Mechanical arm operation track optimization method based on generative adversarial network

Abstract

The invention discloses a mechanical arm operation track optimization method based on a generative adversarial network. The method comprises the steps that mechanical arm operation track data are collected, a generative adversarial network model is constructed, and a generator generates a mechanical arm path track according to random noise and task characteristic parameters; distinguishing the generated path track and the real path track through a discriminator, and obtaining an optimization generation result; according to the optimization generation result, an optimal track meeting task requirements is screened out through track feature scoring and a screening mechanism; by means of the method, the efficiency and adaptability of mechanical arm trajectory planning are remarkably improved, task execution energy consumption is effectively reduced, task completion time is shortened, trajectory smoothness and accuracy are improved, and the method is suitable for mechanical arm operation optimization in the fields of industrial manufacturing, logistics carrying and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of robot arm operation trajectory optimization under machine learning, and in particular to a robot arm operation trajectory optimization method based on a generative adversarial network. Background Art

[0002] In recent years, with the rapid development of industrial automation and intelligent manufacturing technology, robotic arms, as key execution equipment, have been widely used in many fields such as production and manufacturing, logistics handling, and medical services. The operation trajectory of the robotic arm directly affects the efficiency and accuracy of task completion. Therefore, trajectory planning and optimization technology has gradually become a hot topic of research. Traditional trajectory planning methods are mostly based on deterministic algorithms, such as path calculation methods based on inverse kinematics and trajectory optimization methods based on graph theory. Although these methods can solve common trajectory generation problems, they have limited performance in complex environments (such as multi-objective optimization requirements or dynamic interference). In addition, with the rise of deep learning technology, more and more studies have introduced data-driven models such as neural network models into trajectory generation to improve the flexibility and adaptability of robotic arm operation trajectory planning.

[0003] Although the existing trajectory optimization technology has made some progress, it still has shortcomings in terms of adaptability and actual efficiency improvement in multi-task scenarios. Specifically, the trajectories generated by traditional algorithms often find it difficult to balance the complex relationship between path length, task execution time, etc., especially in high-precision tasks, where the jitter and discontinuity problems between trajectory points are particularly significant. In addition, although the trajectory generation method based on deep learning has strong generalization capabilities, it is limited by the efficiency of utilizing real task data, and the stability of the generated trajectory and the degree of physical constraint satisfaction are lacking. In this case, the generated trajectory may cause unstable motion of the robot arm in actual operation, thereby affecting the overall system performance. Therefore, how to improve the actual execution efficiency and task adaptability of trajectory planning is an important research direction in the current field. Summary of the invention

[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0005] In view of the above existing problems, the present invention is proposed. Therefore, the present invention provides a method for optimizing the operation trajectory of a robot arm based on a generative adversarial network to solve the problems mentioned in the background technology.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] The present invention provides a method for optimizing the operation trajectory of a robotic arm based on a generative adversarial network, comprising:

[0008] Collecting trajectory data of the robot arm during operation execution, predicting the current moment of the trajectory point in the trajectory data by using the Kalman filter method and making corrections to obtain a set of smooth trajectory points, normalizing the set of smooth trajectory points, and considering the eigenvector of the trajectory data;

[0009] Create a generative adversarial network model, take the feature vector of the trajectory data and the normalized smooth trajectory point set as an input set, input them into the generative adversarial network model, process the input set by the generator and the discriminator in the generative adversarial network model, design an alternating optimization mechanism for the generator and the discriminator, train the generative adversarial network model, and obtain trajectory characteristic parameters;

[0010] According to the trajectory characteristic parameters, the optimal trajectory of the robot arm when performing operation is selected, so as to improve the efficiency of the robot arm when performing operation.

[0011] As a preferred solution of the robot arm operation trajectory optimization method based on the generative adversarial network described in the present invention, wherein: the current moment of the trajectory point in the trajectory data is predicted and corrected by the Kalman filtering method to obtain a smooth trajectory point set, including:

[0012] The Kalman filter method is used to predict the current time of the trajectory point in the trajectory data, and at the same time, the observation value of the trajectory data is collected to correct the predicted value obtained at the current time of the trajectory point in the predicted trajectory data;

[0013] The predicted values ​​are taken as a set of smooth trajectory points.

