A target recognition-based automatic control method and system for a robot arm

Abstract

The application relates to the technical field of automatic control of mechanical arms and discloses a mechanical arm automatic control method and system based on target identification. The method collects target object position, speed and mechanical arm joint angle data through vision, speed, angle and other sensors, obtains a prediction value at a control moment by using a prediction model, corrects the prediction value in combination with data matching degree and correlation degree, clusters three-dimensional data, analyzes similarity under different sequence lengths to obtain a historical prediction value, obtains a mechanical arm trajectory prediction value according to the corrected prediction value, the historical prediction value and data similarity, and then carries out motion planning control. The system comprises a memory, a processor and related programs. The application improves the trajectory prediction accuracy and self-adaptability of the mechanical arm, can effectively cope with a dynamic working environment, and improves control precision and efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic control of robotic arms, and in particular to a method and system for automatic control of robotic arms based on target recognition. Background Art

[0002] Robotic arms are increasingly used in industrial automation, ranging from simple material handling to complex precision assembly. Their automated control directly impacts production efficiency and quality. Existing robotic arm control methods typically rely on pre-set trajectories or fixed programs, which often exhibit significant limitations when working with dynamically changing objects or complex working environments.

[0003] The development of intelligent manufacturing is placing higher demands on the flexibility and adaptability of robotic arms. In actual production, parameters such as the position and speed of target objects can change at any time. For example, in logistics sorting scenarios, the position and speed of items on a conveyor belt can be unstable; in assembly operations, the position of parts to be assembled can deviate. Traditional control methods struggle to accurately track these dynamically changing targets in real time, resulting in reduced robot arm motion accuracy and even potential operational errors.

[0004] While existing target recognition-based robotic arm control methods can obtain partial target information through sensors, they lack sufficient data processing and trajectory prediction. For example, simply utilizing data from a single sensor type fails to fully and accurately describe the target object's state. Alternatively, trajectory prediction fails to fully consider the relationship between target position, robotic arm joint angles, and target velocity, resulting in inaccurate predictions.

[0005] Existing methods often fail to effectively classify and analyze historical data, failing to fully utilize the information contained in the data. This results in the robot arm being unable to respond quickly and accurately to similar working conditions. In complex industrial environments, the robot arm must process large amounts of data and adjust its movements in real time based on this data, placing high demands on the efficiency and accuracy of data processing. Existing control methods struggle to meet the data processing efficiency requirements of actual production, resulting in slow robot arm responses and impacting production efficiency.

[0006] Existing robotic arm control methods lack adaptability to different work scenarios and require manual adjustment of control parameters, increasing operational difficulty and labor costs. For example, when the type of target object in the work environment changes, traditional methods require rewriting the program or adjusting parameters, which is not only time-consuming and labor-intensive, but also prone to errors.

[0007] Existing robotic arm automatic control methods have many deficiencies in dynamic target tracking, data processing, trajectory prediction, and adaptability, failing to meet the demands of modern industrial automation. Therefore, a control method and system that can efficiently process multi-source sensor data, accurately predict robotic arm trajectories, and exhibit good adaptability is urgently needed. Summary of the Invention

[0008] The purpose of the present invention is to provide a method and system for automatic control of a robotic arm based on target recognition, so as to solve the problems raised in the above background technology.

[0009] To achieve the above object, the present invention provides the following technical solution: a method for automatic control of a robotic arm based on target recognition, the method comprising:

[0010] Collect the position data, speed data of the target object and the joint angle data of the robotic arm through different types of sensing devices;

[0011] The next moment after the current moment is recorded as the control moment. For a preset number of moments before the current moment, the predicted values ​​of the target position data, the robot arm joint angle data, and the target speed data at the control moment are obtained through the prediction model; different robot arm trajectory prediction values ​​are obtained based on the matching degree between the target position data and the robot arm joint angle data, the matching degree between the target position data and the target speed data, and the predicted values ​​of the robot arm joint angle data and the target position data at the control moment;

[0012] The predicted value of the manipulator joint angle data at the control moment is corrected by combining the respective proportions of the correlation between the target position data and the manipulator joint angle data, and the correlation between the target position data and the target speed data with different manipulator trajectory prediction values ​​to obtain a corrected prediction value;

[0013] The position data, speed data and joint angle data of the target object at each moment are collected by different types of sensing devices to form a three-dimensional data set, and all the three-dimensional data sets are clustered;

[0014] For the cluster group at the current moment, the 3D data set is composed of different sequences based on the target position, robot arm joint angle, and target speed. Sequences of different lengths are analyzed to obtain the similarity between the 3D data sets of different sequence lengths. The actual trajectory value is then combined to form a weight to obtain the historical prediction value of the robot arm joint angle data.

[0015] The robot arm trajectory prediction value is obtained according to the similarity between the corrected prediction value, the historical prediction value and the three-dimensional data set; and the robot arm is controlled by performing motion planning according to the robot arm trajectory prediction value.

[0016] Preferably, the method of collecting the position data, speed data and joint angle data of the target object at each moment and the robot arm by different types of sensing devices is as follows:

[0017] A visual sensor and a speed sensor are set in the target object's activity area, and an angle sensor is set on the robot arm body. Each sensor collects data once every preset period, and collects a certain amount of data. The data collected at all times are recorded as a sequence, and the target position sequence, robot arm angle sequence and target speed sequence are obtained respectively.

[0018] Preferably, the method for obtaining different robot arm trajectory prediction values ​​according to the matching degree between the target position data and the robot arm joint angle data, the matching degree between the target position data and the target speed data, and the predicted values ​​of the robot arm joint angle data and the target position data at the control moment is:

[0019] Calculate the matching degree of the sequence values ​​of the target position sequence and the robot arm angle sequence at the same time, and form a sequence of the matching degrees at all times, which is recorded as the position angle matching sequence; calculate the matching degree of the sequence values ​​of the target position sequence and the target velocity sequence at the same time, and form a sequence of the matching degrees at all times, which is recorded as the position velocity matching sequence;

[0020] The average value of all sequence values ​​in the position angle matching sequence is recorded as the first matching degree, and the average value of all sequence values ​​in the position velocity matching sequence is recorded as the second matching degree;

[0021] The second robot arm trajectory prediction value and the third robot arm trajectory prediction value are obtained according to the first matching degree, the second matching degree, and the predicted values ​​of the robot arm angle sequence and the target speed sequence at the control moment.

