Control method and system for universal mobile mechanical arm composite robot

Abstract

The invention discloses a universal mobile mechanical arm composite robot control method and system, and the method comprises the steps: carrying out the matching of a corresponding operation process template according to a judged current operation scene type, and loading a default operation parameter and a safety control threshold value in a current operation scene from a system database; by taking a fixed time window as a unit, a driving path of a movable chassis of the composite robot and an action posture of a mechanical arm are dynamically planned, and a safety control rule is executed in real time to monitor an operation process; after the job task is finished, performing batch updating on the job parameters based on the recorded data; when detecting that the network communication quality becomes poor or equipment breaks down, automatically switching to a local safe operation mode; and after the fault is recovered, the task execution record and the action track record are compared, and the unfinished operation task is continued. The problems that the operation process is fixed, multi-scene automatic identification and strategy switching are lacked, and skill optimization depends on manual parameter adjustment are solved.

Description

Technical Field

[0001] This invention belongs to the field of robot control technology, specifically relating to a general-purpose mobile robotic arm composite robot control method and system. Background Technology

[0002] In the fields of intelligent manufacturing and IoT applications, mobile robotic arms are widely used in material handling, workpiece assembly, and scene inspection due to their ability to integrate mobility and manipulation functions. However, existing mobile robotic arm solutions struggle to meet the demands for versatility, intelligence, and high efficiency in practical applications.

[0003] The existing mobile robotic arms have largely "hard-coded" operation processes and parameters. They are designed with fixed gripping paths and forces for workpieces of specific sizes and fixed movement speeds for specific channel widths. When scene elements change, they cannot automatically recognize these changes and require manual recalibration of process parameters, leading to long downtime. This inefficiency is particularly pronounced in flexible production scenarios involving multiple batches and product categories. Current solutions rely heavily on human experience for skill optimization. For example, when the robotic arm's gripping success rate decreases, technicians need to repeatedly experiment to adjust parameters such as gripping angle and force; when the mobile chassis path deviation increases, the path algorithm needs to be manually replanned. This not only consumes significant manpower but also limits optimization effectiveness to human experience, making continuous and precise skill improvement difficult. Furthermore, data generated during operation, such as success rate, time consumption, and energy consumption, are not effectively utilized, failing to form a data-driven optimization loop. This results in the robot's operational skills remaining at a fixed level, making it difficult to adapt to increasing scene complexity. Summary of the Invention

[0004] To address these issues, this invention provides a general-purpose mobile robotic arm composite robot control method and system, which solves the problems of fixed work processes, lack of automatic scene recognition and strategy switching, lack of adaptive learning capabilities, and reliance on manual parameter tuning for skill optimization.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a general-purpose mobile robotic arm composite robot control method, comprising the following steps:

[0006] S1. Read the preset task objectives, on-site environmental elements, and system resource constraints, and determine the current operation scenario type based on multimodal perception data;

[0007] S2. Match the corresponding work process template according to the determined current work scenario type, and load the default work parameters and safety control thresholds for the current work scenario from the system database.

[0008] S3. Using a fixed time window as a unit, dynamically plan the travel path of the mobile chassis of the composite robot and the movement posture of the robotic arm, and execute safety control rules in real time to monitor the operation process.

[0009] S4. After the task is completed, record the success or failure status of the task execution, the total time consumed, and the action deviation data, and update the task parameters in batches based on the recorded data;

[0010] S5. When a network communication quality deterioration or equipment failure is detected, automatically switch to local safe operation mode; after the fault is recovered, continue the unfinished work tasks by comparing the task execution record and the action trajectory record.

[0011] As a preferred control method for a general-purpose mobile robotic arm, in step S1, during the scene recognition process based on multimodal perception data to determine the current work scene type, a cosine similarity algorithm is used to calculate the similarity between the input features and the scene template features. The formula is:

[0012]

[0013] in, The input feature vector includes task target features, material property features, on-site obstacle features, channel features, time window features, and equipment capability features; These are scene template feature vectors, stored in the system policy library; For vectors and cosine similarity, when If the value exceeds a preset threshold, the input scene is determined to match the scene template.

[0014] The formula for fusing multimodal sensing data is:

[0015]

[0016] In the formula, To fuse feature vectors, To integrate weights, the weights are dynamically adjusted based on the complexity of the scenario. For visual feature vectors, Language feature vectors; based on fused feature vectors Complete task recognition and scene type determination, and map high-level intents into paths, postures and primitive commands.

[0017] As a preferred control method for a general-purpose mobile robotic arm, step S2 further includes adjusting the operation parameters using a fuzzy control algorithm. The formula for calculating the output of the fuzzy inference rule is as follows:

[0018]

[0019] In the formula, The final output parameters for the job. The number of fuzzy rules activated. For the first The membership degree of a fuzzy rule. For the first Output parameter values ​​corresponding to the fuzzy rules; parameter initialization is completed based on the adjusted job parameters.

[0020] As a preferred control method for a general-purpose mobile robotic arm, in step S3, during the dynamic planning of the mobile chassis's travel path and the robotic arm's posture, a model predictive control algorithm is used to plan the path and posture. The objective function is:

[0021]

[0022] In the formula, The objective function value is used to comprehensively evaluate the optimization effect of path attitude planning; To control the incremental sequence, including the speed adjustment of the moving chassis and the angle adjustment of each joint of the robotic arm; To predict the time domain length, i.e., the number of fixed time windows to be covered forward in each planning iteration; for Time prediction The system output includes the actual travel path deviation of the mobile chassis and the actual attitude deviation of the robotic arm's end effector. for The reference output at any given time is the preset expected travel path parameters and expected robotic arm posture parameters; , , These are respectively the path tracking accuracy weight, the control incremental smoothing weight, and the energy consumption optimization weight; for Time prediction The system's energy consumption value at any given time.

[0023] As a preferred solution for the control method of a general-purpose mobile robotic arm composite robot, step S3 involves using a distance detection algorithm to calculate the real-time distance between the robotic arm and the obstacle. The distance calculation formula is as follows:

[0024]

[0025] In the formula, The distance between the end effector of the robotic arm and the obstacle in three-dimensional space. The real-time coordinates of the robotic arm's end effector. Let the coordinates be those of the obstacle; when When the distance is less than the preset minimum safe distance, a warning or emergency stop command will be triggered.

