Mechanical arm control method and system based on self-adaptive adjustment

Abstract

The invention belongs to the field of mechanism control, and provides a mechanical arm control method and system based on adaptive adjustment, and the method comprises the steps: coding a plurality of mechanical arm control behaviors into memetic fragments, endowing each memetic fragment with survival fitness, and carrying out the selection, intersection and mutation operation on the memetic fragments based on the survival fitness, generating a control behavior evolution graph; constructing a task semantic parameter map, and dynamically activating or disabling the semantic nodes by a monitoring mechanism according to the state of the current task to generate real-time working condition description of the current task; according to the similarity between the working condition embedding vector and the semantic embedding of each memetic fragment, extracting the memetic fragment most relevant to the current working condition from the control behavior evolution graph to form a candidate memetic factor graph; selecting the control path with the highest score as an optimal control path; and performing fine tuning on the control parameters through a local disturbance mechanism, and taking the fine-tuned control parameters as final optimal motion control parameters.

Description

Technical Field

[0001] The present invention belongs to the field of mechanism control, and in particular relates to a method and system for controlling a robotic arm based on adaptive regulation. Background Art

[0002] With the rapid development of intelligent manufacturing and industrial automation technologies, robotic arms, as key execution units, have been widely used in scenarios such as electronic assembly, precision machining, and automated inspection. To adapt to the diverse and flexible demands of industrial production, robotic arm control technology has evolved from traditional static preset modes to dynamic perception and autonomous decision-making. Especially in complex working conditions or environments with frequent task switching, achieving real-time optimization and adaptive adjustment of robotic arm motion control parameters has become a core issue in current research and industrial applications.

[0003] Existing robotic arm control methods are primarily based on preset trajectory planning, force control models, or PID control strategies based on sensor feedback. These methods rely on manually defined task models and parameter tables, and have a certain degree of execution efficiency for known tasks. However, in scenarios where tasks change frequently or the execution environment is complex and ever-changing, these methods often suffer from issues such as response lag, poor parameter adaptability, and excessive control rigidity, making them unable to effectively support the demands of industrial applications that require both high dynamics and high reliability.

[0004] Therefore, there is an urgent need for an intelligent control method that can dynamically generate control strategies based on the current working conditions. This method should have the capabilities of semantic perception, behavior selection and parameter optimization, and be able to adapt to the accuracy and speed requirements of different task stages to improve the overall flexibility and stability of the system. Summary of the Invention

[0005] In order to solve the problems in the prior art, the present invention provides a robotic arm control method based on adaptive adjustment, comprising the following steps:

[0006] Encoding multiple robotic arm control behaviors into meme segments, each of which represents a control behavior module related to a specific task, and assigning a survival fitness to each meme segment. Based on the survival fitness, selection, crossover, and mutation operations are performed on the meme segments to generate a control behavior evolution map, which is used to guide the motion control execution of the robotic arm;

[0007] Establish a distributed semantic monitoring mechanism and construct a task semantic parameter map. The task semantic parameter map includes multiple semantic nodes and their relationships. Each semantic node corresponds to a type of working condition, task goal, or actuator state. The monitoring mechanism dynamically activates or disables the semantic node based on the current task state to generate a real-time working condition description of the current task.

[0008] Converting the real-time working condition description into a working condition embedding vector, and extracting the meme fragment most relevant to the current working condition from the control behavior evolution graph based on the similarity between the working condition embedding vector and the semantic embedding of each meme fragment, to form a candidate meme factor graph;

[0009] Constructing a path evaluation function based on the historical success rate, control cost, and failure rate of the control path in the candidate module factor graph, scoring multiple control paths, and selecting the control path with the highest score as the optimal control path;

[0010] The original control parameters are extracted from the optimal control path, and the control parameters are fine-tuned through a local perturbation mechanism to adapt to the accuracy and dynamic requirements of the current task. The fine-tuned control parameters are used as the final optimal motion control parameters and output to drive the motion control execution of the robotic arm.

[0011] Furthermore, the meme fragment refers to the basic operation unit formed in the process of controlling behavior. Each meme fragment contains action instructions, path planning strategies or execution constraints with independent functions in a certain type of task. It is the smallest module unit with control significance in the process of robotic arm control.

[0012] Furthermore, the survival fitness is a quantitative indicator of the performance of the meme fragment in a specified task or environment, which is obtained by normalizing and calculating the meme fragment based on its historical success rate, task completion efficiency, and control stability factors.

[0013] Furthermore, the evolutionary map is a graph structure composed of a series of meme fragments and the connection relationships between them. The evolutionary map records the connection sequence of meme fragments during multiple task executions in history, reflecting the evolutionary path and adaptability weight of the control strategy.

[0014] Furthermore, the task semantic parameter graph refers to a working condition knowledge network organized in a graph structure, where nodes in the graph represent abstract semantic units and edges represent logical relationships or causal associations between nodes.

[0015] Another aspect of the present invention provides a robotic arm control system based on adaptive regulation, characterized in that the system includes the following modules:

[0016] a meme segment generation module, configured to encode a plurality of robotic arm control behaviors into meme segments, each of the meme segments representing a control behavior module associated with a specific task, and assigning a survival fitness to each of the meme segments;

[0017] a meme evolution construction module, configured to perform selection, crossover, and mutation operations on meme segments based on the survival fitness, and generate a control behavior evolution map, wherein the control behavior evolution map is used to guide motion control execution of the robotic arm;

[0018] A semantic monitoring module is used to establish a distributed semantic monitoring mechanism and construct a task semantic parameter map. The task semantic parameter map includes multiple semantic nodes and their associations. Each semantic node corresponds to a type of working condition, task goal, or actuator state. The monitoring mechanism dynamically activates or disables the semantic node according to the current task state to generate a real-time working condition description of the current task.

[0019] a meme selection module, configured to convert the real-time working condition description into a working condition embedding vector, and extract the meme fragment most relevant to the current working condition from the control behavior evolution graph based on the similarity between the working condition embedding vector and the semantic embedding of each meme fragment, to form a candidate meme factor graph;

[0020] a path evaluation module, configured to construct a path evaluation function based on the historical success rate, control cost, and failure rate of the control path in the candidate module factor graph, score the multiple control paths, and select the control path with the highest score as the optimal control path;

[0021] A parameter fine-tuning module is used to extract the original control parameters from the optimal control path and fine-tune the control parameters through a local perturbation mechanism to adapt to the accuracy and dynamic requirements of the current task. The fine-tuned control parameters are used as the final optimal motion control parameters and output to drive the motion control execution of the robotic arm.

[0022] Furthermore, the meme fragment refers to the basic operation unit formed in the process of controlling behavior. Each meme fragment contains action instructions, path planning strategies or execution constraints with independent functions in a certain type of task. It is the smallest module unit with control significance in the process of robotic arm control.