[0014] As a preferred solution of the robot arm operation trajectory optimization method based on the generative adversarial network described in the present invention, wherein: the set of smooth trajectory points is normalized, and the characteristic vector of the trajectory data is considered at the same time, including:

[0015] Each smooth trajectory point in the smooth trajectory point set is normalized, and the path length and average movement speed of the robot arm during operation execution are calculated from the collected trajectory data to obtain a path length feature vector and a movement speed feature vector.

[0016] As a preferred solution of the robot arm operation trajectory optimization method based on the generative adversarial network described in the present invention, wherein: creating a generative adversarial network model, taking the feature vector of the trajectory data and the normalized smooth trajectory point set as the input set, including:

[0017] Combining the motion speed feature vector and the path length feature vector to obtain a task condition feature vector;

[0018] The normalized smooth trajectory point set and the task condition feature vector are used as the input set of the generative adversarial network model.

[0019] As a preferred solution of the robot arm operation trajectory optimization method based on the generative adversarial network described in the present invention, the input set is processed by the generator and the discriminator in the generative adversarial network model, and an alternating optimization mechanism of the generator and the discriminator is designed, including:

[0020] Introducing random noise into the generator, taking the task condition feature vector and the random noise as inputs of the generator, generating a path trajectory through a deep neural network and gradually outputting a smooth trajectory point sequence;

[0021] The obtained smooth trajectory point sequence is input into the discriminator. The discriminator judges the obtained smooth trajectory point sequence by collecting the real smooth trajectory point sequence to obtain the loss value of the generator. According to the loss value of the generator, the authenticity of the input smooth trajectory point sequence is judged.

[0022] As a preferred solution of the robot arm operation trajectory optimization method based on the generative adversarial network described in the present invention, wherein: an alternating optimization mechanism of the generator and the discriminator is designed, the generative adversarial network model is trained, and the trajectory characteristic parameters are obtained, including:

[0023] The generator is fixed, the discriminator is optimized, the real smooth trajectory point sequence and the obtained smooth trajectory point sequence are input, the discriminator loss value is obtained, and the generator is optimized by back-propagating the discriminator loss value;

[0024] The discriminator is fixed, the generator is optimized, the smooth trajectory point sequence is input, the generator loss value is obtained, and the discriminator is optimized by backpropagation of the generator loss value;

[0025] The process of optimizing the generator and the discriminator is iterated until the obtained smooth trajectory point sequence reaches the optimal state, and the trajectory feature parameters are output.

[0026] As a preferred solution of the robot arm operation trajectory optimization method based on the generative adversarial network described in the present invention, wherein: according to the trajectory characteristic parameters, the optimal trajectory when the robot arm operation is performed is screened, including:

[0027] A score is set for each trajectory feature parameter, and each trajectory feature parameter is arranged and combined to obtain a path trajectory after the arrangement and combination, and a score of the path trajectory is calculated. The path trajectory scores are arranged in descending order, and the highest score is selected as the optimal trajectory when the robot arm is operated.

[0028] Compared with the prior art, the invention has the following beneficial effects:

[0029] 1. By combining the generative adversarial network (GAN) and the trajectory feature screening mechanism, the optimal trajectory with adaptive characteristics can be generated according to different task requirements. Compared with the traditional algorithm, the present invention is more flexible in multi-objective optimization scenarios, especially under complex task conditions, and can balance the constraints between multiple trajectory feature parameters, thus improving the multi-task adaptation capability of trajectory generation;

[0030] 2. Scoring and screening trajectory feature parameters avoids interference of redundant trajectories on robot arm movement, thereby significantly reducing task execution time. At the same time, the trajectory optimization process focuses on smoothness and continuity, reduces jitter and errors between trajectory points, and improves the overall accuracy and efficiency of robot arm operations, enabling it to better adapt to high-precision industrial scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0032] Figure 1 The overall flow chart of a method for optimizing the operation trajectory of a robotic arm based on a generative adversarial network according to an embodiment of the present invention;

[0033] Figure 2 A flowchart of processing a smooth trajectory point sequence by a neural network of a robot arm operation trajectory optimization method based on a generative adversarial network according to an embodiment of the present invention;