[0026] As a preferred control method for a general-purpose mobile robotic arm, in step S4, the operation parameters are updated in batches based on the recorded data using a reinforcement learning model. The reinforcement learning model employs a temporal difference learning algorithm to update the parameters of key action primitives, and the value function update formula is as follows:

[0027]

[0028] In the formula, for Moment State The value function, for Moment State The value function, The learning rate is used to control the step size for parameter updates. From state Transfer to Instant rewards received This is a discount factor used to weigh immediate rewards against future rewards;

[0029] The parameter update also optimizes the parameters of the key action primitives using the gradient descent algorithm. The parameter update formula is as follows:

[0030]

[0031] In the formula, For the updated parameters, The parameters before the update. This is the gradient descent step size. For loss function right gradient, Success rate, time consumption, energy consumption, and deviation calculations based on action primitives.

[0032] As a preferred solution for the control method of a general-purpose mobile robotic arm composite robot, in step S4, during the batch update of operation parameters based on recorded data, the recorded operation data and equipment health data are filtered for valid data using a sliding window algorithm. The formula for determining the validity of data within the window is as follows:

[0033]

[0034] In the formula, The average deviation of the data within the window. To adjust the sliding window size, For the first in the window Data points, The average value of the data within the window; when When the deviation is less than the preset threshold, the data within the window is deemed valid, and batch parameter updates are performed based on the valid data.

[0035] As a preferred solution for the control method of a general-purpose mobile robotic arm composite robot, in step S5, after the fault is recovered, the task execution record and motion trajectory record are compared to continue the unfinished task. During this process, a data reconciliation algorithm is used to verify the continuity of the task. The formula for calculating the reconciliation matching degree is as follows:

[0036]

[0037] In the formula, To ensure accurate reconciliation, To compare the number of dimensions in the data, The first record before the fault Dimensional data, The first data collected after the fault was recovered Dimensional data, For indicator functions, when hour ,otherwise ;when If the amount exceeds the preset reconciliation threshold, the task can be continued.

[0038] This invention also provides a general-purpose mobile robotic arm composite robot control system, comprising:

[0039] The task modeling and scene recognition module is used to read preset task objectives, on-site environmental elements and system resource constraints, and determine the current operation scene type based on multimodal perception data;

[0040] The strategy selection and parameter initialization module is used to match the corresponding work process template according to the determined current work scenario type, and load the default work parameters and safety control thresholds for the current work scenario from the system database.

[0041] The rolling execution and safety control module is used to dynamically plan the travel path of the mobile chassis of the composite robot and the movement posture of the robotic arm in fixed time windows, and execute safety control rules in real time to monitor the operation process.

[0042] The learning and updating module is used to record the success or failure status, total time, and action deviation data of the task execution results after the task is completed, and to update the task parameters in batches based on the recorded data.

[0043] The anomaly handling and recovery module is used to automatically switch to local safe operation mode when a deterioration in network communication quality or a device malfunction is detected; after the fault is recovered, the unfinished work tasks are resumed by comparing the task execution records and action trajectory records.

[0044] As a preferred solution for a general-purpose mobile robotic arm composite robot control system, the task modeling and scene recognition module uses a cosine similarity algorithm to calculate the similarity between input features and scene template features. The formula is as follows:

[0045]

[0046] in, The input feature vector includes task target features, material property features, on-site obstacle features, channel features, time window features, and equipment capability features; These are scene template feature vectors, stored in the system policy library; For vectors and cosine similarity, when If the value exceeds a preset threshold, the input scene is determined to match the scene template.

[0047] As a preferred solution for a general-purpose mobile robotic arm composite robot control system, the fusion formula for multimodal perception data in the task modeling and scene recognition module is as follows:

[0048]

[0049] In the formula, To fuse feature vectors, To integrate weights, the weights are dynamically adjusted based on the complexity of the scenario. For visual feature vectors, Language feature vectors; based on fused feature vectors Complete task recognition and scene type determination, and map high-level intents into paths, postures and primitive commands.

[0050] As a preferred solution for a general-purpose mobile robotic arm composite robot control system, the strategy selection and parameter initialization module is also used to adjust the operation parameters using a fuzzy control algorithm. The calculation formula for the output of the fuzzy inference rule is as follows:

[0051]

[0052] In the formula, The final output parameters for the job. The number of fuzzy rules activated. For the first The membership degree of a fuzzy rule. For the first Output parameter values ​​corresponding to the fuzzy rules; parameter initialization is completed based on the adjusted job parameters.

[0053] As a preferred solution for a general-purpose mobile robotic arm composite robot control system, the rolling execution and safety control module employs a model predictive control algorithm to plan the path and posture, with the objective function being:

[0054]

[0055] In the formula, The objective function value is used to comprehensively evaluate the optimization effect of path attitude planning; To control the incremental sequence, including the speed adjustment of the moving chassis and the angle adjustment of each joint of the robotic arm; To predict the time domain length, i.e., the number of fixed time windows to be covered forward in each planning iteration; for Time prediction The system output includes the actual travel path deviation of the mobile chassis and the actual attitude deviation of the robotic arm's end effector. for The reference output at any given time is the preset expected travel path parameters and expected robotic arm posture parameters; , , These are respectively the path tracking accuracy weight, the control incremental smoothing weight, and the energy consumption optimization weight; for Time prediction The system's energy consumption value at any given time.

[0056] As a preferred solution for a general-purpose mobile robotic arm composite robot control system, the rolling execution and safety control module uses a distance detection algorithm to calculate the real-time distance between the robotic arm and obstacles. The distance calculation formula is as follows:

[0057]

[0058] In the formula, The distance between the end effector of the robotic arm and the obstacle in three-dimensional space. The real-time coordinates of the robotic arm's end effector. Let the coordinates be those of the obstacle; when When the distance is less than the preset minimum safe distance, a warning or emergency stop command will be triggered.

[0059] As a preferred solution for a general-purpose mobile robotic arm composite robot control system, the learning and updating module uses a reinforcement learning model to update the operation parameters in batches based on recorded data. The reinforcement learning model employs a temporal difference learning algorithm to update the parameters of key action primitives, and the value function update formula is as follows:

[0060]

[0061] In the formula, for Moment State The value function, for Moment State The value function, The learning rate is used to control the step size for parameter updates. From state Transfer to Instant rewards received This is a discount factor used to weigh immediate rewards against future rewards.

