subsequent dynamic scheduling (such as screening redundant modal blocks in high-security scenarios or calling reasoning layer blocks in low-cognitive scenarios) to no longer rely on vague experience, but to achieve precise block selection based on clear attributes, thereby achieving scene adaptation and efficient human-robot interaction.

[0027] Specifically, when performing intention block atomization and encapsulation, the intention of the industrial robot to execute a task is decomposed into a series of smallest, independent, and reusable intention blocks. Each intention block is bound with three-dimensional attributes: SAT level: represents the capability level of the intent block, including the perception layer (S1: basic sensor input), the action layer (A1: execute specific actions), and the reasoning / prediction layer (T1: advanced logical reasoning, decision or prediction).

[0028] Task abstraction level: represents the abstraction level of the intent block in the task, including the strategic layer, the tactical layer, and the operational layer.

[0029] Primary input / output modality: represents the input / output modality that the intent block mainly relies on, including vision, hearing, and touch.

[0030] For example, an intent block that detects obstacles in front may have the attributes: SAT level = perception (S1), task abstraction level = operation, and primary modality = vision. An intent block that plans an obstacle avoidance path may have the attributes: SAT level = reasoning (T1), task abstraction level = tactics, and primary modality = vision.

[0031] In one embodiment, the type of scene in which the industrial robot is located is determined according to real-time sensor acquisition information, including: integrating the environment information collected by the sensor and the robot body state information into a scene feature vector; matching the scene feature vector with a pre-set scene rule library to determine the scene type as a high safety scene, a low cognitive load scene, or a regular scene.

[0032] In this embodiment, the scheduling engine monitors the environment scene in which the robot is located, the urgency of the task, and other related state information in real time. According to the monitored scene information, the intent blocks in the DAG are dynamically scheduled and executed. During the scheduling process, dynamic scene adaptation scheduling can be achieved: High safety scene adaptation: In a scene that requires extremely high safety (such as when a robot arm is performing high-speed and precise motion), the scheduling engine will preferentially select and activate intent blocks with redundant modalities or high reliability attributes. For example, when a potential danger is detected, not only visual recognition is relied on, but also vibration signals from tactile sensors are activated simultaneously to form a redundant perception and feedback of visual light + tactile vibration to improve safety margin.

[0033] Low cognitive load scene adaptation: When the robot needs to explain its decision logic to the operator or higher-level system, or in the case of scene complexity, the scheduling engine will enable intent blocks with higher SAT levels (such as T1 layer reasoning / prediction), and can explain the decision process through the auditory modality (such as voice output) to reduce the cognitive load of the operator.

[0034] Regular scene adaptation: If it is matched as a regular scene, the scheduling engine strictly schedules the intent blocks according to the topological order of the task DAG.

[0035] In one embodiment, the corresponding scheduling strategy is selected according to the scene type, including: if the scene type is a high-safety scene, the scheduling strategy is to preferentially screen the intent block with redundant input and output modalities or high-reliability characteristics; if the scene type is a low-cognitive-load scene, the scheduling strategy is to enable the intent block of the inference prediction layer capability level and prepare multi-modal explanation information; and if the scene type is a regular scene, the scheduling strategy is to screen the intent block in topological order of the directed acyclic graph.

[0036] In the present embodiment, the redundant input and output modalities refer to simultaneously enabling multiple perception / execution channels in a critical scene. The cognitive load refers to the amount of mental input required for an operator to understand the robot decision-making process. The present embodiment directly maps the scene type to the scheduling strategy: the high-safety scene preferentially screens the redundant modality or high-reliability intent block, thereby improving the safety margin from the strategy level and reducing the safety risk of high-speed precision motion and other scenes; the low-cognitive-load scene actively enables the inference prediction layer intent block and prepares multi-modal explanations, thereby specifically addressing the pain points of low human-machine interaction efficiency and reducing the mental input (cognitive load) of the operator to understand the decision; and the regular scene is executed in DAG order to ensure the efficiency of regular tasks, thereby achieving a balance of safety, interaction, and efficiency in different scenes and making the scheduling strategy more targeted.

[0037] In one embodiment, according to the scheduling strategy and the three-dimensional scheduling attributes of the node intent block in the directed acyclic graph, the intent block to be executed in the present round is screened, including: screening all node intent blocks whose all preceding dependent intent blocks have been executed from the directed acyclic graph as candidate intent blocks; and performing attribute matching on the candidate intent blocks according to the requirements of the current scheduling strategy on the three-dimensional scheduling attributes, so as to determine the candidate intent blocks that pass the attribute matching as the intent block to be executed in the present round.