[0022] Preferably, the method for obtaining the second manipulator trajectory prediction value and the third manipulator trajectory prediction value according to the first matching degree, the second matching degree, and the predicted values ​​of the manipulator angle sequence and the target speed sequence at the control moment is:

[0023] The predicted values ​​of the robot arm angle sequence at the control moment are proportionally mapped by the first matching degree to obtain the second robot arm trajectory prediction value, and the predicted values ​​of the target speed sequence at the control moment are proportionally mapped by the second matching degree to obtain the third robot arm trajectory prediction value.

[0024] Preferably, the method for correcting the predicted value of the manipulator joint angle data at the control moment by combining the respective proportions of the correlation degree between the target position data and the manipulator joint angle data and the correlation degree between the target position data and the target speed data with different manipulator trajectory prediction values ​​to obtain the corrected predicted value is:

[0025] Calculate the fluctuation range of all sequence values in the bit angle matching sequence, and take the reciprocal of the fluctuation range as the correlation degree of the target position sequence and the mechanical arm angle sequence;

[0026] Calculate the fluctuation range of all sequence values in the bit speed matching sequence, and take the reciprocal of the fluctuation range as the correlation degree of the target position sequence and the target speed sequence;

[0027] Record the predicted value of the mechanical arm joint angle data at the control moment obtained through the prediction model as the first mechanical arm prediction trajectory;

[0028] For the first mechanical arm prediction trajectory, preset a basic weight, and obtain the distribution weight of the second mechanical arm trajectory prediction value and the third mechanical arm trajectory prediction value according to the ratio of the correlation degrees;

[0029] Add the first mechanical arm prediction trajectory, the second mechanical arm trajectory prediction value, the third mechanical arm trajectory prediction value and their respective distribution weights to obtain the corrected prediction value.

[0030] Preferably, for the cluster group where the current moment is located, different sequences are formed according to the target position, the mechanical arm joint angle and the target speed from the three-dimensional data set, the similarity between the three-dimensional data sets under different sequence lengths is obtained by analyzing different length sequences, and the method for obtaining the historical prediction value of the mechanical arm joint angle data by combining the actual trajectory value to form the weight is:

[0031] Record the three-dimensional data set corresponding to the current moment as the target data set, obtain the cluster group where the target data set is located, and record all data sets in this cluster group as group data sets;

[0032] The group data set selects the position and speed of a preset period to obtain the target position sequence, the mechanical arm angle sequence and the target speed sequence corresponding to the group data set;

[0033] According to the similarity of the corresponding sequences of the group data set and the target data set, the similarity between the three-dimensional data sets is obtained, and the similar data sets are determined.

[0034] According to the trajectory value of the corresponding similar data set at the next moment under different sequence lengths, the historical prediction value of the mechanical arm angle sequence is obtained.

[0035] Preferably, the method for obtaining the similarity between the three-dimensional data sets according to the similarity of the corresponding sequences of the group data set and the target data set is:

[0036] For each group data set, calculate the similarity of the corresponding target position sequence, mechanical arm angle sequence and target speed sequence with the target position sequence, mechanical arm angle sequence and target speed sequence at the current moment, respectively;

[0037] The similarities between the three sequences of the group dataset and the target dataset are averaged and then linearly normalized to obtain the similarity between the group dataset and the target dataset;

[0038] The similarity between all group data sets and the target data set is obtained. The similarity between all group data sets and the target data set is classified by the threshold partitioning algorithm. The mean similarity of each class after classification is obtained. The group data set in the class with the largest similarity mean is recorded as the similar data set.

[0039] Preferably, the method for obtaining the historical prediction value of the robot arm angle sequence according to the trajectory value of the similar data set corresponding to different sequence lengths at the next moment is:

[0040] For each sequence length, the mean of the trajectory values ​​of all similar datasets at the next adjacent moment is used as the prediction reference value at the control moment; under this sequence length, the dispersion of the trajectory values ​​of all similar datasets at the next moment is calculated; the inverse of the dispersion is used as the credibility parameter; the ratio of each credibility parameter to the sum of all credibility parameters is used as the sequence weight;

[0041] The historical prediction value is obtained by weighting the prediction reference values ​​under all sequence lengths by the sequence weight.

[0042] Preferably, the method for obtaining the robot arm trajectory prediction value based on the similarity between the corrected prediction value, the historical prediction value and the three-dimensional data set is:

[0043] Obtain all similar data sets under the current sequence length, and obtain the revised prediction values ​​and historical prediction values ​​of similar data sets;

[0044] The corrected prediction value and historical prediction value of each similar data set are subtracted from the actual trajectory value at the next moment to form the corrected deviation sequence and the historical deviation sequence;

[0045] For each deviation sequence, the product of the dispersion of the deviation sequence and the average value of all sequence values ​​in the deviation sequence is used as the weight factor;

[0046] The ratio of the weight factor corresponding to the revised deviation sequence to the sum of the weight factors is used as the weight coefficient of the revised forecast value, and the ratio of the weight factor corresponding to the historical deviation sequence to the sum of the weight factors is used as the weight coefficient of the historical forecast value;

[0047] The corrected prediction value and the historical prediction value are weighted and added together by their respective weight coefficients to obtain the robot arm trajectory prediction value.

[0048] Preferably, the present invention also includes a robotic arm automatic control system based on target recognition, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, and when the processor executes the computer program, the steps of the above-mentioned robotic arm automatic control method based on target recognition are implemented.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The target recognition-based automatic control method and system for a robotic arm, provided by this invention, uses different types of sensing devices to collect position and velocity data of the target object, as well as the joint angle data of the robotic arm, providing comprehensive and accurate data support for the precise control of the robotic arm. The combined use of multiple sensor types enables acquisition of target and robotic arm status information from multiple dimensions, effectively overcoming the limitations of single sensor data and enabling the system to more comprehensively understand the dynamic changes in the working environment and the target object.

[0051] In terms of data processing, a prediction model is used to calculate the predicted values ​​of the target position, manipulator joint angle, and target speed at the control moment. These values ​​are then corrected based on the matching and correlation between the different data points, improving the accuracy of trajectory prediction. This approach fully considers the inherent connections between the target position and manipulator joint angles, and between the target position and target speed. By analyzing the matching and correlation, the relationship between the data can be more accurately grasped, resulting in a corrected prediction value that is closer to the actual situation.

[0052] The three-dimensional dataset is clustered and analyzed for similarities across different sequence lengths. The actual trajectory values ​​are then weighted to generate historical predictions. This process fully leverages the information contained in the historical data. Cluster analysis groups similar operating condition data into a single category. When processing current data, similar historical data can be quickly found, allowing the experience gained from this data to predict the robot's trajectory. Similarity analysis across different sequence lengths can exploit data features at different time scales, making historical predictions more accurate and reliable.