[0062] As a preferred solution for a general-purpose mobile robotic arm composite robot control system, the learning and updating module further optimizes the parameters of key motion primitives through a gradient descent algorithm. The parameter update formula is as follows:

[0063]

[0064] In the formula, For the updated parameters, The parameters before the update. This is the gradient descent step size. For loss function right gradient, Success rate, time consumption, energy consumption, and deviation calculations based on action primitives.

[0065] As a preferred solution for a general-purpose mobile robotic arm composite robot control system, in the learning and updating module, during the batch updating of operation parameters based on recorded data, the recorded operation data and equipment health data are filtered for valid data using a sliding window algorithm. The formula for determining the validity of data within the window is as follows:

[0066]

[0067] In the formula, The average deviation of the data within the window. To adjust the sliding window size, For the first in the window Data points, The average value of the data within the window; when When the deviation is less than the preset deviation threshold, the data in the window is determined to be valid, and batch parameter updates are performed based on the valid data.

[0068] In the anomaly handling and recovery module, a data reconciliation algorithm is used to verify task continuity. The formula for calculating the reconciliation matching degree is:

[0069]

[0070] In the formula, To ensure accurate reconciliation, To compare the number of dimensions in the data, The first record before the fault Dimensional data, The first data collected after the fault was recovered Dimensional data, For indicator functions, when hour ,otherwise ;when If the amount exceeds the preset reconciliation threshold, the task can be continued.

[0071] The beneficial effects of this invention are as follows:

[0072] First, the present invention can automatically identify the work scenario and resource constraints, match or combine the corresponding processes and parameters without manual recalibration, avoid work interruption caused by scenario changes, and is especially suitable for flexible production needs of multiple batches and multiple categories, greatly improving work continuity and efficiency.

[0073] Secondly, this invention can automatically complete small-batch updates of key primitive parameters based on job data, eliminating the reliance on manual parameter tuning in existing technologies, enabling continuous and precise improvement of job skills, switching to local safety mode in case of failure, and reconciling and resuming operations after recovery, solving the problem of inefficient exception handling, reducing job interruptions and losses caused by failures, and ensuring continuous system operation.

[0074] Third, the unified monitoring and data closed-loop module enables centralized collection, storage, and correlation analysis of operational data and equipment health data. Combined with data playback and small-batch verification mechanisms, it forms an improved process of data collection-analysis-optimization-verification-full update, solving the problems of data dispersion and difficulty in closing the loop in existing technologies, and providing data support for long-term system reliability improvement and functional iteration. Attached Figure Description

[0075] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0076] Figure 1 This is a schematic diagram of the control method for a general-purpose mobile robotic arm composite robot provided in an embodiment of the present invention;

[0077] Figure 2 A technical roadmap for a general-purpose mobile robotic arm composite robot control method provided in this embodiment of the invention;

[0078] Figure 3 This is a diagram illustrating the architecture of a general-purpose mobile robotic arm composite robot control system provided in an embodiment of the present invention. Detailed Implementation

[0079] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0080] Example 1

[0081] See Figure 1 and Figure 2 This invention provides a general-purpose mobile robotic arm composite robot control method, comprising the following steps:

[0082] S1. Read the preset task objectives, on-site environmental elements, and system resource constraints, and determine the current operation scenario type based on multimodal perception data; by integrating task objectives, environmental elements, and resource constraints, and combining multimodal perception data, construct the cognitive basis of the operation scenario, and provide scenario boundary information for decision-making;

[0083] S2. Match the corresponding work process template according to the determined current work scenario type, and load the default work parameters and safety control thresholds for the current work scenario from the system database; match the process template and load the parameters based on the scenario determination result to realize the automatic correspondence between scenario and policy, and avoid parameter mismatch caused by manual intervention;

[0084] S3. Using fixed time windows as units, the travel path of the mobile chassis of the composite robot and the movement posture of the robotic arm are dynamically planned, and safety control rules are executed in real time to monitor the operation process. Dynamic planning with fixed time windows can adjust the path and posture according to the real-time environment, and monitor in real time through safety control rules, thus balancing the flexibility and safety of the operation.

[0085] S4. After the task is completed, record the success or failure status, total time and action deviation data of the task execution results, and update the task parameters in batches based on the recorded data; the batch update of parameters based on the task data allows the system to accumulate experience from historical tasks and improve the accuracy and efficiency of subsequent tasks.

[0086] S5. When a deterioration in network communication quality or a device malfunction is detected, the system automatically switches to a local safe operation mode. After the fault is resolved, unfinished tasks are resumed by comparing task execution records with action trajectory records. By handling sudden faults in local safe mode and resuming tasks through record comparison, the system ensures that the overall workflow is not interrupted by faults, guaranteeing the continuity and reliability of system operation.

[0087] In this embodiment, in step S1, during the scene recognition process of determining the current work scene type based on multimodal perception data, the cosine similarity algorithm is used to calculate the similarity between the input features and the scene template features. The formula is as follows:

[0088]

[0089] in, The input feature vector includes task target features, material property features, on-site obstacle features, channel features, time window features, and equipment capability features; These are scene template feature vectors, stored in the system policy library; For vectors and cosine similarity, when If the value exceeds a preset threshold, the input scene is determined to match the scene template.

[0090] Specifically, the core of the cosine similarity algorithm is to quantify the degree of feature similarity by using the cosine value of the angle between vectors. The smaller the angle, the closer the cosine value is to 1, and the higher the feature similarity. Input feature vectors It covers multiple dimensions of information, including tasks, materials, environment, and equipment, ensuring a comprehensive and accurate description of the current work scenario; scenario template vectors. It is a set of standard scene features pre-stored by the system. Each template corresponds to a typical work scenario. For example, the precision assembly scenario template includes features such as small-sized materials, narrow passages, and high-precision equipment capabilities. Through calculation... and The cosine similarity can quickly determine the fit between the current scene and the standard template. When the similarity exceeds the preset threshold, the scene type can be determined. This process does not require human intervention, which greatly improves the efficiency and accuracy of scene recognition and solves the problems of time-consuming and error-prone traditional manual scene determination.

[0091] In this embodiment, the fusion formula for multimodal sensing data in step S1 is:

[0092]

[0093] In the formula, To fuse feature vectors, To integrate weights, the weights are dynamically adjusted based on the complexity of the scenario. For visual feature vectors, Language feature vectors; based on fused feature vectors Complete task recognition and scene type determination, and map high-level intents into paths, postures and primitive commands.