[0023] Furthermore, the survival fitness is a quantitative indicator of the performance of the meme fragment in a specified task or environment, which is obtained by normalizing and calculating the meme fragment based on its historical success rate, task completion efficiency, and control stability factors.

[0024] Furthermore, the evolutionary map is a graph structure composed of a series of meme fragments and the connection relationships between them. The evolutionary map records the connection sequence of meme fragments during multiple task executions in history, reflecting the evolutionary path and adaptability weight of the control strategy.

[0025] Furthermore, the task semantic parameter graph refers to a working condition knowledge network organized in a graph structure, where nodes in the graph represent abstract semantic units and edges represent logical relationships or causal associations between nodes.

[0026] The present invention establishes a high-level semantic association between task conditions and control behaviors by constructing a control behavior evolution map and a task semantic parameter map, enabling the system to autonomously select adapted meme segments based on the current embedded information of the working conditions, thereby significantly improving the adaptive control capability and response efficiency of the robotic arm in a multi-task environment.

[0027] The present invention introduces a path evaluation function, comprehensively considers the historical success rate, control cost and failure rate, screens the optimal control path, and further fine-tunes the control parameters through a local perturbation mechanism, thereby achieving dynamic optimization of the robot arm control strategy, effectively reducing the control error rate and improving the execution accuracy.

[0028] The method of the present invention has a clear structure and independent operating logic. It can be widely applied to various types of industrial robotic arm systems, especially to complex operation scenarios with frequent task switching and drastic changes in environmental conditions. It has strong generalization, scalability and engineering practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0030] Figure 1 is a flow chart of the method of the present invention;

[0031] Figure 2 It is a schematic diagram of the evolutionary map before evolution;

[0032] Figure 3 It is a schematic diagram of the evolutionary map;

[0033] Figure 4 It is a schematic diagram of the task semantic parameter map. DETAILED DESCRIPTION

[0034] The invention is preferably described below in conjunction with the accompanying drawings and specific embodiments.

[0035] See also Figure 1In one embodiment, the present invention proposes a robotic arm control method based on adaptive adjustment. This method aims to address the difficulty in accurately matching robotic arm control parameters in diverse operational tasks and complex industrial environments by providing a control strategy with autonomous perception, intelligent selection, and dynamic optimization capabilities. By constructing an evolutionary control behavior graph, combined with a task semantic perception mechanism and historical feedback-driven adjustment logic, this method enables adaptive adjustment of the robotic arm's control parameters under different operating and task conditions, improving its flexible execution capabilities and motion stability in multi-position and multi-process scenarios, thereby meeting the high-efficiency, high-precision, and highly adaptable control requirements of modern intelligent manufacturing.

[0036] The method specifically comprises the following steps:

[0037] Encoding multiple robotic arm control behaviors into meme segments, each of which represents a control behavior module related to a specific task, and assigning a survival fitness to each meme segment. Based on the survival fitness, selection, crossover, and mutation operations are performed on the meme segments to generate a control behavior evolution map, which is used to guide the motion control execution of the robotic arm;

[0038] Establish a distributed semantic monitoring mechanism and construct a task semantic parameter map. The task semantic parameter map includes multiple semantic nodes and their relationships. Each semantic node corresponds to a type of working condition, task goal, or actuator state. The monitoring mechanism dynamically activates or disables the semantic node based on the current task state to generate a real-time working condition description of the current task.

[0039] Converting the real-time working condition description into a working condition embedding vector, and extracting the meme fragment most relevant to the current working condition from the control behavior evolution graph based on the similarity between the working condition embedding vector and the semantic embedding of each meme fragment, to form a candidate meme factor graph;

[0040] Constructing a path evaluation function based on the historical success rate, control cost, and failure rate of the control path in the candidate module factor graph, scoring multiple control paths, and selecting the control path with the highest score as the optimal control path;

[0041] The original control parameters are extracted from the optimal control path, and the control parameters are fine-tuned through a local perturbation mechanism to adapt to the accuracy and dynamic requirements of the current task. The fine-tuned control parameters are used as the final optimal motion control parameters and output to drive the motion control execution of the robotic arm.

[0042] Each step is described in detail below.

[0043] Step S10: encode multiple robotic arm control behaviors into meme fragments, each of which represents a control behavior module related to a specific task, and assign survival fitness to each of the meme fragments. Based on the survival fitness, selection, crossover and mutation operations are performed on the meme fragments to generate a control behavior evolution map, which is used to guide the motion control execution of the robotic arm.

[0044] To achieve rapid deployment and intelligent control of robotic arms in different mission scenarios, a mechanism is needed that can automatically generate the most appropriate control strategy based on mission requirements. Traditional robotic arm control methods often rely on fixed parameter configurations and preset motion processes, making them difficult to adapt to industrial environments with changing working conditions and frequent task switching. By introducing the concept of meme evolution, the robotic arm control behavior is abstracted into meme fragments with behavioral meaning and adaptability, and a control behavior evolutionary map is constructed. This allows for the structured expression and dynamic evolution of control behavior, providing the optimal control path for different tasks and automatically adjusting the control strategy as the environment changes, thereby improving the versatility, adaptability, and intelligence of the control system.

[0045] In the present invention, a meme fragment refers to a basic operation unit formed in the process of controlling behavior. Each meme fragment contains a type of action instruction, path planning strategy or execution constraint condition with independent function in a certain type of task, which can be understood as the smallest module unit with control significance.

[0046] Specifically, for example:

[0047] Meme clips for CNC machining loading and unloading operations may include:

[0048] Memetic Fragment A (Terminal Positioning Grab)

[0049] Task semantics: In the workpiece area after processing, the end effector of the robot arm is positioned to a fixed coordinate point and grabs the stationary workpiece.

[0050] Feature description: Use vision system to assist positioning, the end descends slowly, and the closing jaw action time is 0.6 seconds.

[0051] Meme segment B (oblique lift to avoid obstacles)

[0052] Task semantics: Extract from the workpiece area in a narrow space and quickly lift it up to avoid obstacles in the processing equipment.

[0053] Feature description: Based on the path planning algorithm, five oblique trajectories are generated, the middle turning point is equipped with speed transition control, and the lifting angle is greater than 45 degrees.

[0054] Memetic segments of the power screw assembly process might include:

[0055] Meme Segment C (Visual Aid Counterpoint)

[0056] Task semantics: Obtain the position of the assembly station based on the visual system and automatically correct the end deviation.

[0057] Feature description: Fusion of depth vision and image recognition data to adjust the end point X / Y / Z position to ±0.5mm accuracy.

[0058] Meme Clip D (Slow Twist Down)

[0059] Task semantics: To avoid backlash during screw tightening, the actuator contacts the target surface with a slow downward pressure.

[0060] Feature description: The downward pressure speed of the end Z axis is limited to 0.01m / s, triggering the torque monitor to start torque recording.