[0053] The robot's trajectory is predicted based on the corrected predictions, historical predictions, and similarities between 3D datasets. This prediction is then used for motion planning and control, enabling the robot to more accurately track the target object. This trajectory prediction method, which considers multiple factors, effectively improves the robot's motion accuracy and response speed, enabling it to accurately complete various operational tasks in a dynamically changing working environment.

[0054] The method and system have good adaptability and can automatically adjust the control strategy according to different working scenes and changes of target objects. Through real-time processing and analysis of data, the system can continuously learn and optimize the control parameters, improve the control effect of the mechanical arm under different working conditions, reduce manual intervention, and reduce the operation difficulty and cost.

[0055] In addition, the method has significant improvement in data processing efficiency, can quickly process a large amount of sensor data, and generate accurate trajectory prediction values in real time, meeting the requirements of fast response of the mechanical arm in industrial production, improving production efficiency and quality. In summary, the present application can effectively solve the problems of inaccurate dynamic target tracking, low data processing efficiency, poor adaptability and other problems in the traditional mechanical arm control method, and has wide application prospect and significant economic benefit. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 The working principle diagram of the target recognition-based automatic control method of the mechanical arm is shown in the present application.

[0057] Figure 2 The working principle diagram of the data acquisition of the sensing device is shown in the present application.

[0058] Figure 3 The working principle diagram of the trajectory prediction value calculation of the mechanical arm is shown in the present application.

[0059] Figure 4 The working principle diagram of the final trajectory prediction value calculation of the mechanical arm is shown in the present application. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0061] Please refer to Figures 1-4 The target recognition-based automatic control method of the mechanical arm according to the present application is described as follows:

[0062] The position data, speed data of the target object and the joint angle data of the mechanical arm are collected by different types of sensing devices.

[0063] The next moment after the current moment is recorded as the control moment. For a preset number of moments before the current moment, the predicted values ​​of the target position data, the robot arm joint angle data, and the target speed data at the control moment are obtained through the prediction model; different robot arm trajectory prediction values ​​are obtained based on the matching degree between the target position data and the robot arm joint angle data, the matching degree between the target position data and the target speed data, and the predicted values ​​of the robot arm joint angle data and the target position data at the control moment;

[0064] The predicted value of the manipulator joint angle data at the control moment is corrected by combining the respective proportions of the correlation between the target position data and the manipulator joint angle data, and the correlation between the target position data and the target speed data with different manipulator trajectory prediction values ​​to obtain a corrected prediction value;

[0065] The target position, robot arm joint angle, and target speed at each moment are combined into a three-dimensional data set, and all three-dimensional data sets are clustered. For the cluster group at the current moment, the three-dimensional data set is divided into different sequences based on the target position, robot arm joint angle, and target speed. Sequences of different lengths are analyzed to obtain the similarity between three-dimensional data sets of different sequence lengths. The similarity is then combined with the actual trajectory value to form a weight to obtain the historical prediction value of the robot arm joint angle data.

[0066] The robot arm trajectory prediction value is obtained according to the similarity between the corrected prediction value, the historical prediction value and the three-dimensional data set; and the robot arm is controlled by performing motion planning according to the robot arm trajectory prediction value.

[0067] Example 1:

[0068] In this embodiment, the data collection process of the sensing device is described in detail. First, a visual sensor and a speed sensor are appropriately placed in the target object's activity area, and an angle sensor is placed on the robotic arm body to achieve comprehensive data collection of the target object and the robotic arm.

[0069] When setting up a visual sensor, considering the scope and characteristics of the target object's activity area, it is necessary to ensure that the visual sensor's field of view can cover the entire target object's activity area, thereby ensuring that the target object can be accurately captured at any position within the activity area. For example, if the target object's activity area is a rectangular space, the visual sensor can be installed at the upper center of the rectangular space to achieve comprehensive monitoring of the entire area. The selected visual sensor must have a high resolution and frame rate to ensure the accuracy and real-time nature of the collected target object position data. Industrial cameras are a commonly used visual sensor with high image acquisition accuracy and speed, which can meet the needs of collecting target object position data in robotic arm automatic control.

[0070] The speed sensor's placement also needs to consider the target object's motion characteristics. To accurately obtain the target object's velocity data, the speed sensor's measurement direction should align with the target object's primary direction of motion. If the target object may move in multiple directions, multiple speed sensors can be used to measure the target object's velocity in multiple directions. A laser Doppler velocimeter is a commonly used speed sensor. It determines the object's velocity by measuring the Doppler frequency shift caused by laser light irradiating the moving object. It offers advantages such as high measurement accuracy and fast response time.

[0071] When installing angle sensors on a robotic arm, they must be installed at each joint to measure the angle of each joint in real time. The joints of a robotic arm are crucial for achieving motion, and changes in the angle of each joint affect the position and posture of the end arm. High-precision encoders are commonly used angle sensors, converting the rotation angle of the robotic arm's joints into electrical signals, enabling precise measurement of joint angles.

[0072] The acquisition cycle for each sensor is a crucial parameter in the data collection process. The preset cycle setting should be determined based on the actual application scenario and control accuracy requirements. If high-precision control is required, a shorter preset cycle is recommended to achieve high-frequency data acquisition. If lower-precision control is required, a longer preset cycle can be used to reduce the amount of data collected and processed. For example, in industrial automation scenarios with high control accuracy requirements, a preset cycle of 0.01 seconds can ensure rapid response and precise control of the robot arm over the target object.

[0073] During data acquisition, each sensor collects data according to a preset cycle and records the collected data in chronological order to form a corresponding sequence. For the target object's position data, each acquired position information contains the target object's coordinate values ​​in three-dimensional space. Arranging these coordinate values ​​collected at all times in chronological order forms the target position sequence. The robot arm angle sequence is formed by recording the angle measurements of each joint at different times in chronological order. The target velocity sequence is obtained by arranging the target object's velocity values ​​measured by the velocity sensor at different times in chronological order.

[0074] To ensure data accuracy and reliability, the impact of interfering factors must be considered during data collection. For example, ambient lighting changes can affect the accuracy of visual sensors, while vibrations during robotic arm movement can affect angle sensor measurements. Therefore, appropriate measures must be taken to mitigate the impact of these interfering factors. For visual sensors, light shields or adjustments to the light source can be used to ensure consistent lighting. For angle sensors, vibration damping can be implemented in the robotic arm joints to reduce the impact of vibration on measurement results.

[0075] Furthermore, data storage and transmission are crucial components of the data collection process. Collected data must be stored promptly and accurately in memory for subsequent processing and analysis. Furthermore, data integrity and security must be maintained during transmission to prevent data loss or tampering. High-speed data transmission interfaces and reliable data transmission protocols can be used to achieve efficient and secure data transmission.