[0094] Specifically, visual feature vectors Image data extraction can intuitively reflect the physical characteristics of the on-site environment, such as material shape, obstacle location, and passage width, but it is easily affected by environmental factors such as lighting and occlusion; language feature vectors Extracting data from the task description text clarifies the high-level intent of the task, such as material handling and workpiece assembly, but lacks detailed descriptions of the actual environment. (Weighting / Integration) Dynamically adjust based on scene complexity; increase [adjustment] in complex environments such as insufficient lighting or mixed materials. To increase the proportion of visual features and ensure that scene judgments accurately reflect the actual environment; when the environment is simple but the task intent is clear, reduce... To highlight the role of language features and ensure that scene determination does not deviate from the task objective. (Fused) It combines environmental details and task intent, providing feature support for task recognition and scene determination, thereby achieving accurate mapping from high-level intent to specific paths, postures and primitive commands.

[0095] In this embodiment, step S2 further includes adjusting the job parameters using a fuzzy control algorithm. The formula for calculating the output of the fuzzy inference rule is as follows:

[0096]

[0097] In the formula, The final output parameters for the job. The number of fuzzy rules activated. For the first The membership degree of a fuzzy rule. For the first Output parameter values ​​corresponding to the fuzzy rules; parameter initialization is completed based on the adjusted job parameters.

[0098] Specifically, fuzzy control algorithms are designed to address problems where operational parameters are difficult to describe with precise mathematical models, and are suitable for operational scenarios involving fuzziness and uncertainty. The system is pre-set... A set of fuzzy rules, each corresponding to a type of scenario feature and the relationship between operational parameters, such as fragile materials corresponding to smaller gripping force, and heavy materials corresponding to larger gripping force. Membership degree. Used to measure the current work scenario and the first The degree of fit of the fuzzy rules; the higher the degree of fit, the better. The larger the value; It is the first The reference operation parameter values ​​corresponding to each rule provide a basis for parameter adjustment. This is achieved through the activation of... The fuzzy rules are used to perform a weighted average calculation to obtain the final output operation parameters U, realizing the conversion from fuzzy scene features to precise operation parameters. Compared with the traditional fixed parameter initialization method, the parameters adjusted by the fuzzy control algorithm can better adapt to the characteristics of different scenarios, avoid operation defects caused by parameter rigidity, and improve the rationality and adaptability of parameter initialization.

[0099] In this embodiment, in step S3, during the dynamic planning of the mobile chassis's travel path and the robotic arm's movement posture, a model predictive control algorithm is used to plan the path and posture, and the objective function is:

[0100]

[0101] In the formula, The objective function value is used to comprehensively evaluate the optimization effect of path attitude planning; To control the incremental sequence, including the speed adjustment of the moving chassis and the angle adjustment of each joint of the robotic arm; To predict the time domain length, i.e., the number of fixed time windows to be covered forward in each planning iteration; for Time prediction The system output includes the actual travel path deviation of the mobile chassis and the actual attitude deviation of the robotic arm's end effector. for The reference output at any given time is the preset expected travel path parameters and expected robotic arm posture parameters; , , These are respectively the path tracking accuracy weight, the control incremental smoothing weight, and the energy consumption optimization weight; for Time prediction The system's energy consumption value at any given time.

[0102] Specifically, the prediction time domain length N determines the forward-looking nature of the planning. A larger N value allows the system to predict environmental and operational states further in advance, thus better avoiding potential obstacles and optimizing paths, but it also increases computational load. The N value needs to be set reasonably based on the complexity of the operational scenario and real-time requirements. Controlling the incremental sequence... The adjustment range of chassis speed and robotic arm joint angles directly determines the key variables for achieving path and attitude adjustments. In the objective function, the first term ensures path tracking accuracy by calculating the sum of squared deviations between the actual output and the reference output, with weights... It can be adjusted according to the accuracy requirements of the operation, such as the need to increase the size for precision assembly scenarios. Prioritizing accuracy; secondly, by controlling the sum of squares of the control increments, to avoid sudden changes in control commands that could lead to equipment instability, weighting... This can be increased in scenarios where equipment stability is critical; the third item optimizes energy consumption by predicting energy consumption values, with weighted... It can be scaled up for long-term operations or in scenarios with limited energy. By comprehensively optimizing the three objectives, the model predictive control algorithm can meet the requirements of operational accuracy while also taking into account equipment operational stability and energy economy, thus overcoming the limitations of single-objective optimization in traditional planning algorithms.

[0103] In this embodiment, step S3 involves using a distance detection algorithm to calculate the real-time distance between the robotic arm and the obstacle. The distance calculation formula is as follows:

[0104]

[0105] In the formula, The distance between the end effector of the robotic arm and the obstacle in three-dimensional space. The real-time coordinates of the robotic arm's end effector. Let the coordinates be those of the obstacle; when When the distance is less than the preset minimum safe distance, a warning or emergency stop command will be triggered.

[0106] Specifically, the real-time coordinates of the robotic arm's end effector The coordinates of the obstacle are collected in real time by position sensors to ensure that the movement position of the end effector is dynamically reflected; Visual sensors identify and locate obstacles, ensuring timely capture of changes in their position. By calculating the three-dimensional spatial distance D between the robotic arm and the obstacle, the relative positional relationship between them can be monitored in real time. The system pre-sets a minimum safe distance based on the operational scenario. This distance takes into account factors such as the robotic arm's movement accuracy and the type of obstacle. For example, the minimum safe distance needs to be appropriately increased for dynamic obstacles such as personnel, and relatively decreased for static obstacles such as fixed equipment. When D is less than the minimum safe distance, the system first triggers a warning command, such as controlling the robotic arm to decelerate. If D continues to decrease and approaches the danger threshold, an emergency stop command is triggered, forcing the robotic arm to stop moving. This forms a dual safety protection mechanism of "warning-emergency stop," effectively preventing collisions between the robotic arm and obstacles and ensuring the safety of equipment and personnel during operations.