[0038] In one embodiment, the conflict resolution of the intent block to be executed includes: presetting the safety weight priority of each intent block, and when it is detected that the execution logic of the intent block has mutual exclusion or resource competition, the high-priority intent block is retained and the low-priority intent block is suppressed or delayed.

[0039] In the present embodiment, the safety weight priority of each intent block is preset, and when it is detected that the execution logic of the intent block has mutual exclusion or resource competition: if it is an irreconcilable safety conflict, the high-priority intent block is retained and the low-priority intent block is suppressed; and if it is a conflict that can be adjusted in time sequence, the high-priority intent block is retained and the low-priority intent block is delayed until the high-priority intent block is executed.

[0040] It can be understood that by presetting the priority (such as the safety class intention block being prior to the regular task block), the execution of urgent or critical instructions is ensured to be prior, the safety risk is reduced, meanwhile, based on the three-dimensional scheduling attribute of the intention block, the corresponding resource is called (such as the visual modal intention block calling the camera), the accurate matching of hardware and software resources in the execution process is ensured, the accuracy of task execution is improved, and the structured to-be-executed queue further guarantees the orderliness of task execution, and finally the balance between system safety and efficiency is realized.

[0041] In one specific embodiment, as shown in Figure 2 , a flowchart of task intention scheduling and execution for dynamic scenarios is provided. It includes: (S101) When the system receives a new industrial task instruction, for example, "move P001 workpiece in A point warehouse to B point assembly table", the task planner first analyzes the instruction.

[0042] (S102) The scheduling engine first determines whether the task is a new task. For example, by comparing the task ID or task content hash value, it is checked whether a corresponding task model exists in the system. If it exists, it directly jumps to step S106.

[0043] (S103) If it is a new task, the "intention block encapsulator" is called to select or dynamically generate the minimum intention block unit required to constitute the task from the "intention block library". The intention block library pre-stores modularized intention blocks of various basic operations (such as clamping, moving, visual detection) and high-level behaviors (such as path planning, abnormality judgment).

[0044] (S104) Each selected intention block binds its three-dimensional attributes through the "intention block attribute definition" module. For example: "Detect P001 workpiece" intention block: SAT level = perception (S1), task abstraction layer = operation, main modal = vision.

[0045] "Plan obstacle avoidance path" intention block: SAT level = reasoning (T1), task abstraction layer = tactics, main modal = vision.

[0046] (S105) The logical dependency relationship of all intention blocks is input into the "DAG construction module" to generate the "task DAG" of the task, which represents the execution order and dependency relationship of the intention blocks. For details of the detailed steps of the construction process, please refer to Figure 3 , which is as follows: (S105a) Define and input dependency relationship: Each intention block will be attached with a dependency relationship list. For example, the dependency list of the "grab workpiece" intention block is ["move above the workpiece", "open the clamping jaw"]. The DAG construction module receives all intention blocks and their dependency relationship lists as input.

[0047] (S105b) Creating nodes and directed edges: The module first creates each intent block as a node of a graph. Then, it adds directed edges between nodes according to the dependency list, for example, from the "move above the workpiece" node to the "grab the workpiece" node.

[0048] (S105c) Cycle dependency check: After all edges are added, the module performs a topological sorting algorithm (such as Kahn's algorithm or an algorithm based on depth-first search) to detect whether there is a circular dependency in the graph. If a cycle is detected (for example, A depends on B, and B depends on A), the system will throw an exception, indicating a task logic error, thus ensuring that the generated DAG is a directed acyclic graph. After the check passes, the generated DAG is solidified in the form of an adjacency list or adjacency matrix data structure.

[0049] (S106) If the task already exists, directly load its pre-built task DAG from storage.

[0050] (S107) The "scene monitor" module in the scheduling engine enters a continuous monitoring loop. It continuously collects information about the environment around the robot (such as obstacles, personnel positions) through various environmental perception sensors (such as lidar, camera), and combines it with the robot's own state information (such as the current speed of the robot arm's end, the remaining time of the task).