[0076] Through the rational configuration of sensing devices, the scientific determination of acquisition cycles, and strict control of the data acquisition, storage, and transmission processes, comprehensive, real-time, and accurate data collection of the target object's position and velocity data, as well as the robot arm's joint angles, is achieved. This collected data forms the basis for subsequent prediction and control, providing strong data support for the robot arm's accurate motion planning and control based on the target object's state. Throughout the data acquisition process, every link is closely linked, and negligence in any link can affect data quality and, in turn, the robot arm's control effectiveness. Therefore, strict management and control of the data acquisition process is necessary to ensure data accuracy and reliability.

[0077] Example 2:

[0078] In this embodiment, the process of obtaining the corrected prediction value is described in detail. This process requires a matching analysis of the target position sequence, the robot arm angle sequence, and the target velocity sequence, combined with correlation calculation and weight assignment, to achieve correction of the robot arm joint angle prediction value.

[0079] Calculate the matching degree of the sequence values ​​of the target position sequence and the robot arm angle sequence at the same time. Specifically, for each sampling moment, the current value in the target position sequence needs to be matched with the corresponding value in the robot arm angle sequence. The matching degree can be calculated using a variety of algorithmic logics. For example, the absolute value of the difference between the two values ​​is calculated to reflect the degree of difference. The smaller the difference, the higher the matching degree; or the value is converted into a vector form and the directional consistency is measured by cosine similarity. The matching results of each moment are arranged in chronological order to form a position angle matching sequence. Similarly, the same operation is performed on the target position sequence and the target velocity sequence, and the matching degree of the sequence values ​​at the same time is calculated to form a position velocity matching sequence. The construction of these two sequences provides the data foundation for the subsequent matching degree analysis.

[0080] Perform statistical processing on the two matching sequences. Calculate the average value of all sequence values ​​in the position-angle matching sequence and record it as the first matching degree. This average value can reflect the matching trend between the target position and the robot arm angle in the overall time dimension. For example, if the values ​​in the position-angle matching sequence are mostly concentrated at a high level, then the first matching degree will also be correspondingly high, indicating that the two are well matched at most times. Similarly, calculate the average value of the position-velocity matching sequence to obtain the second matching degree, which is used to characterize the overall matching trend between the target position and the target velocity.

[0081] Based on the first and second matching degrees, a preliminary derivation of the predicted arm trajectory is performed. The predicted values ​​of the arm angle sequence at the control moment (i.e., the values ​​pre-derived by the prediction model) are proportionally mapped using the first matching degree. The proportional mapping logic here is as follows: the first matching degree reflects the degree of match between the target position and the arm angle. When the matching degree is high, the predicted arm angle should be closer to the result derived based on the target position; otherwise, the mapping ratio needs to be adjusted. For example, if the first matching degree is 0.7, it means that the target position and the arm angle have an overall 70% match consistency. In this case, the predicted arm angle is multiplied by 0.7 to obtain the second predicted arm trajectory value. Similarly, the predicted values ​​of the target velocity sequence at the control moment are proportionally mapped using the second matching degree to obtain the third predicted arm trajectory value. This step achieves the conversion from matching degree to predicted trajectory value, providing a multi-dimensional prediction reference for subsequent corrections.

[0082] Calculate the correlation between the target position and the robot arm angle and target speed. For the position angle matching sequence, it is necessary to determine the fluctuation range of all its sequence values, that is, the difference between the maximum and minimum values ​​in the sequence. The size of the fluctuation range reflects the stability of the matching degree. The smaller the fluctuation range, the smaller the change in the matching degree at different times, and the higher the correlation between the two. Therefore, the reciprocal of the fluctuation range is used as the correlation between the target position sequence and the robot arm angle sequence. For example, if the fluctuation range of the position angle matching sequence is 0.2, its reciprocal 5 means that the correlation between the two is 5. The same method is applied to the position speed matching sequence, and the reciprocal of its fluctuation range is calculated to obtain the correlation between the target position and the target speed. These two correlation values ​​reflect the degree of dependence between different data pairs and provide a basis for weight allocation.

[0083] After obtaining the correlation, weight assignment and prediction value correction begin. First, the predicted value of the robot arm joint angle data at the control moment, obtained by the prediction model, is recorded as the first robot arm predicted trajectory. A base weight is preset for this predicted trajectory. The setting of this base weight takes into account the reliability of the prediction model and historical data experience. For example, it can be set to 0.4. This weight represents the basic proportion of the first robot arm predicted trajectory in the final corrected prediction value.

[0084] The weights assigned to the second and third robotic arm trajectory prediction values ​​are determined based on the correlation ratio. Assuming the correlation between the target position and the robotic arm angle is 5, and the correlation with the target velocity is 4, with a ratio of 5:4, the weight assigned to the second robotic arm trajectory prediction value is 5 / (5+4), and the weight assigned to the third robotic arm trajectory prediction value is 4 / (5+4). This weight distribution method based on the correlation ratio ensures that the trajectory prediction values ​​corresponding to data with high correlation receive greater weight during the correction process, thereby more fully reflecting the actual dependency relationship between the data.

[0085] The three predicted trajectory values are weighted and added to obtain a corrected predicted value. The specific calculation process is as follows: the first robot arm predicted trajectory is multiplied by its basic weight, and then the second robot arm trajectory predicted value is multiplied by its assigned weight, and then the third robot arm trajectory predicted value is multiplied by its assigned weight, and the sum of the three is the final corrected predicted value. For example, if the first robot arm predicted trajectory is 30 degrees, the second robot arm trajectory predicted value is 24 degrees, and the third robot arm trajectory predicted value is 20 degrees, the corresponding weights are 0.4, 5 / 9*0.6, and 4 / 9*0.6, respectively. The corrected predicted value is obtained by weighted calculation. This correction process considers multiple dimensions of prediction information and dynamically adjusts the weights according to the correlation between data, so that the corrected predicted value can more accurately reflect the actual trend of the robot joint angle at the control time, providing a more reliable basis for subsequent robot motion planning. The entire process of obtaining the corrected predicted value realizes the optimization of the predicted value through rigorous data processing logic and weight distribution mechanism, improving the accuracy and reliability of robot control.

[0086] Embodiment 3:

[0087] In this embodiment, the process of obtaining the historical predicted value is described in detail. This process needs to be based on the clustering analysis of three-dimensional data sets, combined with sequence similarity calculation and weight distribution, to realize the derivation of the historical predicted value of the robot joint angle data.