[0107] In this embodiment, in step S4, the task parameters are updated in batches based on the recorded data using a reinforcement learning model. The reinforcement learning model uses a temporal difference learning algorithm to update the parameters of the key action primitives, and the value function update formula is as follows:

[0108]

[0109] In the formula, for Moment State The value function, for Moment State The value function, The learning rate is used to control the step size for parameter updates. From state Transfer to Instant rewards received This is a discount factor used to weigh immediate rewards against future rewards. Status and Representing time t and The job status at any given time, such as the state before and after fetching, and the value function. This is used to measure the "quality" of different states; the more favorable the state is for completing the task, the better. The higher the value, the greater the instant reward. Based on the task results after state transition, a positive reward is given for successful capture, and a negative reward is given for failed capture. The reward signal guides the system to adjust parameters in an optimization direction. Learning rate. Control the magnitude of parameter updates. An excessively large value may cause drastic fluctuations in the parameters. A value that is too small will slow down parameter updates; it needs to be set reasonably according to the stability requirements of the operating scenario; discount factor Used to balance the weight of immediate rewards and future rewards. When the value is close to 1, the system places more emphasis on long-term cumulative rewards. When the value is close to 0, the system focuses more on immediate rewards, through Adjustments can allow the system to find a balance between short-term operational efficiency and long-term operational performance. This is achieved through continuous iterative updates to the value function. The system can gradually learn the optimal operation parameters, realize the autonomous optimization of key action primitive parameters, and get rid of the dependence on manual parameter tuning.

[0110] The parameter update also optimizes the parameters of the key action primitives using the gradient descent algorithm. The parameter update formula is as follows:

[0111]

[0112] In the formula, For the updated parameters, The parameters before the update. This is the gradient descent step size. For loss function right gradient, Success rate, time consumption, energy consumption, and bias calculation based on action primitives. Loss function. Taking into account multiple dimensions such as success rate, time consumption, energy consumption, and deviation of action primitives, the parameters The worse the corresponding operating performance, The larger the value, the smaller the gradient. Indicator loss function at current parameters The direction of change at a given point indicates whether adjusting the parameters along that direction will increase the loss function, or vice versa. Gradient descent step size. Control the magnitude of each parameter adjustment. An excessively large value may cause the parameter to exceed the minimum value of the loss function. If the value is too small, it will prolong the parameter optimization cycle; therefore, it needs to be set according to the accuracy and efficiency requirements of parameter optimization. By updating the formula, the parameters can be gradually adjusted along the direction of decreasing loss function until they converge to the optimal parameters that minimize the loss function. This allows for precise quantification of the impact of parameters on job performance, enabling refined parameter optimization and further improving the job quality of action primitives, thus compensating for the shortcomings of temporal difference learning algorithms in parameter optimization accuracy.

[0113] In this embodiment, during step S4, when updating the operation parameters in batches based on the recorded data, the recorded operation data and equipment health data are filtered for valid data using a sliding window algorithm. The formula for determining the validity of data within the window is as follows:

[0114]

[0115] In the formula, The average deviation of the data within the window. To adjust the sliding window size, For the first in the window Data points, The average value of the data within the window; when When the deviation is less than a preset threshold, the data within the window is considered valid, and batch parameter updates are performed based on the valid data. (Sliding window size) This determines the range of data to be filtered each time. The larger the value, the better the filtered data reflects long-term operational trends, but the slower the response speed to recent data changes will be. A smaller value indicates a more sensitive response to recent data changes, but it may be more susceptible to short-term anomalies and should be set according to data stability and update frequency requirements. (The last part, "in the window," appears to be an unrelated instruction and is left untranslated.) Data points It is raw data collected in real time, average value It represents the average level and average deviation of all data points within the window. The dispersion of data within a window is quantified by calculating the average of the absolute deviations of each data point from the average value. The smaller the value, the more concentrated and stable the data within the window is, and the absence of obvious outliers. A larger value indicates a higher degree of data dispersion within the window, potentially containing outliers. The system has a preset deviation threshold. When the value is less than the threshold, the data within the window is considered valid and can be used for parameter updates; when... When the value exceeds the threshold, the data is considered abnormal and the data in that window must be discarded to avoid deviations in parameter update direction caused by abnormal data. This ensures the accuracy and stability of parameter updates and provides reliable data support for the system's autonomous optimization.

[0116] In this embodiment, step S5, after the fault is recovered, involves comparing the task execution record with the action trajectory record to continue the unfinished work tasks. During this process, a data reconciliation algorithm is used to verify the continuity of the tasks. The formula for calculating the reconciliation matching degree is as follows:

[0117]

[0118] In the formula, To ensure accurate reconciliation, To compare the number of dimensions in the data, The first record before the fault Dimensional data, The first data collected after the fault was recovered Dimensional data, For indicator functions, when hour ,otherwise ;when If the amount exceeds the preset reconciliation threshold, the task can be continued.

[0119] Specifically, the number of dimensions of the comparison data It covers key status indicators for task execution, such as task progress, robotic arm joint angles, mobile chassis position, and operational parameter settings. A larger value indicates a more comprehensive reflection of the system's state consistency, but it also increases computational complexity. The value needs to be set based on task complexity and the importance of key state indicators. The i-th dimension of data recorded before the failure. It is a snapshot of the critical state of the system before the failure occurred, and the i-th dimension of data collected after the failure is recovered. This refers to the real-time status data after system recovery. An indicator function I is used to determine whether the pre-failure data and post-recovery data for the corresponding dimension are consistent; a match is counted as 1, and an inconsistency as 0. The reconciliation matching degree M is the ratio of the number of consistent dimensions to the total number of dimensions. The larger the M value, the closer the system's post-recovery state is to its pre-failure state, and the higher the feasibility of task continuation. The system presets a reconciliation threshold. When M is greater than the threshold, the system state is deemed to meet the continuation conditions, and incomplete action instructions can be reissued based on the pre-failure task execution records and action trajectory records to achieve task continuation. When M is less than the threshold, the system state is deemed to differ significantly from the pre-failure state, and continuation may lead to job failure, requiring task restart. This algorithm ensures the safety and feasibility of task continuation, avoiding job risks caused by inconsistent states.

[0120] The application scenarios of this invention are as follows:

[0121] Scenario 1: Flexible production scenario with multiple product categories:

[0122] In multi-category, small-batch production scenarios such as electronic component assembly and small mechanical parts processing, production lines frequently need to switch work objects, such as switching from assembling mobile phone motherboards to assembling computer graphics cards. This requires rapid replacement of the corresponding mobile chassis and robotic arm end-effector tools, and the scene elements—workpiece size, assembly precision requirements, and workspace—change frequently. This invention achieves plug-and-play functionality through standardized interfaces, eliminating the need for case-by-case interface debugging. It automatically identifies new scenes using multimodal perception and cosine similarity algorithms, matching corresponding assembly process templates and parameters. Combined with reinforcement learning, it dynamically optimizes assembly parameters, significantly reducing manual intervention during scene switching, shortening production line adjustment cycles, and adapting to the high-frequency switching needs of flexible production.