[0061] Memetic fragments from the laser engraving or marking process may include:

[0062] Meme Fragment E (Rectangular Area Outline Scan)

[0063] Task semantics: On a fixed platform, the end effector patrols the edge of the target area along a set rectangular trajectory for visual confirmation.

[0064] Feature description: The path is a closed curve consisting of four straight lines, with arc transitions between each line segment.

[0065] Meme Fragment F (Lifting and Lowering Synchronous Waiting)

[0066] Task semantics: When the laser engraving head is raised or lowered on the Y-axis, the robotic arm maintains its current position and waits for the processing signal to be completed.

[0067] Feature description: The robot arm enters the position holding mode and periodically polls the signal bus to confirm whether to continue.

[0068] Survival fitness is a quantitative indicator of the performance of a meme fragment in a specific task or environment, usually calculated based on factors such as its historical success rate, task completion efficiency, and control stability.

[0069] For example:

[0070] CNC cutting and grabbing action

[0071] Performance statistics (based on 100 historical calls):

[0072] Success rate S=96% (no falling off after grasping, accurate positioning)

[0073] Average completion time T = 1.2 seconds

[0074] Average end trajectory error E = 0.8 mm

[0075] Abnormal interruption times = 0

[0076] Fitness calculation (schematic):

[0077] F=αS+β(1-E_norm)-γ*T_norm

[0078] in:

[0079] E_norm=E / E_max=0.8 / 5=0.16 (error normalization)

[0080] T_norm=T / T_max=1.2 / 3=0.4 (time normalization)

[0081] α=0.5, β=0.3, γ=0.2

[0082] F=0.5×0.96+0.3×(1-0.16)-0.2×0.4=0.48+0.252-0.08=0.652

[0083] This meme fragment performs stably, with high accuracy and moderate speed, and has high adaptability, and can be retained into the next generation.

[0084] Another specific example is the end contact control during screw tightening:

[0085] Performance statistics:

[0086] Success rate S=99%

[0087] Average contact error E=0.2 mm

[0088] Average task completion time T = 2.8 seconds

[0089] Dynamic torque fluctuation peak = high, tightening is unstable

[0090] Fitness decision-making mechanism (combined with meme weight):

[0091] The system determines that the meme is stable in low-speed, high-precision tasks, but not suitable for high-beat scenarios, so it assigns a scenario weight coefficient W, defined as:

[0092] W_high_precision=1.0

[0093] W_high_speed=0.4

[0094] The final fitness is:

[0095] In high-precision tasks: F = W × S = 1.0 × 0.99 = 0.99

[0096] In high-speed scenarios: F = W × S = 0.4 × 0.99 = 0.396

[0097] This meme has obvious scene dependence, and its fitness value has task adaptation characteristics. It can be used preferentially in specific scenes and automatically avoided in other scenes.

[0098] In this paper, an evolutionary graph refers to a graph structure composed of a series of control behavior modules (meme fragments) and the connections between them. This structure not only records the connection sequence of meme fragments during multiple historical task executions, but also reflects the evolutionary path and adaptive weight of the control strategy. Compared with traditional static flow charts, evolutionary graphs have the ability to dynamically grow, self-adjust, and optimize paths. As tasks are executed, they can select, crossover, and mutate the combination of meme fragments to form a stable and efficient control strategy cluster.

[0099] In a specific example, Figure 2 As shown in the figure, the composition of the control behavior evolution map is shown using the laptop computer testing process as an example. This process includes the following common behavioral memes:

[0100] M1: Gripper aligns and grips the product

[0101] M2: Raise to standard operating height

[0102] M3: Move to the test platform

[0103] M4: Start test action and monitor status

[0104] M5: After completion, place it in the good product area

[0105] M6: Failed products are placed in the defective area

[0106] M7: Test exception, initiate retry action

[0107] M8: Timeout interrupt and initiate alarm process

[0108] Evolutionary paths formed in past missions include:

[0109] Normal process: M1->M2->M3->M4->M5

[0110] Abnormal branch 1: M4 fails -> M6

[0111] Abnormal branch 2: M4 timeout -> M7 -> M4 (retry) -> M5

[0112] Abnormal branch 3: M7 still fails -> M8

[0113] In this invention, selection, crossover, and mutation are the core mechanisms for evolving meme fragments and updating control behavior maps. This approach draws on the concept of genetic algorithms from biological evolution theory and incorporates the characteristics of control systems for a structured design. These three types of operations are used to continuously optimize the composition and order of meme fragments, gradually generating control paths that perform well under various operating conditions.

[0114] The goal of the selection operation is to retain high-performing segments from the existing collection of meme segments for subsequent behavioral graph construction or evolution operations. By setting a survival fitness threshold or using a proportional selection mechanism, the system prioritizes segments with high mission success rates, high control accuracy, and high execution efficiency, while gradually eliminating inefficient, unstable, or high-risk segments.

[0115] Implementation methods include:

[0116] Set the fitness threshold, for example, F ≥ 0.75;

[0117] Or use roulette wheel selection, tournament selection and other methods for probability screening;

[0118] The selected meme fragments will participate in building the next generation of control behavior evolution map.

[0119] In a specific example:

[0120] Among the 100 tasks, meme fragment M4 (test action) had a success rate of 95%, an error rate of 1%, a short execution time, and a survival fitness of 0.92, which was higher than the threshold of 0.75. Therefore, it was retained by the system and participated in subsequent evolution.

[0121] The crossover operation simulates the process of biological sexual reproduction, combining two or more highly adaptive meme fragments into a new composite fragment, attempting to achieve better control behavior through strategic reorganization. This operation supports information fusion between memes, enhancing the diversity and innovation of control strategies.

[0122] Implementation methods include:

[0123] Identify pairs of meme segments that frequently appear consecutively in historical tasks, such as M2 (lift) + M3 (move);

[0124] Construct a new segment M9, encoding the two behaviors as a joint execution segment, such as diagonal lifting movement;

[0125] New segments will have independent identification and initial fitness, and their actual performance will be evaluated during execution.

[0126] In a specific example:

[0127] Combining M4 (test action) and M7 (anomaly detection) into a composite meme M10 implements an integrated behavior of retrying immediately after automatically detecting an anomaly, simplifying the control process and reducing latency.

[0128] Mutation is used to introduce novel meme fragments to enhance the system's exploration capabilities and avoid being trapped in local optimal solutions. By randomly modifying the structure, control parameters, and perception mechanisms of existing meme fragments, new strategy candidates are generated. The system then decides whether to retain them based on their performance.

[0129] Implementation methods include:

[0130] Randomly select a meme segment, such as M3 (movement);

[0131] Modify its path strategy, for example, by adding a visual recognition module to form M11 (visually assisted positioning);

[0132] The new fragments generated by mutation will enter the control behavior map with lower fitness, and their value will be gradually verified through task performance;

[0133] Constraints can be added to prevent the system from introducing unexecutable or excessively risky segments.