[0088] The three-dimensional data set corresponding to the current time is determined as the target data set. The three-dimensional data set is composed of the target position, the robot joint angle and the target speed at the current time, and these three dimensions of data collectively reflect the state of the target object and the robot at the current time. By clustering all three-dimensional data sets using a clustering algorithm, data with similar characteristics can be grouped into the same class, thereby determining the cluster group to which the target data set belongs. The selection of the clustering algorithm needs to consider the data characteristics and computational efficiency, for example, the K-means clustering algorithm, which divides the data into K clusters through iterative calculation, so that the data in the same cluster has high similarity, and the data in different clusters has low similarity. After determining the cluster group, all data sets in the group are recorded as group data sets, which have similar characteristics with the target data set and provide a data basis for subsequent similarity analysis.

[0089] The group data sets are processed in sequence. The group data sets need to select the position and speed data of a preset period to form the corresponding target position sequence, robot angle sequence and target speed sequence. The selection of the preset period needs to be determined according to the actual application scenario and data characteristics, for example, the data of the first 10 sampling periods can be selected to ensure that the sequence can reflect certain trends and rules. By arranging the position and speed data in the group data set in chronological order, the sequence corresponding to the target data set is formed, which is ready for subsequent similarity calculation.

[0090] Calculate the similarity between the group dataset and the target dataset. For each group dataset, calculate the similarity between its corresponding target position sequence, robotic arm angle sequence, and target velocity sequence and the current target position sequence, robotic arm angle sequence, and target velocity sequence, respectively. Similarity can be calculated using various methods, such as the dynamic time warping (DTW) algorithm, which effectively handles similarity comparisons between sequences of different lengths and calculates similarity by finding the optimal matching path between time series; or using the Euclidean distance, which calculates the sum of the squared distances between the corresponding time values ​​of the two sequences; the smaller the distance, the higher the similarity. Average the similarities of the three sequences between the group dataset and the target dataset to obtain the overall similarity between the group dataset and the target dataset. To facilitate comparison and classification, the overall similarity is linearly normalized and mapped to the interval [0, 1] to make the similarities between different group datasets comparable.

[0091] After obtaining the similarities between all group datasets and the target dataset, a thresholding algorithm is used to classify the similarities. This algorithm can employ a K-means clustering algorithm or a density-based clustering algorithm, such as the DBSCAN algorithm, to classify the datasets into different categories based on the distribution characteristics of the similarities. By analyzing the mean similarity values ​​for each category, the category with the largest mean similarity value is selected as the similar dataset. These similar datasets have a high degree of similarity with the target dataset and can better reflect the characteristics and trends of the target dataset, providing a more reliable reference for subsequent historical forecast calculations.

[0092] Calculate historical prediction values ​​for different sequence lengths. For each preset sequence length (e.g., 5 sampling periods, 10 sampling periods, etc.), process similar datasets within that sequence length. First, take the average of the trajectory values ​​(i.e., the actual trajectory values ​​of the robot arm joint angles at the control moment) of all similar datasets at the next adjacent moment and use this as the prediction reference value for that sequence length. This prediction reference value reflects the average level of the trajectory values ​​of similar datasets at the next moment within that sequence length.

[0093] Calculate the dispersion of the trajectory values ​​of all similar datasets at the next moment within the sequence length. Dispersion can be measured using standard deviation or variance, reflecting the degree of dispersion of the trajectory values. Smaller dispersion indicates a more concentrated trajectory value across similar datasets, leading to higher credibility; larger dispersion indicates a more dispersed trajectory value, leading to lower credibility. Therefore, the inverse of dispersion is used as the credibility parameter: smaller dispersion indicates a larger credibility parameter, and vice versa.

[0094] Calculate sequence weight. Take the ratio of each confidence parameter to the sum of all confidence parameters as the sequence weight of the sequence length. For example, if there are two sequence lengths, and the corresponding confidence parameters are 0.8 and 0.4 respectively, then their sequence weights are 0.8 / (0.8+0.4) and 0.4 / (0.8+0.4) respectively. Sequence weight reflects the importance of prediction reference value under different sequence lengths in historical prediction value. The higher the confidence of the sequence length, the greater the corresponding sequence weight.

[0095] Sum the prediction reference values under all sequence lengths by weighting with the corresponding sequence weight to obtain the historical prediction value of the joint angle data of the robot arm. For example, if there are two sequence lengths, and the prediction reference values are 25 degrees and 28 degrees respectively, and the sequence weights are 0.6 and 0.4 respectively, then the historical prediction value is 25x0.6+28x0.4. In this way, the information of similar data sets under different sequence lengths is considered comprehensively, and the trends and rules in historical data are fully utilized, so that the historical prediction value can more accurately reflect the trend of the joint angle of the robot arm, providing an important reference for subsequent trajectory prediction of the robot arm. The entire process of obtaining the historical prediction value is strictly processed through multiple links such as cluster analysis, similarity calculation, dispersion evaluation and weight distribution, ensuring the reliability and accuracy of the historical prediction value, and providing strong support for the automatic control of the robot arm.

[0096] Embodiment 4:

[0097] In this embodiment, the process of obtaining the trajectory prediction value of the robot arm is described in detail. This process needs to be based on the modified prediction value and the historical prediction value of the similar data set, combined with bias analysis and weight distribution, to realize accurate prediction of the trajectory of the robot arm.

[0098] Determine all similar data sets under the current sequence length. The determination of similar data sets needs to be based on the previous cluster analysis and similarity calculation, that is, from the cluster group where the target data set is located, the group data sets with high similarity to the target data set are selected through the threshold division algorithm, and these group data sets constitute the similar data set. For example, assuming that the current sequence length is set to 10 sampling periods, and through the previous cluster and similarity calculation, 15 group data sets with high similarity to the target data set are determined, which are the similar data sets under the current sequence length.

[0099] Corrected prediction values ​​and historical prediction values ​​are extracted from each similar dataset. The corrected prediction value is obtained by correcting the predicted value of the robot arm joint angle data at the control moment. The calculation process involves matching analysis of the target position, robot arm angle, and target speed, as well as the assignment of correlation weights. The historical prediction value is calculated based on the clustering group similarity of the three-dimensional dataset and the mean trajectory value under different sequence lengths. For example, in these 15 similar datasets, each dataset contains corresponding corrected prediction values ​​and historical prediction values. For example, the corrected prediction value of the first similar dataset is 26.5 degrees, and the historical prediction value is 27.2 degrees; the corrected prediction value of the second similar dataset is 25.8 degrees, and the historical prediction value is 26.9 degrees, and so on.