[0123] Scenario 2: Intelligent Handling in Warehouse Logistics

[0124] In scenarios requiring large-scale material handling, such as e-commerce warehousing and industrial raw material warehouses, the types of materials being handled (e.g., cardboard boxes, metal parts, fragile items), the warehouse aisle environment (e.g., temporarily piled obstacles, aisles of varying widths), and handling time windows (e.g., peak-hour expedited delivery demands) vary significantly. This invention uses a distance detection algorithm to calculate the distance between the robotic arm and obstacles in real time, ensuring safe handling in narrow aisles. It dynamically adjusts handling parameters based on a fuzzy control algorithm, such as automatically reducing gripping force when handling fragile items and increasing chassis movement speed when widening aisles. An anomaly handling mechanism can cope with warehouse network fluctuations; for example, switching to local safety mode when the network signal in the shelving area is weak, and resuming unfinished handling tasks after fault recovery through data reconciliation, avoiding material accumulation or task interruption, and improving the efficiency and reliability of warehouse handling.

[0125] Scenario 3: Maintenance of Intelligent Manufacturing Equipment

[0126] In scenarios such as automobile production lines and heavy machinery manufacturing, critical components of large equipment, such as engine bearings and machine tool guideways, require regular inspection, cleaning, or minor maintenance. These operating environments are characterized by complex equipment structures (multiple obstructions, narrow spaces), high precision requirements (e.g., avoiding damage to component surfaces during cleaning), and the need for real-time monitoring of equipment operation. This invention achieves precise operation in complex environments through multimodal perception data fusion, visual recognition of component positions, and force sensing control of cleaning intensity. A model predictive control algorithm plans the robotic arm's posture in narrow spaces to avoid collisions with other equipment components. A sliding window algorithm filters equipment health data, and reinforcement learning optimizes maintenance parameters. Simultaneously, a unified data loop stores maintenance records, providing data support for subsequent equipment fault prediction and ensuring the stable operation of intelligent manufacturing equipment.

[0127] Example 2

[0128] See Figure 3 Embodiment 2 of the present invention also provides a general-purpose mobile robotic arm composite robot control system, comprising:

[0129] The task modeling and scene recognition module 100 is used to read preset task objectives, on-site environmental elements and system resource constraints, and determine the current operation scene type based on multimodal perception data;

[0130] The strategy selection and parameter initialization module 200 is used to match the corresponding work process template according to the determined current work scenario type, and load the default work parameters and safety control thresholds under the current work scenario from the system database.

[0131] The rolling execution and safety control module 300 is used to dynamically plan the travel path of the mobile chassis of the composite robot and the action posture of the robotic arm in fixed time windows, and execute safety control rules in real time to monitor the operation process.

[0132] The learning and updating module 400 is used to record the success or failure status, total time consumption, and action deviation data of the task execution results after the task is completed, and to update the task parameters in batches based on the recorded data.

[0133] The anomaly handling and recovery module 500 is used to automatically switch to local safe operation mode when a deterioration in network communication quality or a device malfunction is detected; after the fault is recovered, the unfinished work tasks are resumed by comparing the task execution record and the action trajectory record.

[0134] In this embodiment, the task modeling and scene recognition module 100 uses a cosine similarity algorithm to calculate the similarity between the input features and the scene template features. The formula is as follows:

[0135]

[0136] in, The input feature vector includes task target features, material property features, on-site obstacle features, channel features, time window features, and equipment capability features; These are scene template feature vectors, stored in the system policy library; For vectors and cosine similarity, when If the value exceeds a preset threshold, the input scene is determined to match the scene template.

[0137] In this embodiment, the fusion formula for multimodal perception data in the task modeling and scene recognition module 100 is as follows:

[0138]

[0139] In the formula, To fuse feature vectors, To integrate weights, the weights are dynamically adjusted based on the complexity of the scenario. For visual feature vectors, Language feature vectors; based on fused feature vectors Complete task recognition and scene type determination, and map high-level intents into paths, postures and primitive commands.

[0140] In this embodiment, the strategy selection and parameter initialization module 200 is further used to adjust the job parameters using a fuzzy control algorithm. The formula for calculating the output of the fuzzy inference rule is as follows:

[0141]

[0142] In the formula, The final output parameters for the job. The number of fuzzy rules activated. For the first The membership degree of a fuzzy rule. For the first Output parameter values ​​corresponding to the fuzzy rules; parameter initialization is completed based on the adjusted job parameters.

[0143] In this embodiment, the rolling execution and safety control module 300 uses a model predictive control algorithm to plan the path and attitude, with the objective function being:

[0144]

[0145] In the formula, The objective function value is used to comprehensively evaluate the optimization effect of path attitude planning; To control the incremental sequence, including the speed adjustment of the moving chassis and the angle adjustment of each joint of the robotic arm; To predict the time domain length, i.e., the number of fixed time windows to be covered forward in each planning iteration; for Time prediction The system output includes the actual travel path deviation of the mobile chassis and the actual attitude deviation of the robotic arm's end effector. for The reference output at any given time is the preset expected travel path parameters and expected robotic arm posture parameters; , , These are respectively the path tracking accuracy weight, the control incremental smoothing weight, and the energy consumption optimization weight; for Time prediction The system's energy consumption value at any given time.

[0146] In this embodiment, the rolling execution and safety control module 300 uses a distance detection algorithm to calculate the real-time distance between the robotic arm and the obstacle. The distance calculation formula is as follows:

[0147]

[0148] In the formula, The distance between the end effector of the robotic arm and the obstacle in three-dimensional space. The real-time coordinates of the robotic arm's end effector. Let the coordinates be those of the obstacle; when When the distance is less than the preset minimum safe distance, a warning or emergency stop command will be triggered.

[0149] In this embodiment, the learning and updating module 400 uses a reinforcement learning model to batch update the job parameters based on recorded data. The reinforcement learning model uses a temporal difference learning algorithm to update the parameters of key action primitives, and the value function update formula is:

[0150]

[0151] In the formula, for Moment State The value function, for Moment State The value function, The learning rate is used to control the step size for parameter updates. From state Transfer to Instant rewards received This is a discount factor used to weigh immediate rewards against future rewards.