[0134] In a specific example:

[0135] M3 performed poorly in multiple tasks (high error, spatial collisions). The system automatically introduced M11, combined its movement path with a visual recognition correction mechanism, to achieve dynamic path fine-tuning and gradually evaluate whether this strategy is better than the original solution.

[0136] Through the above three evolutionary mechanisms, the system has the following characteristics in the process of constructing and optimizing the control behavior map:

[0137] Selecting operations ensures that the system inherits high-performance strategies and has strong stability;

[0138] Cross-operation enables experience integration and portfolio innovation, enhancing strategy diversity;

[0139] Mutation operations promote the exploration of untried strategies and enhance the system's adaptive capabilities;

[0140] The behavioral map ultimately formed by the system continuously strengthens high-fitness paths, eliminates inefficient paths, and dynamically adapts to different tasks and working conditions.

[0141] These operations constitute the core basis for the formation and evolution of the evolutionary graph in the present invention, and are also the key points that distinguish it from traditional fixed process control systems.

[0142] In a complete and specific example, the above-mentioned notebook computer factory testing example is continued.

[0143] Evaluate the fitness of each meme segment based on the most recent 100 task execution history data:

[0144] M1: F=0.95 (good stability)

[0145] M2: F=0.92 (standard action)

[0146] M3: F=0.68 (occasionally collides in a small space)

[0147] M4: F=0.90 (more reliable)

[0148] M5: F=0.96 (high success rate)

[0149] M6: F=0.85 (successful obstacle avoidance but low efficiency)

[0150] M7: F=0.65 (average retry effect)

[0151] M8: F=0.40 (terminate the process, low frequency of use)

[0152] When performing an evolution operation:

[0153] Step 1: Select an action

[0154] Retain segments with meme F ≥ 0.85: M1, M2, M4, M5, M6;

[0155] Eliminate M8 (high alarm frequency and no performance improvement);

[0156] M3 and M7 are at the edge of fitness and enter the mutation and crossover stage.

[0157] Step 2: Crossover Operation

[0158] Cross 1: M2+M3->M9 (raise and move sideways)

[0159] The combined actions of M2 and M3 are often called consecutively;

[0160] The system combines them into an integrated trajectory to reduce attitude adjustment time;

[0161] M9: Motion path optimization, fitness to be evaluated.

[0162] Crossover 2: M4+M7->M10 (test with judgment retry)

[0163] Integrate test actions and exception judgment logic to form a composite segment;

[0164] M10 has a test-retry logic judgment function.

[0165] Step 3: Mutation Operation

[0166] Mutate M3 and introduce a fine-tuning strategy to form a new fragment:

[0167] M11 (visual-assisted positioning): A variation of M3, introducing image recognition to correct terminal deviations;

[0168] The initial fitness of this segment is set to a medium level (F = 0.7) and can be adjusted according to the execution performance.

[0169] Generate a control behavior evolution graph such as Figure 3 shown.

[0170] When performing a task, if the system detects a narrow space, it will prioritize the following path:

[0171] M1->M2->M9->M4->M5

[0172] If the positioning accuracy of the test module decreases, the system selects the following path:

[0173] M1->M2->M3->M11->M4->M5

[0174] If intermittent anomalies occur in the test history, enable the fused fragment path:

[0175] M1->M2->M3->M4->M5 / M6

[0176] In this step, the graph is not static reference data, but a real-time decision-making and execution guidance system that the system relies on in each cycle. By using the control behavior evolution graph as the core basis for actual control decisions, the present invention can free the robotic arm control system from the limitations of static program flow and achieve dynamic task analysis, adaptive path generation, and parameter-level control optimization. The graph contains behavioral modules, connection strategies, execution history, and strategy weights, giving the system a closed-loop capability of task understanding—behavior selection—parameter regulation—path correction, greatly improving the intelligence, flexibility, and scenario adaptability of the control system.

[0177] Step S20: establish a distributed semantic monitoring mechanism and construct a task semantic parameter map. The task semantic parameter map includes multiple semantic nodes and their association relationships. Each semantic node corresponds to a type of working condition, task goal or actuator status. The monitoring mechanism dynamically activates or disables the semantic node according to the current task status to generate a real-time working condition description of the current task.

[0178] In order to achieve dynamic matching between the robot control strategy and the working environment, the system needs to be able to perceive the current task conditions in real time and quickly adapt the control path. In actual industrial production processes, there are often significant differences in the types of tasks, workstation status, environmental conditions and equipment response states faced by the robot. Traditional control systems often rely solely on the direct acquisition of physical quantities, and are unable to form a structured understanding of task semantics, nor can they make precise adjustments based on the essential differences in tasks. The present invention establishes a distributed semantic monitoring mechanism, constructs a task semantic parameter map, and deeply integrates the control logic with the perception structure, so that the system can achieve task-oriented real-time working condition modeling, and support the graphical selection and meme scheduling of subsequent control paths.

[0179] In the present invention, distributed semantic monitoring refers to a network of monitoring devices arranged on multiple perception nodes, with task semantics as the driving goal, to collect, identify, associate and update task-related status information, including hardware sensors, edge computing nodes and controller communication units.

[0180] In a specific example, the task goal is: the robotic arm grasps the laptop computer and places it on the test platform, completes a series of automated tests, and places it according to the test results.

[0181] The operation area includes the following typical sensing nodes:

[0182] Robotic arm body (gripper, joint, motor)

[0183] Test station (test platform, data interface)

[0184] Environmental sensing area (temperature, humidity, oil mist, light, etc.)

[0185] The corresponding relationship between the monitoring unit distribution and task semantic nodes is shown in Table 1.

[0186] Table 1 Correspondence between monitoring unit distribution and task semantic nodes

[0187] The task semantic parameter graph refers to a working condition knowledge network organized in a graph structure. The nodes in the graph represent abstract semantic units, such as high work intensity, decreased visual accuracy, and high grasping failure rate. The edges represent the logical relationship or causal association between nodes.

[0188] The types of semantic nodes include:

[0189] Working condition semantic nodes: such as severe oil mist interference and high temperature;

[0190] Task target semantic nodes: such as the requirement for high-precision placement of test beats;

[0191] Actuator status semantic node: such as gripping failure and frequent end offset.

[0192] The association relationship is the logical connection between semantic nodes, including:

[0193] Cause and effect (e.g., severe oil mist interference -> blurred image recognition);

[0194] Collaborative relationships (e.g. unstable grasping ∧ blurred image -> increased need for path adjustment);

[0195] Inferring relationships (e.g., the necessity of inferring higher-level behavioral strategies from multiple lower-level states).

[0196] In a specific example, following the aforementioned example of the laptop computer automatic testing process, the semantic factors that affect the robot arm control decision in this process may include:

[0197] Is the work pace tight?

[0198] Whether the gripper has repeated failures;

[0199] Test whether the platform responds slowly;

[0200] Whether the current environment is a high concentration of oil mist;

[0201] Are there difficulties with image recognition?