[0100] Calculate the corrected deviation sequence and the historical deviation sequence. For each similar dataset, subtract its corrected prediction value from the actual trajectory value at the next moment to obtain the corrected deviation value; subtract its historical prediction value from the actual trajectory value at the next moment to obtain the historical deviation value. Arrange the corrected deviation values ​​of all similar datasets in sequence to form a corrected deviation sequence; and arrange all historical deviation values ​​in sequence to form a historical deviation sequence. For example, assuming the actual trajectory value at the next moment of the first similar dataset is 26.8 degrees, its corrected deviation value is 26.5-26.8=-0.3 degrees, and the historical deviation value is 27.2-26.8=0.4 degrees. For the second similar dataset, the actual trajectory value at the next moment is 26.2 degrees, its corrected deviation value is 25.8-26.2=-0.4 degrees, and the historical deviation value is 26.9-26.2=0.7 degrees. And so on, ultimately forming a corrected deviation sequence and a historical deviation sequence containing 15 data points.

[0101] Calculate the weighting factor. For the corrected deviation sequence, first calculate its dispersion. Dispersion can be measured by calculating the standard deviation of the sequence, reflecting the degree of dispersion of the deviation values. For example, if the values ​​of the corrected deviation sequence are [-0.3, -0.4, 0.1, -0.2, 0.3, -0.1, 0.2, -0.3, 0.1, -0.2, 0.2, -0.1, 0.3, -0.2, 0.1]. Calculate its standard deviation to determine the dispersion. Then, calculate the average of all the values ​​in the deviation sequence. For example, the average of the corrected deviation sequence above is [(-0.3) + (-0.4) + 0.1 + (-0.2) + 0.3 + (-0.1) + 0.2 + (-0.3) + 0.1 + (-0.2) + 0.2 + (-0.1) + 0.3 + (-0.2) + 0.1] / 15 to obtain the average value. Multiply the dispersion and the average value to obtain the weighting factor of the corrected deviation sequence. The same method is applied to the historical deviation series to calculate its dispersion, average value, and weight factor. The size of the weight factor reflects the comprehensive characteristics of the deviation series. The greater the dispersion and the larger the average value, the larger the weight factor, and vice versa.

[0102] Calculate the weight coefficients. The ratio of the weight factor corresponding to the corrected deviation sequence to the sum of the weight factors of the corrected deviation sequence and the historical deviation sequence is used as the weight coefficient for the corrected forecast value. The ratio of the weight factor corresponding to the historical deviation sequence to the sum of the weight factors of the two is used as the weight coefficient for the historical forecast value. For example, if the weight factor of the corrected deviation sequence is 0.5 and the weight factor of the historical deviation sequence is 0.7, then the weight coefficient of the corrected forecast value is 0.5 / (0.5+0.7), and the weight coefficient of the historical forecast value is 0.7 / (0.5+0.7). The calculation of the weight coefficients reflects the contribution ratio of the two forecast values ​​to the final trajectory forecast value. Forecast values ​​with better deviation sequence characteristics (such as small dispersion and average value close to zero) will receive higher weight coefficients.

[0103] Calculate the predicted trajectory of the robot arm. Multiply the revised and historical predictions by their respective weight coefficients and add them together to obtain the predicted trajectory. For example, assuming the average value of the revised prediction is 26.2 degrees, the average value of the historical prediction is 27.1 degrees, the weight coefficient of the revised prediction is 0.45, and the weight coefficient of the historical prediction is 0.55, then the predicted trajectory of the robot arm is 26.2 × 0.45 + 27.1 × 0.55. This weighted calculation comprehensively considers the deviation between the revised and historical predictions, making the final trajectory prediction closer to the actual trajectory value, providing a reliable basis for the robot arm's motion planning.

[0104] Throughout the entire process, accurate screening of similar datasets is fundamental and directly impacts the quality of the deviation sequence calculation. For example, insufficient similarity between the similar dataset and the target dataset may result in significant deviations between the corrected and historical deviation sequences and the actual trajectory values, thus affecting the calculation of weighting factors and coefficients. Rigorous deviation analysis is also crucial. The calculation of the discreteness and average value must accurately reflect the characteristics of the deviation sequence to ensure that the weighting factors can reasonably characterize the reliability of both predictions. The weight allocation mechanism dynamically adjusts the weights of the corrected and historical predictions to optimize the combination of different prediction information. This allows the robot trajectory prediction to fully leverage the patterns in historical data and the multi-dimensional information from the current correction process, improving prediction accuracy and adaptability. This process, without relying on specific formula derivation, achieves dynamic prediction of the robot trajectory through data processing logic and weight allocation strategies. This allows for better adaptation to real-time state changes of the target object and the robot arm, providing key technical support for the robot's automatic control.

[0105] Example 5:

[0106] In this embodiment, an implementation method of a robotic arm automatic control system based on target recognition is described in detail.

[0107] The system's hardware components include memory, a processor, and various sensors. Memory, a core component for data storage, can be a solid-state drive (SSD) or dynamic random access memory (DRAM). It stores collected raw data, processed intermediate data, and control programs. For example, during robotic arm operation, the memory needs to store real-time target position image data captured by the vision sensor, real-time velocity data streams from the velocity sensor, and historical measurements of the robotic arm's joint angles. This data volume may continue to increase with operational time, so the memory must have sufficient storage capacity and read / write speed. The processor is the computing core of the system and can be a high-performance central processing unit (CPU) or graphics processing unit (GPU), such as Intel's i7 series CPU or NVIDIA's RTX series GPU. It executes data processing algorithms, predictive model calculations, and motion planning logic. The processor must possess strong parallel computing capabilities to handle the real-time processing of multi-sensor data, such as simultaneous cluster analysis and similarity calculation of target position sequences, robotic arm angle sequences, and target velocity sequences.

[0108] The sensor component is closely linked to the data acquisition phase of the method. Visual sensors and speed sensors are deployed in the target object's activity area. For example, Basler's industrial cameras, which can capture 30 frames of images per second with a resolution of 1280×720, are used to extract the target object's position coordinates. An OPTEX laser speed sensor, with a measurement range of 0.1-10 m / s and an accuracy of ±0.5%, is used to obtain target speed data in real time. Angle sensors, such as Renishaw's incremental encoders, with a resolution of up to 0.001 degrees, are installed at each joint of the robotic arm to ensure accurate measurement of the robotic arm's joint angles. These sensors are connected to the processor via data interfaces. For example, the industrial camera transmits image data via a GigE interface, and the encoder transmits angle data via an RS-485 interface, ensuring real-time and stable data transmission.