[0051] (S108) The scheduling engine evaluates the current scene type based on these real-time information and pre-set scene rules. The detailed steps of this evaluation and decision-making process are described in Figure 4 , as follows: (S108a) Scene feature vectorization: The scene monitor integrates the collected multi-source information, such as the distance to the nearest obstacle, the confidence of the presence of personnel, the speed of the robot arm, into a real-time scene feature vector V_scene.

[0052] (S108b) Scene rule base matching: The scheduling engine matches V_scene with the internal maintained scene rule base. This rule base defines the mapping relationship between different scene types and feature vectors, for example: Rule 1 (High safety scene): IF (distance to the nearest obstacle < 0.5 meters OR personnel confidence > 0.8) AND (robot arm speed > 0.5 meters / second) THEN scene type = "High safety scene".

[0053] Rule 2 (Low cognitive load / interpretation demand scene): IF (number of task execution failures > 2 OR operator issues "interpretation" instruction) THEN scene type = "Low cognitive load / interpretation demand scene".

[0054] Rule 3 (Regular Scenario): ELSE Scenario Type = "Regular Scenario".

[0055] (S109-S112) The scheduling engine selects and executes the corresponding scheduling strategy from the strategy library according to the matched scenario type: (S110) High Safety Strategy: If the match is "High Safety Scenario", the scheduling engine will prioritize scheduling intent blocks with redundant modalities (such as enabling visual and tactile obstacle avoidance at the same time) or high reliability (such as reducing speed, increasing checkpoints).

[0056] (S111) Interpretation Priority Strategy: If the match is "Low Cognitive Load / Interpretation Demand Scenario", the scheduling engine will enable higher-order T1 layer intent blocks (such as "decision logic analysis") and prepare to output interpretation information through voice and other means.

[0057] (S112) Standard Sequence Strategy: If the match is "Regular Scenario", the scheduling engine strictly follows the topological order of the task DAG to schedule intent blocks.

[0058] (S113-S114) After the scheduling engine determines the intent blocks that need to be executed in this round, the "conflict resolver" checks whether there are potential conflicts between these intent blocks and external new instructions (such as the operator pressing the emergency stop button).

[0059] (S115) Once a conflict is detected, for example, the "emergency stop" intent block and the "continue executing task" intent block are triggered at the same time, the conflict resolution mechanism will sort the conflicting intent blocks according to the preset "safety weight priority", and dynamically suppress or delay the execution of low-priority intent blocks by inserting a high-priority instruction or removing / suspending a low-priority instruction in the "to-be-executed queue" of the scheduling engine, to ensure that high-priority intent blocks (such as emergency stop) can be executed first, thereby resolving the conflict.

[0060] (S116) The intent blocks that do not have conflicts or have been successfully resolved are determined as the final execution units in this round and are added to a first-in-first-out (FIFO) structure "to-be-executed queue" by the system, waiting for the robot body to call.

[0061] (S117) The execution module of the robot body takes out the instructions from the "to-be-executed queue" and cooperatively calls the corresponding hardware and software resources according to the three-dimensional attributes (SAT level, task abstraction layer, main modality) bound to the intent blocks. For example, if the main modality of the intent block is "vision", the vision sensor and image processing unit will be driven; if it is "action" layer, the servo motor will be driven to execute physical actions.

[0062] (S118) During the execution, the system continuously monitors the execution status of the intention block. Once a completion signal is received from the underlying hardware or a preset completion condition is met (e.g., a sensor detects that the gripper has been closed to the position), it is determined that the current intention block execution is completed.

[0063] (S119) After each intention block is completed, the scheduling engine checks whether all nodes (intention blocks) in the task DAG have been marked as "executed".

[0064] (S120) If all intention blocks have been executed, the system determines that the entire industrial task is completed, and the process ends. If there are still unexecuted intention blocks in the task DAG, the process will return to step S107 to enter the next round of "scene monitoring and scheduling" loop to cope with the possible changes in the environment and task state, and continue to execute the subsequent tasks.

[0065] It should be understood that, although Figure 1 the steps in the flowchart of the system are displayed in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated in this article, the execution of these steps has no strict order restriction, and these steps can be executed in other orders. Moreover, Figure 1 at least part of the steps in the system can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be alternately executed with other steps or at least part of the sub-steps or stages of other steps.