[0202] Whether test exceptions occur continuously;

[0203] Whether it is in a low-light environment at night;

[0204] Whether higher grasping stability is required;

[0205] Whether to call the visual aid function.

[0206] There are certain causal or synergistic relationships between these semantic nodes. For example, image recognition difficulties may be caused by high oil mist concentration or low light environment, and test abnormalities may be related to unstable grippers or workpiece misalignment.

[0207] like Figure 4 As shown, a task semantic parameter map is formed in a specific example.

[0208] The task semantic parameter graph, comprising multiple semantic nodes and their relationships, enables the control system to not only identify anomalies in a specific physical state but also integrate multiple factors to make semantically combined judgments, thereby triggering more precise and context-sensitive memetic path selection. This graph, characterized by its abstractness, structure, and dynamic nature, is a key support structure for the system's intelligent control adaptation.

[0209] The activation and deactivation of semantic nodes in the present invention are not the traditional opening and closing of data, but refer to:

[0210] Activating a semantic node means that the system determines that a specific task semantic state has occurred or satisfied the conditions in the current task condition based on perception information, rule reasoning, or statistical calculation results, and marks the corresponding semantic node as active, so that it participates in the calculation and strategy judgment of subsequent control paths in the graph;

[0211] Disabling a semantic node means that the system determines that a certain semantic state has not occurred or has no judgment significance, and temporarily blocks the node from the graph, and does not use it as a basis for path construction or parameter adjustment of the control graph.

[0212] The activation state of the semantic nodes in the graph essentially constitutes the semantic vector or context description of the current task, which acts like the activation function in a neural network to control which features participate in the current decision-making process.

[0213] In the implementation, let the graph node set be S={s1,s2,...,s n}, each node s i ∈S corresponds to a semantic state.

[0214] Define a mapping function A(t): S->{0,1}, which indicates whether a semantic node is activated at time t:

[0215] If A(s i ,t)=1, indicating that node s i be activated;

[0216] If A(s i ,t)=0, indicating that node s i is disabled.

[0217] The state of the entire graph at time t can be expressed as a vector V(t)=[A(s1,t),A(s2,t),...,A(s n ,t)].

[0218] In a specific example,

[0219] During the automatic testing process for laptop computers, clamping and placing operations are being performed.

[0220] Sensor input:

[0221] The gripping force at the end of the gripper is lower than the set threshold;

[0222] The oil mist concentration is 360μg / m³;

[0223] The image recognition edge confidence is 76%;

[0224] The light intensity fluctuation range was 22%, exceeding the ±15% upper limit;

[0225] The last two rounds of testing resulted in bit errors and failed.

[0226] Semantic mapping results:

[0227] Crawl stability decreased -> activated;

[0228] ImageRecognitionBlur->Activate;

[0229] Oil mist interference is serious->activate;

[0230] External light interference abnormality->activate;

[0231] Test abnormally frequently -> activate;

[0232] Path fine-tuning requirements are high -> derived from the previous state -> activated;

[0233] Test interface abnormal -> not triggered -> disabled;

[0234] Mission beat is tight -> not triggered -> disabled.

[0235] The semantic vector formed by the system at this time:

[0236] V(t)=[1,1,1,1,1,1,0,0]

[0237] In the control behavior evolution graph, the system limits the control path to include:

[0238] Visual Assist Fine-tuning

[0239] Low-speed path adjustment

[0240] Capture redundant confirmation

[0241] Subgraphs of isometic fragments,executing adaptive control strategies.

[0242] By activating and deactivating semantic nodes, the task semantic parameter map is no longer a static description, but a real-time, dynamically structured working condition recognition model. This mechanism ensures that the system can make precise and context-aware control decisions based on the actual task status.

[0243] Step S30, converting the real-time working condition description into a working condition embedding vector, and extracting the meme fragment most relevant to the current working condition from the control behavior evolution graph based on the similarity between the working condition embedding vector and the semantic embedding of each meme fragment, to form a candidate meme factor graph.

[0244] To enable adaptive control strategy selection for the robotic arm in different task scenarios, the system needs to convert the real-time perceived working state into a structured representation that can be used for graph search and meme screening. Because semantic nodes are typically discrete activation signals or labels and cannot be directly used to measure similarity, an embedding method is used to map the working state description into a representation in a continuous vector space. This is then compared with the semantic embeddings of each meme fragment in the control behavior evolution graph to identify the set of control strategies that best matches the current task. This mechanism essentially establishes a connection between the working state semantic space and the meme behavior space, and is a key step in achieving a closed-loop working state-behavior mapping.

[0245] In this paper, a working condition embedding vector refers to the system's analysis of the currently activated semantic nodes in the task's semantic parameter map when executing a specific task. These discrete semantic state combinations are converted into a set of continuous numerical vectors with vector space positional characteristics, forming a comprehensive, structured semantic representation of the current task environment. This vector then serves as input for meme screening and control path inference, enabling context-based adaptive control strategy generation.

[0246] In a specific example, during a test round, the system detects the following status:

[0247] The gripper has poor stability during grasping;

[0248] The image recognition module feedback indicates that the recognition edge fuzziness is high;

[0249] The oil mist interference value exceeds the set threshold during mission execution;

[0250] Path adjustment requests are frequent.

[0251] After the semantic monitoring mechanism, the system identifies and activates the following semantic nodes:

[0252] Decreased crawling stability

[0253] Image recognition is fuzzy

[0254] Serious oil mist interference

[0255] Path fine-tuning requires high

[0256] Each semantic node predefines the embedding weights on four control semantic dimensions, such as:

[0257] The decrease in grasping stability can be represented as a vector [0.9, 0.1, 0.3, 0.1], where

[0258] Grasping reliability (0.9): This state seriously affects the stability of the end-grasping action;

[0259] Visual interference (0.1): only slightly affects visual recognition;

[0260] Path Strategy (0.3): Minor path adjustments may be required to avoid multiple attempts;

[0261] System load (0.1): Low requirements for dynamic updates of control parameters.

[0262] The image recognition fuzzy map is a vector [0.2, 0.9, 0.6, 0.2], where:

[0263] Grasping reliability (0.2): Image blur may indirectly affect the grasping action;

[0264] Visual interference (0.9): Severe interference with the visual perception system, with blurred edge recognition;

[0265] Path strategy (0.6): affects vision-based trajectory generation;

[0266] System load (0.2): May trigger a reset of visual parameters or enable auxiliary algorithms.

[0267] The oil mist interference is severely mapped to [0.3, 0.7, 0.5, 0.3], where:

[0268] Grasping reliability (0.3): Oil mist lubricates the surface of the fixture or object, slightly affecting the grasping;

[0269] Visual interference (0.7): Oil mist causes image blur and affects recognition;

[0270] Path strategy (0.5): may trigger path fine-tuning to adapt to fuzzy recognition;

[0271] System load (0.3): Filters, infrared assistance, and other countermeasures need to be enabled to increase the frequency of system parameter adjustments.