[0109] The system's software component is represented by a computer program stored in memory and executed by a processor. This program, developed in programming languages ​​such as C++ or Python, includes a data acquisition module, a prediction model module, a cluster analysis module, a trajectory prediction module, and a motion planning module. The data acquisition module is responsible for controlling the sensor to collect data at a preset interval (e.g., 10ms) and organizing the raw data into a target position sequence, a robotic arm angle sequence, and a target velocity sequence. For example, after the visual sensor captures a frame of image, the data acquisition module extracts the pixel coordinates of the target object using image processing algorithms (such as edge detection and feature point recognition). These are then converted into three-dimensional position data in the world coordinate system using camera calibration parameters and stored in the target position sequence.

[0110] The prediction model module processes data from a preset number of moments (e.g., 20) prior to the current moment to obtain predicted values ​​for the target position, robotic arm joint angles, and target velocity at the control moment. This module can employ deep learning models such as recurrent neural networks (RNNs) or long short-term memory networks (LSTMs) to train model parameters using historical data. For example, an LSTM model can be trained using target position data from the previous 20 moments to predict the target position value for the next moment. Model training is performed offline, and the trained model parameters are directly used for prediction calculations during online operation.

[0111] The cluster analysis module clusters 3D datasets (each containing target position, robotic arm joint angles, and target velocity). For example, using the K-means clustering algorithm, the 3D dataset is divided into five clusters, each corresponding to a typical combination of target object and robotic arm state. When a new 3D dataset is generated, the cluster analysis module calculates its distance to each cluster center and determines the cluster group to which it belongs, providing a basis for subsequent similarity analysis.

[0112] The trajectory prediction module integrates the calculation logic for revised predictions and historical predictions. For example, assuming that the cluster group to which the current 3D dataset belongs contains 100 historical datasets at a specific control moment, the trajectory prediction module first calculates the position-angle matching sequence between the target position sequence and the robot arm angle sequence, obtaining a first matching degree of 0.75. It then calculates the position-velocity matching sequence, obtaining a second matching degree of 0.68. The predicted robot arm angle value (e.g., 30 degrees) is proportionally mapped using the first matching degree to obtain a second robot arm trajectory prediction value of 22.5 degrees. The predicted target velocity value (e.g., 0.5 m / s) is mapped using the second matching degree to obtain a third robot arm trajectory prediction value of 0.34 m / s. The fluctuation range of the position-angle matching sequence is calculated as 0.15, and its reciprocal, 6.67, is used as the correlation between the target position and the robot arm angle. The fluctuation range of the position-velocity matching sequence is calculated as 0.2, and its reciprocal, 5, is used as another correlation degree. Weights are assigned based on the correlation ratio of 6.67:5. Combined with the base weight of 0.4 for the first robot arm predicted trajectory, this weighted value yields a revised prediction value of 25.6 degrees. At the same time, 20 historical data sets with high similarity to the current data set were selected from the cluster group. For the case where the sequence length is 15, the mean of the trajectory values ​​of these similar data sets at the next moment is calculated to be 26.3 degrees, the discreteness is 0.8, and the credibility parameter is 1.25. Combined with the weights of other sequence lengths, the weighted historical prediction value is 26.1 degrees.

[0113] The motion planning module generates motion control instructions for the robotic arm based on the predicted trajectory value (e.g., 25.9 degrees) output by the trajectory prediction module. This module needs to consider the inverse kinematic solution of the robotic arm, convert the trajectory of the end effector into a sequence of angles for each joint, and incorporate velocity planning and acceleration planning to ensure the smoothness of the robotic arm's motion. For example, when the trajectory prediction value is 25.9 degrees for joint 1, the motion planning module calculates the motion trajectory from the current angle (e.g., 20 degrees) to the target angle, sets the time parameters for the acceleration, uniform speed, and deceleration segments, generates a smooth joint angle change curve, and sends it to the robotic arm's servo driver via a pulse signal or analog voltage signal to control the motor operation.

[0114] The system operates as follows: sensors collect data at a preset interval and transmit it to the processor. The data acquisition module organizes the data into sequences and stores them in memory. The prediction model module uses historical sequence data to calculate the predicted value for the control moment. The cluster analysis module determines the cluster group to which the current dataset belongs. The trajectory prediction module combines matching, correlation, and similarity to calculate the revised predicted value and the historical predicted value to obtain the predicted value of the robot arm trajectory. The motion planning module generates control instructions based on the predicted trajectory value to drive the robot arm. This entire process is repeated in a cycle, achieving real-time control of the robot arm.

[0115] In practical applications, the system's hardware selection must be tailored to the robot's load capacity, control accuracy, and operating environment. For example, high-precision assembly scenarios require higher-resolution angle sensors and more powerful processors; high-speed motion scenarios require increased sensor sampling frequency and data transmission bandwidth. Software algorithm parameters (such as preset cycles, number of clusters, and sequence length) also require optimization through debugging to accommodate the varying motion characteristics of the target object and the robot's dynamics.

[0116] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0117] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for automatic control of a robotic arm based on target recognition, characterized in that: The method comprises the following steps: Collect the position data, speed data of the target object and the joint angle data of the robotic arm through different types of sensing devices; The next moment after the current moment is recorded as the control moment. For a preset number of moments before the current moment, the predicted values ​​of the target position data, the robot arm joint angle data, and the target speed data at the control moment are obtained through the prediction model; Obtain different robot arm trajectory prediction values ​​according to the matching degree between the target position data and the robot arm joint angle data, the matching degree between the target position data and the target speed data, and the predicted values ​​of the robot arm joint angle data and the target position data at the control time; The predicted value of the manipulator joint angle data at the control moment is corrected by combining the respective proportions of the correlation between the target position data and the manipulator joint angle data, and the correlation between the target position data and the target speed data with different manipulator trajectory prediction values ​​to obtain a corrected prediction value; The position data, speed data and joint angle data of the target object at each moment are collected by different types of sensing devices to form a three-dimensional data set, and all the three-dimensional data sets are clustered; For the cluster group at the current moment, the 3D data set is composed of different sequences based on the target position, robot arm joint angle, and target speed. Sequences of different lengths are analyzed to obtain the similarity between the 3D data sets of different sequence lengths. The actual trajectory value is then combined to form a weight to obtain the historical prediction value of the robot arm joint angle data. The robot arm trajectory prediction value is obtained according to the similarity between the corrected prediction value, the historical prediction value and the three-dimensional data set; and the robot arm is controlled by performing motion planning according to the robot arm trajectory prediction value.

2. The method for automatic control of a robotic arm based on target recognition according to claim 1, wherein: The method for collecting the position data, speed data and joint angle data of the target object at each moment and the robot arm through different types of sensing devices is as follows: A visual sensor and a speed sensor are set in the target object's activity area, and an angle sensor is set on the robot arm body. Each sensor collects data once every preset period, and collects a certain amount of data. The data collected at all times are recorded as a sequence, and the target position sequence, robot arm angle sequence and target speed sequence are obtained respectively.