[0152] In this embodiment, the learning and updating module 400 further optimizes the parameters of the key action primitives through a gradient descent algorithm. The parameter update formula is as follows:

[0153]

[0154] In the formula, For the updated parameters, The parameters before the update. This is the gradient descent step size. For loss function right gradient, Success rate, time consumption, energy consumption, and deviation calculations based on action primitives.

[0155] In this embodiment, during the batch update of job parameters based on recorded data in the learning and updating module 400, the recorded job data and equipment health data are filtered for valid data using a sliding window algorithm. The formula for determining the validity of data within the window is as follows:

[0156]

[0157] In the formula, The average deviation of the data within the window. To adjust the sliding window size, For the first in the window Data points, The average value of the data within the window; when When the deviation is less than the preset threshold, the data within the window is deemed valid, and batch parameter updates are performed based on the valid data.

[0158] In this embodiment, the anomaly handling and recovery module 500 uses a data reconciliation algorithm to verify task continuity. The formula for calculating the reconciliation matching degree is:

[0159]

[0160] In the formula, To ensure accurate reconciliation, To compare the number of dimensions in the data, The first record before the fault Dimensional data, The first data collected after the fault was recovered Dimensional data, For indicator functions, when hour ,otherwise ;when If the amount exceeds the preset reconciliation threshold, the task can be continued.

[0161] It should be noted that the information interaction and execution process between the modules of the above system are based on the same concept as the method embodiment in Embodiment 1 of this application, and the resulting technical effects are the same as those in the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in this application, and it will not be repeated here.

[0162] Example 3

[0163] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium storing program code of a general-purpose mobile robotic arm composite robot control method. The program code includes instructions for executing the general-purpose mobile robotic arm composite robot control method of Embodiment 1 or any possible implementation thereof.

[0164] Computer-readable storage media can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0165] Example 4

[0166] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0167] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor can execute the general mobile robotic arm composite robot control method of Embodiment 1 or any possible implementation thereof by calling the program instructions.

[0168] Specifically, a processor can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0169] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0170] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0171] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A general-purpose control method for a mobile robotic arm composite robot, characterized in that, Includes the following steps: S1. Read the preset task objectives, on-site environmental elements, and system resource constraints, and determine the current operation scenario type based on multimodal perception data; S2. Match the corresponding work process template according to the determined current work scenario type, and load the default work parameters and safety control thresholds for the current work scenario from the system database. S3. Using a fixed time window as a unit, dynamically plan the travel path of the mobile chassis of the composite robot and the movement posture of the robotic arm, and execute safety control rules in real time to monitor the operation process. S4. After the task is completed, record the success or failure status of the task execution, the total time consumed, and the action deviation data, and update the task parameters in batches based on the recorded data; S5. When a network communication quality deterioration or equipment failure is detected, automatically switch to local safe operation mode; after the fault is recovered, continue the unfinished work tasks by comparing the task execution record and the action trajectory record.

2. The control method for the general-purpose mobile robotic arm composite robot according to claim 1, characterized in that, In step S1, during the scene recognition process based on multimodal perception data to determine the current work scene type, the cosine similarity algorithm is used to calculate the similarity between the input features and the scene template features. The formula is as follows: ； in, The input feature vector includes task target features, material property features, on-site obstacle features, channel features, time window features, and equipment capability features; These are scene template feature vectors, stored in the system policy library; For vectors and cosine similarity, when If the value exceeds a preset threshold, the input scene is determined to match the scene template. The formula for fusing multimodal sensing data is: ； In the formula, To fuse feature vectors, To integrate weights, the weights are dynamically adjusted based on the complexity of the scenario. For visual feature vectors, For language feature vectors; based on fused feature vectors Complete task recognition and scene type determination, and map high-level intents into paths, postures and primitive commands.

3. The control method for the general-purpose mobile robotic arm composite robot according to claim 1, characterized in that, Step S2 also includes adjusting the operation parameters using a fuzzy control algorithm. The formula for calculating the output of the fuzzy inference rule is as follows: ； In the formula, The final output parameters for the job. The number of fuzzy rules activated. For the first The membership degree of a fuzzy rule. For the first The output parameter values ​​corresponding to the fuzzy rules; Parameter initialization is completed based on the adjusted job parameters.

4. The control method for the general-purpose mobile robotic arm composite robot according to claim 1, characterized in that, Step S3, in the process of dynamically planning the travel path of the composite robot's mobile chassis and the motion posture of the robotic arm, a model predictive control algorithm is used to plan the path and posture. The objective function is: ； In the formula, The objective function value is used to comprehensively evaluate the optimization effect of path attitude planning; To control the incremental sequence, including the speed adjustment of the moving chassis and the angle adjustment of each joint of the robotic arm; To predict the time domain length, i.e., the number of fixed time windows to be covered forward in each planning iteration; for Time prediction The system output includes the actual travel path deviation of the mobile chassis and the actual attitude deviation of the robotic arm's end effector. for The reference output at any given time is the preset expected travel path parameters and expected robotic arm posture parameters; , , These are respectively the path tracking accuracy weight, the control incremental smoothing weight, and the energy consumption optimization weight; for Time prediction Real-time system energy consumption value; Step S3: Calculate the real-time distance between the robotic arm and the obstacle using a distance detection algorithm. The distance calculation formula is: ； In the formula, The distance between the end effector of the robotic arm and the obstacle in three-dimensional space. The real-time coordinates of the robotic arm's end effector. Let the coordinates be those of the obstacle; when When the distance is less than the preset minimum safe distance, a warning or emergency stop command will be triggered.

5. The control method for a general-purpose mobile robotic arm composite robot according to claim 1, characterized in that, In step S4, the task parameters are updated in batches based on the recorded data using a reinforcement learning model. The reinforcement learning model uses a temporal difference learning algorithm to update the parameters of the key action primitives, and the value function update formula is as follows: ； In the formula, for Moment State The value function, for Moment State The value function, The learning rate is used to control the step size for parameter updates. From state Transfer to Instant rewards received This is a discount factor used to weigh immediate rewards against future rewards; The parameter update also optimizes the parameters of the key action primitives using the gradient descent algorithm. The parameter update formula is as follows: ； In the formula, For the updated parameters, The parameters before the update. This is the gradient descent step size. For loss function right gradient, Calculations based on success rate, time consumption, energy consumption, and deviation of action primitives; In step S4, during the batch update of operation parameters based on recorded data, the recorded operation data and equipment health data are filtered for valid data using a sliding window algorithm. The formula for determining the validity of data within the window is as follows: ； In the formula, The average deviation of the data within the window. To adjust the sliding window size, For the first in the window Data points, The average value of the data within the window; when When the deviation is less than the preset threshold, the data within the window is deemed valid, and batch parameter updates are performed based on the valid data.