[0034] Figure 3 This is a comparison diagram of the path trajectory and energy consumption of a robot arm operation trajectory optimization method based on a generative adversarial network described in an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0036] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0037] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0038] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0039] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0040] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0041] Example 1

[0042] Reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a method for optimizing the operation trajectory of a robotic arm based on a generative adversarial network, comprising:

[0043] S1. Collect the trajectory data of the robot arm during operation, predict the current moment of the trajectory point in the trajectory data by Kalman filtering method and make corrections to obtain a set of smooth trajectory points, normalize the set of smooth trajectory points, and consider the eigenvector of the trajectory data;

[0044] Specifically, a high-precision encoder is installed at the joint of the robot arm (to collect robot arm angle and robot arm speed data), a pressure sensor is installed at the end (to collect the contact force between the tool end and the environment), and a camera is configured (to collect visual data when the robot arm is operating) to obtain the trajectory data of the robot arm during operation.

[0045] Specifically, the trajectory data D of the robot arm during operation is expressed as:

[0046] D={(x i ,y i , z i );θ i , v i , t i , f i}

[0047] Among them, (x i ,y i , z i ) is expressed as spatial coordinates (x, y, z), which is obtained by collecting visual data of the robot arm during operation by the camera; θ i Expressed as the angle of the robotic arm; v i Expressed as the speed of the robot arm; t i Indicates the timestamp of the current state of the robot; f i It is expressed as the contact force between the tool end of the robot arm and the environment;

[0048] Furthermore, the current moment of the trajectory point in the trajectory data is predicted and corrected by the Kalman filter method to obtain a set of smooth trajectory points;

[0049] It should be explained that Kalman Filter is a recursive algorithm for dynamic system state estimation, which is mainly used in the case of noise or uncertainty in the system. It can obtain the optimal estimate of the system state by weighted average of the current measurement value and the prior estimate. Since the scheme of the present invention requires the use of sensors, encoders and cameras to collect data, and these devices may cause jumps in the collected data under the state of uneven sampling, and on the other hand, the device itself may not be accurate enough, resulting in unreasonable collected data. Therefore, it is necessary to use the Kalman filter method to dynamically predict and observe the trajectory data D.

[0050] Specifically, each data point in the trajectory data of the robot arm during operation execution is called a trajectory point, and the set of trajectory points formed by connecting all trajectory points is called a path trajectory, also called the operation trajectory of the robot arm;

[0051] Furthermore, the mathematical expression of the current moment of the trajectory point in the predicted trajectory data is:

[0052]

[0053] in, is the predicted value, A is the state transfer matrix, which represents the operation state of the robot from the previous moment to the current moment, Du t is the control quantity, which represents the control input of the robot at the current moment, and its input is the trajectory data D;

[0054] Furthermore, by collecting the observed values ​​of the trajectory data, the predicted values ​​are corrected The formula is:

[0055]

[0056] Among them, z t is the observed value of the collected trajectory data; K t is the Kalman gain, which is used to weigh the credibility of the predicted value and the observed value; H is the observation matrix, which is used to convert the state vector x t Mapped into the observation space; x t is the state vector, and the output result is represented as a smooth trajectory point;

[0057] It should be noted that although there are other data smoothing methods (such as low-pass filtering, wavelet transform, and moving average), they cannot replace Kalman filtering for the following reasons: First, low-pass filtering is suitable for processing frequency domain noise, but cannot capture nonlinear changes in dynamic trajectories; second, moving average will erase the dynamic characteristics of the trajectory through averaging processing and is difficult to adapt to real-time changes; third, wavelet decomposition requires batch calculation after all data collection is completed, which is not suitable for real-time updates;

[0058] Specifically, the set of smooth trajectory points D smooth It is expressed as:

[0059]

[0060] Where n represents the total number of smooth trajectory points + 1;

[0061] It should be noted that the corrected trajectory points are closer to the real trajectory, which can prevent the subsequent generator from learning the wrong robot arm operation state and improve the training accuracy of the generative network adversarial model (GAN);

[0062] Furthermore, for the smooth trajectory point set D smooth Normalize each smooth trajectory point in to get:

[0063]