3. The method for automatic control of a robotic arm based on target recognition according to claim 2, wherein: The method for obtaining different robot arm trajectory prediction values ​​according to the matching degree between the target position data and the robot arm joint angle data, the matching degree between the target position data and the target speed data, and the predicted values ​​of the robot arm joint angle data and the target position data at the control time is: Calculate the matching degree of the sequence values ​​of the target position sequence and the robot arm angle sequence at the same time, and form a sequence of the matching degrees at all times, which is recorded as the position angle matching sequence; calculate the matching degree of the sequence values ​​of the target position sequence and the target velocity sequence at the same time, and form a sequence of the matching degrees at all times, which is recorded as the position velocity matching sequence; The average value of all sequence values ​​in the position angle matching sequence is recorded as the first matching degree, and the average value of all sequence values ​​in the position velocity matching sequence is recorded as the second matching degree; The second robot arm trajectory prediction value and the third robot arm trajectory prediction value are obtained according to the first matching degree, the second matching degree, and the predicted values ​​of the robot arm angle sequence and the target speed sequence at the control moment.

4. The method for automatic control of a robotic arm based on target recognition according to claim 3, wherein: The method for obtaining the second manipulator trajectory prediction value and the third manipulator trajectory prediction value according to the first matching degree, the second matching degree, and the predicted values ​​of the manipulator angle sequence and the target speed sequence at the control moment is: The predicted values ​​of the robot arm angle sequence at the control moment are proportionally mapped by the first matching degree to obtain the second robot arm trajectory prediction value, and the predicted values ​​of the target speed sequence at the control moment are proportionally mapped by the second matching degree to obtain the third robot arm trajectory prediction value.

5. The method for automatic control of a robotic arm based on target recognition according to claim 2, wherein: The method for correcting the predicted value of the robotic arm joint angle data at the control moment by combining the respective proportions of the correlation between the target position data and the robotic arm joint angle data and the correlation between the target position data and the target speed data with different robotic arm trajectory prediction values ​​to obtain the corrected predicted value is: Calculate the fluctuation range of all sequence values ​​in the position angle matching sequence, and use the inverse of the fluctuation range as the correlation between the target position sequence and the robot arm angle sequence; Calculate the fluctuation range of all sequence values ​​in the position-velocity matching sequence, and use the inverse of the fluctuation range as the correlation between the target position sequence and the target velocity sequence; The predicted value of the robot arm joint angle data at the control time obtained by the prediction model is recorded as the first robot arm predicted trajectory; A basic weight is preset for the predicted trajectory of the first robotic arm, and the distribution weights of the predicted trajectory value of the second robotic arm and the predicted trajectory value of the third robotic arm are obtained according to the ratio of the correlation degrees; The first robotic arm predicted trajectory, the second robotic arm trajectory predicted value, the third robotic arm trajectory predicted value and their corresponding allocation weights are weightedly added to obtain a corrected predicted value.

6. The method for automatic control of a robotic arm based on target recognition according to claim 2, wherein: For the cluster group at the current moment, the three-dimensional data set is composed of different sequences according to the target position, robot arm joint angle, and target speed. Sequences of different lengths are analyzed to obtain the similarity between the three-dimensional data sets under different sequence lengths. The method of combining the actual trajectory value to form a weight to obtain the historical prediction value of the robot arm joint angle data is as follows: The three-dimensional dataset corresponding to the current moment is recorded as the target dataset, the cluster group where the target dataset is located is obtained, and all datasets in this cluster group are recorded as group datasets; The group data set selects the position and speed of the preset period to obtain the target position sequence, the robot arm angle sequence and the target speed sequence corresponding to the group data set; Obtaining similarity between three-dimensional data sets based on similarity of corresponding sequences of the group data set and the target data set, and determining similar data sets; The historical prediction value of the robot arm angle sequence is obtained according to the trajectory value of the next moment of similar data sets corresponding to different sequence lengths.

7. The method for automatic control of a robotic arm based on target recognition according to claim 6, wherein: The method of obtaining the similarity between the three-dimensional datasets based on the similarity of the corresponding sequences of the group dataset and the target dataset and determining the similar datasets is as follows: For each group of data sets, the similarity between the corresponding target position sequence, robot arm angle sequence and target speed sequence and the current target position sequence, robot arm angle sequence and target speed sequence is calculated respectively; The similarities between the three sequences of the group dataset and the target dataset are averaged and then linearly normalized to obtain the similarity between the group dataset and the target dataset; The similarity between all group data sets and the target data set is obtained. The similarity between all group data sets and the target data set is classified by the threshold partitioning algorithm. The mean similarity of each class after classification is obtained. The group data set in the class with the largest similarity mean is recorded as the similar data set.

8. The method for automatic control of a robotic arm based on target recognition according to claim 6, wherein: The method for obtaining the historical prediction value of the robot arm angle sequence based on the trajectory value of the next moment of the similar data set corresponding to different sequence lengths is: For each sequence length, the mean of the trajectory values ​​of all similar datasets at the next adjacent moment is used as the prediction reference value at the control moment; under this sequence length, the dispersion of the trajectory values ​​of all similar datasets at the next moment is calculated; the inverse of the dispersion is used as the credibility parameter; the ratio of each credibility parameter to the sum of all credibility parameters is used as the sequence weight; The historical prediction value is obtained by weighting the prediction reference values ​​under all sequence lengths by the sequence weight.

9. The method for automatic control of a robotic arm based on target recognition according to claim 2, wherein: The method for obtaining the robot arm trajectory prediction value based on the similarity between the corrected prediction value, the historical prediction value and the three-dimensional data set is: Obtain all similar data sets under the current sequence length, and obtain the revised prediction values ​​and historical prediction values ​​of similar data sets; The corrected prediction value and historical prediction value of each similar data set are subtracted from the actual trajectory value at the next moment to form the corrected deviation sequence and the historical deviation sequence; For each deviation sequence, the product of the dispersion of the deviation sequence and the average value of all sequence values ​​in the deviation sequence is used as the weight factor; The ratio of the weight factor corresponding to the revised deviation sequence to the sum of the weight factors is used as the weight coefficient of the revised forecast value, and the ratio of the weight factor corresponding to the historical deviation sequence to the sum of the weight factors is used as the weight coefficient of the historical forecast value; The corrected prediction value and the historical prediction value are weighted and added together by their respective weight coefficients to obtain the robot arm trajectory prediction value.

10. A robot arm automatic control system based on target recognition, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method for automatic control of a robotic arm based on target recognition as described in any one of claims 1 to 9 are implemented.

Images