[0272] Path fine-tuning requires a high mapping of [0.4, 0.3, 0.8, 0.5], where:

[0273] Grasping reliability (0.4): Multiple path iterations may affect the grasping posture;

[0274] Visual interference (0.3): Path changes have a slight impact on the visual system;

[0275] Path strategy (0.8): highly dependent on local trajectory updates;

[0276] System load (0.5): The system needs to frequently fine-tune control instructions, and the control load is high.

[0277] The system then combines these four embedding vectors and uses an averaging strategy to form the current working state embedding vector for the task:

[0278] Adding the vectors by dimension yields: [1.8, 2.0, 2.2, 1.1]

[0279] Take the average value of the four nodes: [0.45, 0.5, 0.55, 0.275]

[0280] The final embedding vector of the current working condition is: [0.45, 0.5, 0.55, 0.275].

[0281] The semantic embedding of a meme fragment refers to the semantic vector representation established in advance for each meme fragment. This vector is generated by analyzing information such as the task scenario applicable to the meme fragment, the corresponding typical semantic state, and historical performance characteristics.

[0282] In a specific example, the meme is adaptively adjusted for gripper pressure.

[0283] This meme fragment represents a control strategy: when the system detects that the grasping stability decreases or the object slip increases, it automatically fine-tunes the tension control parameters of the gripper by analyzing the torque feedback of the end effector and the contact surface state to improve the gripping stability and avoid physical damage.

[0284] Its applicable task scenarios include:

[0285] Grasping stability is reduced;

[0286] The fixture feedback force is unstable;

[0287] Precision parts handling;

[0288] Increased repeat crawl failure rate.

[0289] Typical activation semantic nodes:

[0290] Grasping stability is reduced;

[0291] Path fine-tuning requirements are high (minor);

[0292] The test interface is difficult to access (weakly related);

[0293] Historical execution characteristics (from system task database statistics):

[0294] High success rate (94%);

[0295] The number of parameter adjustments is moderate;

[0296] Low control resource usage;

[0297] Short response delay (suitable for fast grasping tasks);

[0298] Average performance in scenes with strong visual interference (relies on force control, not image recognition).

[0299] Similar to the working condition embedding vector, the semantic embedding is a four-dimensional vector corresponding to the following four control semantic dimensions:

[0300] Dimension 1: the strength of the impact of grasp reliability;

[0301] Dimension 2: Visual interference responsiveness;

[0302] Dimension 3: degree of path strategy adaptation;

[0303] Dimension 4: System control load level.

[0304] The semantic embedding vector of this meme fragment is constructed as follows:

[0305] [0.95,0.15,0.35,0.25]

[0306] in:

[0307] 0.95 (grasping reliability): The main purpose of this meme is to enhance gripping stability, and the effect is significant;

[0308] 0.15 (visual interference): does not rely on image data, works only through force sensor feedback, and has weak response to visual interference;

[0309] 0.35 (Path Strategy): Has a certain degree of path adaptation capability, but mainly relies on tension compensation and does not dominate path planning;

[0310] 0.25 (system load): The control strategy is simple, the system resource consumption is low, and it is suitable for resource-constrained environments.

[0311] After determining the embedding vector, the working condition embedding vector formed by the current task semantic state is used as an input feature and matched against all meme segments with predefined semantic embeddings. By calculating the similarity function between the embedding vectors, the set of meme segments that best fit the current working condition is identified. The meme segments in this set form a directed subgraph structure, called the candidate meme graph, which is used to guide the evaluation and selection of subsequent control paths.

[0312] To illustrate this process more intuitively, the following is an example of a specific task.

[0313] During the automated testing of laptop computers, a robotic arm must remove the device under test from a tray and precisely place it on the test interface. During a particular task cycle, the system identified the following issues: decreased gripper stability, blurred visual recognition images, moderate trajectory adjustment requirements, and average control system load. Based on this state information, the system activated semantic nodes and calculated the following working condition embedding vector:

[0314] The current working condition embedding vector is: [0.8, 0.2, 0.4, 0.3]

[0315] The four dimensions in this vector represent the weights of the task in terms of grasping reliability, visual interference, path strategy adjustment, and control system load.

[0316] Several meme fragments have been defined in the control behavior evolution graph. Each meme fragment has its own semantic embedding vector, for example:

[0317] Meme A is the adaptive adjustment of gripper tension, and its semantic embedding is [0.95, 0.15, 0.35, 0.25];

[0318] Meme B is visual positioning compensation, and the semantic embedding is [0.3, 0.9, 0.7, 0.4];

[0319] Meme C is a path buffer smoothing strategy with semantic embedding of [0.4, 0.3, 0.85, 0.2];

[0320] Meme D is the environment switching controller, and its semantic embedding is [0.6, 0.5, 0.7, 0.9].

[0321] The system calculates cosine similarity or Euclidean distance between the working condition embedding and the semantic embeddings of each meme. The results show that meme A has the highest similarity to the current working condition, followed by meme C, with memes B and D showing a lower degree of match. A similarity threshold (for example, 0.8) is set, and the system selects memes A and C as candidate memes. The system then further examines their connections in the evolutionary graph. If meme A can naturally transition to meme C during actual execution, a directed candidate meme graph is formed.

[0322] This subgraph represents the control behavior path that the system should prioritize under the current working conditions. It begins with adaptive gripper tension adjustment as the starting meme, connected to the path buffering and smoothing strategy to form a control sequence. This sequence is evaluated in the subsequent path scoring phase, ultimately selecting the control parameters to drive the robot arm to perform the current task.

[0323] The advantage of this mechanism is that it not only achieves contextual alignment between meme behavior selection and the current task semantic state, but also significantly compresses the control path search space by constructing a candidate meme factor graph, improving the real-time and accuracy of control strategy generation, thereby enhancing the system's adaptive control capability under complex and changeable working conditions.

[0324] Step S40 , constructing a path evaluation function based on the historical success rate, control cost and failure rate of the control path in the candidate module factor graph, scoring the multiple control paths, and selecting the control path with the highest score as the optimal control path.

[0325] When multiple feasible control paths exist within the candidate meme graph, the system can scientifically and quantitatively evaluate and screen each path based on historical execution performance and current task requirements. This ensures that the selected path not only has excellent historical performance but also offers low cost and minimal risk, thereby improving the overall stability, efficiency, and robustness of the control system in multi-task scheduling scenarios. Because the execution of each meme fragment affects aspects such as grasping accuracy, system load, and execution stability, a path evaluation function must be used to comprehensively evaluate each control path in multiple dimensions to select the optimal control path.

[0326] In an optional specific implementation, step S40 specifically includes the following sub-steps:

[0327] Step S401: Obtain all reachable control paths in the candidate meme graph, traverse the path set, and identify the meme sequence structure in each path.