6. The control method for the general-purpose mobile robotic arm composite robot according to claim 1, characterized in that, Step S5: After the fault is resolved, the task execution record is compared with the action trajectory record to continue the unfinished work tasks. During this process, a data reconciliation algorithm is used to verify the continuity of the tasks. The formula for calculating the reconciliation matching degree is: ； In the formula, To ensure accurate reconciliation, To compare the number of dimensions in the data, The first record before the fault Dimensional data, The first data collected after the fault was recovered Dimensional data, For indicator functions, when hour ,otherwise ;when If the amount exceeds the preset reconciliation threshold, the task can be continued.

7. A general-purpose mobile robotic arm composite robot control system, characterized in that, include: The task modeling and scene recognition module is used to read preset task objectives, on-site environmental elements and system resource constraints, and determine the current operation scene type based on multimodal perception data; The strategy selection and parameter initialization module is used to match the corresponding work process template according to the determined current work scenario type, and load the default work parameters and safety control thresholds for the current work scenario from the system database. The rolling execution and safety control module is used to dynamically plan the travel path of the mobile chassis of the composite robot and the movement posture of the robotic arm in fixed time windows, and execute safety control rules in real time to monitor the operation process. The learning and updating module is used to record the success or failure status, total time, and action deviation data of the task execution results after the task is completed, and to update the task parameters in batches based on the recorded data. The anomaly handling and recovery module is used to automatically switch to local safe operation mode when a deterioration in network communication quality or a device malfunction is detected; after the fault is recovered, the unfinished work tasks are resumed by comparing the task execution records and action trajectory records.

8. The universal mobile robotic arm composite robot control system according to claim 7, characterized in that, In the task modeling and scene recognition module, the cosine similarity algorithm is used to calculate the similarity between the input features and the scene template features. The formula is as follows: ； in, The input feature vector includes task target features, material property features, on-site obstacle features, channel features, time window features, and equipment capability features; These are scene template feature vectors, stored in the system policy library; For vectors and cosine similarity, when If the value exceeds a preset threshold, the input scene is determined to match the scene template. In the task modeling and scene recognition module, the fusion formula for multimodal perception data is as follows: ； In the formula, To fuse feature vectors, To integrate weights, the weights are dynamically adjusted based on the complexity of the scenario. For visual feature vectors, For language feature vectors; based on fused feature vectors Complete task recognition and scene type determination, and map high-level intents into paths, postures and primitive commands.

9. A general-purpose mobile robotic arm composite robot control system according to claim 7, characterized in that, The strategy selection and parameter initialization module is also used to adjust the operation parameters using a fuzzy control algorithm. The formula for calculating the output of the fuzzy inference rule is: ； In the formula, The final output parameters for the job. The number of fuzzy rules activated. For the first The membership degree of a fuzzy rule. For the first The output parameter values ​​corresponding to the fuzzy rules; Parameter initialization is completed based on the adjusted job parameters; In the rolling execution and safety control module, a model predictive control algorithm is used to plan the path and attitude, and the objective function is: ； In the formula, The objective function value is used to comprehensively evaluate the optimization effect of path attitude planning; To control the incremental sequence, including the speed adjustment of the moving chassis and the angle adjustment of each joint of the robotic arm; To predict the time domain length, i.e., the number of fixed time windows to be covered forward in each planning iteration; for Time prediction The system output includes the actual travel path deviation of the mobile chassis and the actual attitude deviation of the robotic arm's end effector. for The reference output at any given time is the preset expected travel path parameters and expected robotic arm posture parameters; , , These are respectively the path tracking accuracy weight, the control incremental smoothing weight, and the energy consumption optimization weight; for Time prediction Real-time system energy consumption value; In the rolling execution and safety control module, a distance detection algorithm is used to calculate the real-time distance between the robotic arm and the obstacle. The distance calculation formula is as follows: ； In the formula, The distance between the end effector of the robotic arm and the obstacle in three-dimensional space. The real-time coordinates of the robotic arm's end effector. Let the coordinates be those of the obstacle; when When the distance is less than the preset minimum safe distance, a warning or emergency stop command will be triggered.

10. The universal mobile robotic arm composite robot control system according to claim 7, characterized in that, In the learning and updating module, the task parameters are updated in batches based on recorded data using a reinforcement learning model. The reinforcement learning model employs a temporal difference learning algorithm to update the parameters of key action primitives, and the value function update formula is as follows: ； In the formula, for Moment State The value function, for Moment State The value function, The learning rate is used to control the step size for parameter updates. From state Transfer to Instant rewards received This is a discount factor used to weigh immediate rewards against future rewards; In the learning and updating module, parameter updating further optimizes the parameters of key action primitives using a gradient descent algorithm. The parameter update formula is as follows: ； In the formula, For the updated parameters, The parameters before the update. This is the gradient descent step size. For loss function right gradient, Calculations based on success rate, time consumption, energy consumption, and deviation of action primitives; In the learning and updating module, during the batch updating of job parameters based on recorded data, the recorded job data and equipment health data are filtered for valid data using a sliding window algorithm. The formula for determining the validity of data within the window is as follows: ； In the formula, The average deviation of the data within the window. To adjust the sliding window size, For the first in the window Data points, The average value of the data within the window; when When the deviation is less than the preset deviation threshold, the data in the window is determined to be valid, and batch parameter updates are performed based on the valid data. In the anomaly handling and recovery module, a data reconciliation algorithm is used to verify task continuity. The formula for calculating the reconciliation matching degree is: ； In the formula, To ensure accurate reconciliation, To compare the number of dimensions in the data, The first record before the fault Dimensional data, The first data collected after the fault was recovered Dimensional data, For indicator functions, when hour ,otherwise ;when If the amount exceeds the preset reconciliation threshold, the task can be continued.

Images