[0066] In one embodiment, as shown in Figure 5 , a dynamic scene-oriented task intention scheduling and execution system is provided, including an intention block management and generation module, a task and scheduling control module, and a robot body module. Wherein: The intention block management and generation module is responsible for building atomized and reusable intention blocks, including an intention block library, an intention block encapsulator, and an intention block attribute definition unit. The intention block library is used to store basic intention blocks, the intention block encapsulator is used to encapsulate the intention blocks, and the intention block attribute definition unit is used to pre-bind three-dimensional scheduling attributes for the intention blocks, and generate intention block instances; A task and scheduling control module is responsible for receiving target tasks and dynamically scheduling intention blocks based on real-time scenes, and includes a task planner, a DAG construction module, a scene monitor, a scheduling engine, and a conflict resolver. The task planner is used to receive task instructions and provide task logic. The DAG construction module is used to generate a task directed acyclic graph containing intention block dependency relationships. The scene monitor is used to receive real-time data from environment perception sensors to determine scene types. The scheduling engine is used to filter to-be-executed intention blocks in combination with scene types and task directed acyclic graphs. The conflict resolver is used to resolve conflicts of to-be-executed intention blocks, and then generate a to-be-executed queue. A robot body module is responsible for executing functions and implementing human-machine interaction, and includes a communication module, a perception module, and an execution module. The communication module is used to process and interpret instructions and feedback instructions, and can also output multi-modal interpretation information to operators or external systems. The perception module is used to execute perception instructions and feed back perception results. The execution module is used to output physical actions.

[0067] In one embodiment, a task intention scheduling and execution device for dynamic scenes is provided, which includes: An intention block definition module is used to construct an intention block library. The intention block library stores a plurality of atomized and reusable intention blocks. Each intention block is pre-defined with a function type and bound to three-dimensional scheduling attributes. The function type corresponds to a basic operation or a high-level behavior of an industrial robot. The three-dimensional scheduling attributes include a capability level, a task abstraction level, and a main input / output modality. A task model calling module is used to receive a target task of an industrial robot and call a task model matched with the target task. The task model includes a directed acyclic graph with nodes being intention blocks. Connection relationships between the nodes represent execution dependency relationships of the intention blocks. A task intention scheduling module is used to determine a scene type in which the industrial robot is located according to real-time sensor acquisition information, select a corresponding scheduling strategy according to the scene type, and filter to-be-executed intention blocks of this round according to the scheduling strategy and three-dimensional scheduling attributes of intention blocks in the directed acyclic graph. A task intention execution module is used to resolve conflicts of to-be-executed intention blocks, generate a to-be-executed queue according to a conflict resolution result, and call corresponding hardware and software resources to execute corresponding functions according to three-dimensional scheduling attributes of intention blocks in the to-be-executed queue.

[0068] The specific limitations of the task intention scheduling execution device for dynamic scenarios can refer to the limitations of the task intention scheduling execution method for dynamic scenarios described above, and will not be described here. Each module in the task intention scheduling execution device for dynamic scenarios described above can be realized by software, hardware and combinations thereof, in whole or in part. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each module.

[0069] In one embodiment, a computer device, which can be a terminal, has an internal structure diagram as shown in Figure 6 The computer device includes a processor, a memory, a network interface, a display screen and an input device connected by a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement a task intention scheduling execution method for dynamic scenarios. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad provided on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0070] Those skilled in the art can understand that Figure 6 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0071] In one embodiment, a computer device is provided, which includes a memory and a processor. The memory stores a computer program, and the processor implements the steps of the method in the above-mentioned embodiments when executing the computer program.

[0072] The technical features of the above embodiments can be combined in any way. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist, they should be considered as the scope of the present disclosure.

[0073] The above embodiments only express several implementation ways of the present application, and the description is more specific and detailed, but it should not be understood as a limitation to the scope of the application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A task intent scheduling and execution method for dynamic scenarios, characterized in that, The method includes: An intent block library is constructed, which stores multiple atomic and reusable intent blocks. Each intent block predefines a function type and binds a three-dimensional scheduling attribute. The function type corresponds to the basic operation or advanced behavior of the industrial robot. The three-dimensional scheduling attribute includes capability level, task abstraction level, and main input / output modality. The system receives the target task of the industrial robot and invokes a task model that matches the target task. The task model includes a directed acyclic graph where nodes are intent blocks, and the connection relationships between nodes represent the execution dependencies of each intent block. The scenario type of the industrial robot is determined based on the real-time sensor data, and a corresponding scheduling strategy is selected based on the scenario type. Based on the scheduling strategy and the three-dimensional scheduling attributes of the intent blocks of each node in the directed acyclic graph, the intent blocks to be executed in this round are selected. The intent blocks to be executed are conflict-resolved, and an execution queue is generated based on the conflict resolution results. The industrial robot calls the corresponding hardware and software resources to execute the corresponding functions according to the three-dimensional scheduling attributes of the intent blocks in the execution queue.