[0328] In step S402 , based on the historical task database, the success rate, control cost, and failure rate records corresponding to each path in past tasks are extracted, and these records are standardized so that data of different dimensions can be uniformly evaluated.

[0329] Step S403: construct a path evaluation function. The evaluation function may use a weighted comprehensive scoring model, for example:

[0330] Score=α×SuccessRate-β×Cost-γ×FaultRate

[0331] Score represents the path's scoring result, SuccessRate represents the historical success rate, Cost represents the control cost, and FaultRate represents the failure rate. α, β, and γ are system-preset weighting coefficients used to balance the importance of these three indicators in different application scenarios. These parameters can be dynamically optimized through empirical settings or online learning.

[0332] Step S404 : Score all paths and sort them according to the score results. The path with the highest score is selected as the optimal control path to generate the final control parameter sequence.

[0333] The advantage of the above-mentioned path evaluation mechanism is that it introduces a comprehensive scoring model based on historical behavior and current resource status, which can take into account reliability, efficiency and risk control when selecting paths, thereby effectively avoiding the system instability problem caused by a single optimal performance path, and improving the practicality and scenario generalization ability of the robot arm control strategy.

[0334] The following describes the candidate meme factor graph example from the aforementioned laptop computer automated testing process. In step S30 , the system selects the adaptive gripper tension adjustment strategy for meme A and the path buffering and smoothing strategy for meme C based on the working condition embedding vector, forming the following candidate meme factor graph: Meme A -> Meme C.

[0335] Assume that there are three reachable paths:

[0336] Path P1: Meme A

[0337] Path P2: Meme A -> Meme C

[0338] Path P3: Meme C

[0339] The system extracts the following indicator data from historical task records:

[0340] The success rate of path P1 is 0.92, the cost is 0.4, and the failure rate is 0.02

[0341] The success rate of path P2 is 0.88, the cost is 0.3, and the failure rate is 0.01

[0342] The success rate of path P3 is 0.78, the cost is 0.2, and the failure rate is 0.03.

[0343] Set α=0.6, β=0.3, γ=0.1, and substitute into the evaluation function:

[0344] Score(P1)=0.6×0.92-0.3×0.4-0.1×0.02=0.552-0.12-0.002=0.43

[0345] Score(P2)=0.6×0.88-0.3×0.3-0.1×0.01=0.528-0.09-0.001=0.437

[0346] Score(P3)=0.6×0.78-0.3×0.2-0.1×0.03=0.468-0.06-0.003=0.405

[0347] The path with the highest score is path P2, that is, meme A->meme C. Based on this, the system selects this path as the optimal control path for the current task, which is used to drive the robotic arm to complete the task at this stage.

[0348] In step S50, the original control parameters are extracted from the optimal control path, and the control parameters are fine-tuned through a local perturbation mechanism to adapt to the accuracy and dynamic requirements of the current task. The fine-tuned control parameters are used as the final optimal motion control parameters and output to drive the motion control execution of the robotic arm.

[0349] Step S50 is designed to further improve the system's accuracy and stability in executing tasks in a dynamic environment. Although the optimal control path has been selected using the path evaluation function, the control parameters contained therein are typically empirical values drawn from historical tasks and cannot fully adapt to slight changes in the current operating conditions. To this end, a local perturbation mechanism is introduced in this step. While retaining the original control path structure, the control parameters are fine-tuned. This ensures that the resulting control instructions have greater task adaptability and dynamic response capabilities, ensuring that the robot arm completes its operational tasks with high precision and low risk in real-time conditions.

[0350] In an optional specific implementation, step S50 specifically includes the following sub-steps:

[0351] Step S501: extract control parameters from the optimal control path node by node to construct an original control parameter set. The parameter set may include multiple control dimensions, such as posture target, actuator output, clamping strength, trajectory time window, etc.

[0352] In step S502 , a disturbance range and a disturbance step are set for each control parameter. The disturbance range can be set based on the task type or historical execution data. For example, the disturbance range of the gripper tension can be ±10%, and the disturbance step can be 1%.

[0353] Step S503 : generating a plurality of disturbance samples based on the disturbance rule, each sample corresponding to a set of complete control parameter combinations for candidate evaluation.

[0354] Step S504 , performing a rapid simulation evaluation of the disturbance sample or a small-scale actual execution trial to obtain the performance indicators of each set of parameters under the target working conditions. The evaluation indicators may include execution accuracy, completion time, load balancing, etc.

[0355] Step S505 : selecting a set of control parameters with the best performance as final motion control parameters according to the evaluation result, and outputting them to the execution control unit to drive the robot arm to move.

[0356] The advantage of the above mechanism is that it breaks the static dependence on preset parameters, allowing the system to perform perturbation-based multi-strategy verification and optimization before task execution. It has strong environmental adaptability and robustness, can significantly reduce the control offset caused by parameter errors, and improve the overall response quality and execution accuracy of the system.

[0357] The following explanation continues with the example of the aforementioned laptop computer automated testing process. In step S40, the system selects the optimal control path consisting of adaptive gripper tension adjustment and path buffering and smoothing strategies. The system extracts the original control parameters from this path, including gripper tension of 1.2N, path transition speed of 0.4m / s, and transition time of 0.6s.

[0358] After entering step S50, the system sets a disturbance range of ±10% and a step size of 0.05N for the gripper tension, and a disturbance range of ±15% and a step size of 0.02m / s for the path speed, forming several parameter combinations, for example:

[0359] Combination 1: [1.1N, 0.38m / s, 0.6s]

[0360] Combination 2: [1.2N, 0.4m / s, 0.58s]

[0361] Combination 3: [1.25N, 0.42m / s, 0.62s]

[0362] Combination 4: [1.3N, 0.36m / s, 0.6s]

[0363] The system evaluated the impact of these combinations on the docking accuracy of the laptop test interface through local simulation, and finally found that combination 2 had both clamping stability and path transition smoothness during execution. Therefore, this combination was selected as the optimal motion control parameter and output to the execution module to drive the robotic arm to complete this task cycle.

[0364] In another embodiment, the present invention further provides a robotic arm control system based on adaptive regulation, comprising:

[0365] a meme segment generation module, configured to encode a plurality of robotic arm control behaviors into meme segments, each of the meme segments representing a control behavior module associated with a specific task, and assigning a survival fitness to each of the meme segments;

[0366] a meme evolution construction module, configured to perform selection, crossover, and mutation operations on meme segments based on the survival fitness, and generate a control behavior evolution map, wherein the control behavior evolution map is used to guide motion control execution of the robotic arm;

[0367] A semantic monitoring module is used to establish a distributed semantic monitoring mechanism and construct a task semantic parameter map. The task semantic parameter map includes multiple semantic nodes and their associations. Each semantic node corresponds to a type of working condition, task goal, or actuator state. The monitoring mechanism dynamically activates or disables the semantic node according to the current task state to generate a real-time working condition description of the current task.

[0368] a meme selection module, configured to convert the real-time working condition description into a working condition embedding vector, and extract the meme fragment most relevant to the current working condition from the control behavior evolution graph based on the similarity between the working condition embedding vector and the semantic embedding of each meme fragment, to form a candidate meme factor graph;

[0369] a path evaluation module, configured to construct a path evaluation function based on the historical success rate, control cost, and failure rate of the control path in the candidate module factor graph, score the multiple control paths, and select the control path with the highest score as the optimal control path;

[0370] A parameter fine-tuning module is used to extract the original control parameters from the optimal control path and fine-tune the control parameters through a local perturbation mechanism to adapt to the accuracy and dynamic requirements of the current task. The fine-tuned control parameters are used as the final optimal motion control parameters and output to drive the motion control execution of the robotic arm.

[0371] It should be noted that the explanation of the above-mentioned embodiment of the robotic arm control method based on adaptive adjustment is also applicable to the device of the embodiment of the present application and will not be repeated here.

[0372] Those skilled in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented using a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0373] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0374] In the several embodiments provided in this application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of this application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), magnetic disk or optical disk, and other media that can store program code.

[0375] The above is only a specific embodiment of the present application. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in this application, which should be included in the scope of protection of this application. For some module structures that are not particularly clear in the present invention, the content recorded in the prior art shall prevail. The prior art mentioned in the above background technology section and the specific embodiment section of the present invention can be regarded as part of the present invention and is used to understand the meaning of some technical features or parameters.

Claims

1. A robotic arm control method based on adaptive regulation, characterized in that: The method comprises the following steps: Encoding multiple robotic arm control behaviors into meme segments, each of which represents a control behavior module related to a specific task, and assigning a survival fitness to each meme segment. Based on the survival fitness, selection, crossover, and mutation operations are performed on the meme segments to generate a control behavior evolution map, which is used to guide the motion control execution of the robotic arm; Establish a distributed semantic monitoring mechanism and construct a task semantic parameter map. The task semantic parameter map includes multiple semantic nodes and their relationships. Each semantic node corresponds to a type of working condition, task goal, or actuator state. The monitoring mechanism dynamically activates or disables the semantic node based on the current task state to generate a real-time working condition description of the current task. Converting the real-time working condition description into a working condition embedding vector, and extracting the meme fragment most relevant to the current working condition from the control behavior evolution graph based on the similarity between the working condition embedding vector and the semantic embedding of each meme fragment, to form a candidate meme factor graph; Constructing a path evaluation function based on the historical success rate, control cost, and failure rate of the control path in the candidate module factor graph, scoring multiple control paths, and selecting the control path with the highest score as the optimal control path; The original control parameters are extracted from the optimal control path, and the control parameters are fine-tuned through a local perturbation mechanism to adapt to the accuracy and dynamic requirements of the current task. The fine-tuned control parameters are used as the final optimal motion control parameters and output to drive the motion control execution of the robotic arm.

2. The method for controlling a robotic arm based on adaptive adjustment according to claim 1, characterized in that: The meme fragment refers to the basic operation unit formed in the process of controlling behavior. Each meme fragment contains action instructions, path planning strategies or execution constraints with independent functions in a certain type of task. It is the smallest module unit with control significance in the process of robotic arm control.

3. The method for controlling a robotic arm based on adaptive adjustment according to claim 1, characterized in that: The survival fitness is a quantitative indicator of the performance of a meme fragment in a specified task or environment, and is calculated by normalizing its historical success rate, task completion efficiency, and control stability factors.

4. The method for controlling a robotic arm based on adaptive adjustment according to claim 1, characterized in that: The evolutionary map is a graph structure composed of a series of meme fragments and the connection relationships between them. The evolutionary map records the connection sequence of meme fragments during multiple task executions in history, reflecting the evolutionary path and adaptability weight of the control strategy.

5. The method for controlling a robotic arm based on adaptive adjustment according to claim 1, characterized in that: The task semantic parameter graph refers to a working condition knowledge network organized in a graph structure, where nodes in the graph represent abstract semantic units and edges represent logical relationships or causal associations between nodes.

6. A robotic arm control system based on adaptive regulation, characterized in that: The system includes the following modules: a meme segment generation module, configured to encode a plurality of robotic arm control behaviors into meme segments, each of the meme segments representing a control behavior module associated with a specific task, and assigning a survival fitness to each of the meme segments; a meme evolution construction module, configured to perform selection, crossover, and mutation operations on meme segments based on the survival fitness, and generate a control behavior evolution map, wherein the control behavior evolution map is used to guide motion control execution of the robotic arm; A semantic monitoring module is used to establish a distributed semantic monitoring mechanism and construct a task semantic parameter map. The task semantic parameter map includes multiple semantic nodes and their associations. Each semantic node corresponds to a type of working condition, task goal, or actuator state. The monitoring mechanism dynamically activates or disables the semantic node according to the current task state to generate a real-time working condition description of the current task. a meme selection module, configured to convert the real-time working condition description into a working condition embedding vector, and extract the meme fragment most relevant to the current working condition from the control behavior evolution graph based on the similarity between the working condition embedding vector and the semantic embedding of each meme fragment, to form a candidate meme factor graph; a path evaluation module, configured to construct a path evaluation function based on the historical success rate, control cost, and failure rate of the control path in the candidate module factor graph, score the multiple control paths, and select the control path with the highest score as the optimal control path; A parameter fine-tuning module is used to extract the original control parameters from the optimal control path and fine-tune the control parameters through a local perturbation mechanism to adapt to the accuracy and dynamic requirements of the current task. The fine-tuned control parameters are used as the final optimal motion control parameters and output to drive the motion control execution of the robotic arm.

7. The adaptive adjustment-based robotic arm control system according to claim 6, characterized in that: The meme fragment refers to the basic operation unit formed in the process of controlling behavior. Each meme fragment contains action instructions, path planning strategies or execution constraints with independent functions in a certain type of task. It is the smallest module unit with control significance in the process of robotic arm control.

8. The adaptive adjustment-based robotic arm control system according to claim 6, characterized in that: The survival fitness is a quantitative indicator of the performance of a meme fragment in a specified task or environment, and is calculated by normalizing its historical success rate, task completion efficiency, and control stability factors.

9. The adaptive adjustment-based robotic arm control system according to claim 6, characterized in that: The evolutionary map is a graph structure composed of a series of meme fragments and the connection relationships between them. The evolutionary map records the connection sequence of meme fragments during multiple task executions in history, reflecting the evolutionary path and adaptability weight of the control strategy.

10. The adaptive adjustment-based robotic arm control system according to claim 6, characterized in that: The task semantic parameter graph refers to a working condition knowledge network organized in a graph structure, where nodes in the graph represent abstract semantic units and edges represent logical relationships or causal associations between nodes.

Images