2. The method according to claim 1, characterized in that, The method further includes: After the intent blocks in the queue to be executed are completed, it is determined whether all node intent blocks in the directed acyclic graph have been executed. If not completed, the next round of intent blocks to be executed is selected, and the target task is determined to be completed when all intent blocks of all nodes in the directed acyclic graph have been executed.

3. The method according to claim 1, characterized in that, The method further includes: If no task model matches the target task, an intent block matching the target task requirements is selected from the intent block library or dynamically generated. Three-dimensional scheduling attributes are bound to the selected or generated intent blocks, and a directed acyclic graph is constructed based on the execution dependencies of each intent block. This directed acyclic graph is then used as the new task model corresponding to the target task.

4. The method according to claim 1, characterized in that, In the three-dimensional scheduling attributes, the capability levels include a perception layer, an action layer, and a reasoning and prediction layer; the task abstraction levels include a target layer, a planning layer, and an operation layer; and the main input-output modalities include visual, auditory, and tactile.

5. The method according to claim 1, characterized in that, The process of determining the scene type of the industrial robot based on real-time sensor data includes: The environmental information and robot body state information collected by the sensors are integrated into a scene feature vector; The scene feature vector is matched with a preset scene rule base to determine whether the scene type is a high-security scene, a low-cognitive-load scene, or a normal scene.

6. The method according to claim 1, characterized in that, The step of selecting the corresponding scheduling strategy based on the scenario type includes: If the scenario type is a high-security scenario, the scheduling strategy is to prioritize filtering intent blocks with redundant input / output modalities or high reliability characteristics. If the scenario type is a low cognitive load scenario, the scheduling strategy is to enable the intent block at the inference and prediction layer capability level and prepare multimodal interpretation information. If the scenario type is a regular scenario, the scheduling strategy is to filter intent blocks according to the topological order of the directed acyclic graph.

7. The method according to claim 1, characterized in that, Based on the scheduling strategy and the three-dimensional scheduling attributes of the intent blocks of each node in the directed acyclic graph, the intent blocks to be executed in this round are selected as follows: Select node intent blocks from the directed acyclic graph where all preceding dependent intent blocks have been executed, and use them as candidate intent blocks. Based on the requirements of the current scheduling strategy for the three-dimensional scheduling attributes, the candidate intent blocks are matched for attributes, and the candidate intent blocks that pass the attribute matching are determined as the intent blocks to be executed in this round.

8. The method according to claim 1, characterized in that, The conflict resolution of the intent block to be executed includes: The security weight priority of each intent block is preset. When mutual exclusion or resource contention is detected in the execution logic of an intent block, the high-priority intent block is retained, while the low-priority intent block is suppressed or delayed.

9. A task intent scheduling and execution device for dynamic scenarios, characterized in that, The device includes: The intent block definition module is used to build an intent block library, which stores multiple atomic and reusable intent blocks. Each intent block predefines a function type and binds three-dimensional scheduling attributes. The function type corresponds to the basic operation or advanced behavior of the industrial robot. The three-dimensional scheduling attributes include capability level, task abstraction level, and main input / output modality. The task model invocation module is used to receive the target task of the industrial robot and invoke the task model that matches the target task; the task model includes a directed acyclic graph with nodes as intent blocks, and the connection relationship between nodes represents the execution dependency relationship of each intent block; The task intent scheduling module is used to determine the scene type of the industrial robot based on the real-time sensor data, select the corresponding scheduling strategy based on the scene type, and filter the intent blocks to be executed in the current round based on the scheduling strategy and the three-dimensional scheduling attributes of the intent blocks of each node in the directed acyclic graph. The task intent execution module is used to resolve conflicts in intent blocks to be executed and generate an execution queue based on the conflict resolution results. The industrial robot calls the corresponding hardware and software resources to execute the corresponding functions according to the three-dimensional scheduling attributes of the intent blocks in the execution queue.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

Images