[0064] in, Represented as a smooth trajectory point set D smooth The mean of Represented as a smooth trajectory point set D smooth The standard deviation of x norm Represented as normalized smooth trajectory points;

[0065] Further, a path length feature vector and a motion speed feature vector are obtained from the calculated path length and average motion speed of the robot arm when the operation is performed in the trajectory data D;

[0066] Specifically, the path length feature vector L is expressed as:

[0067]

[0068] Where N represents the number of samples of trajectory data;

[0069] Specifically, the motion velocity feature vector v avg It is expressed as:

[0070]

[0071] Furthermore, the motion velocity feature vector v avg Combined with the path length feature vector L, we get the task condition feature vector Φ task =[L,v avg ];

[0072] Furthermore, the normalized smooth trajectory point set D smooth and the task condition feature vector Φ task =[L,v avg ] as the input set of GAN;

[0073] S2. Create a generative adversarial network model, take the feature vector of the trajectory data and the normalized smooth trajectory point set as the input set, input them into the generative adversarial network model, and process the input set by the generator and discriminator in the generative adversarial network model, design an alternating optimization mechanism for the generator and the discriminator, train the generative adversarial network model, and obtain trajectory characteristic parameters;

[0074] Furthermore, random noise is introduced into the generator The task condition feature vector Φ task =[L,v avg ] and random noise As input to the generator, we get:

[0075]

[0076] Furthermore, the path trajectory is generated by a deep neural network and a smooth trajectory point sequence is gradually output. The process is as follows:

[0077] S201, initializing the task condition feature vector and random noise to generate the starting point T0 of the smooth trajectory point;

[0078] S202, based on the smooth trajectory points and random noise at the previous moment Recursively get the current smooth trajectory point:

[0079]

[0080] Among them, T i+1 is the smooth trajectory point at the previous moment, T i is the smooth trajectory point at the current moment, and G is the generator;

[0081] S203, output smooth trajectory point sequence T gen ={T0, T1, ..., T n};

[0082] Specifically, each deep neural network includes two hidden layers: hidden layer 1, which fuses the task condition feature vector and random noise to generate the trajectory starting point T0; hidden layer 2, which predicts subsequent smooth trajectory points based on the starting point of the smooth trajectory point;

[0083] Furthermore, the obtained smooth trajectory point sequence is input into the discriminator, and the discriminator judges the obtained smooth trajectory point sequence by collecting the real smooth trajectory point sequence to obtain the loss value of the generator, and judges the authenticity of the input smooth trajectory point sequence according to the loss value of the generator;

[0084] Specifically, the generator loss value includes the adversarial loss value and the trajectory feature error value;

[0085] It should be explained that the adversarial loss value is the core goal of the game between the generator and the discriminator, which is used to measure whether the generated smooth trajectory point sequence is realistic enough to "fool" the discriminator; the goal of the generator is to minimize the adversarial loss value, while the goal of the discriminator is to maximize the adversarial loss value;

[0086] Specifically, if the adversarial loss value is large, it means that the discriminator can easily distinguish the generated smooth trajectory point sequence from the real smooth trajectory point sequence, indicating that the smooth trajectory point sequence generated by the generator is not realistic enough; if the adversarial loss value is small, it means that the generated smooth trajectory point sequence is close to the real smooth trajectory point sequence, and the performance of the generator gradually reaches the optimal level;

[0087] It should be explained that the trajectory feature error is the deviation of the generated smooth trajectory point sequence or the real smooth trajectory point sequence in certain specific features, which refers to the deviation of the path length feature vector L and the motion speed feature vector v avg Because in the trajectory generation of the robot arm, it is not enough to just generate a realistic trajectory, but also needs to meet the specific requirements of the task, such as path length, speed and other targets;

[0088] Furthermore, we design an alternating optimization mechanism for the generator and the discriminator, fix the generator, optimize the discriminator, and input the real smooth trajectory point sequence T real And the obtained smooth trajectory point sequence T gen , where the output of the discriminator R(T) is the true probability of the two input trajectories:

[0089] R(T)∈[0,1]

[0090] Then we get:

[0091] R(T real )≈1

[0092] R(T gen )≈0

[0093] It should be noted that since the discriminator aims to maximize the classification accuracy of the two trajectories, the above R(T real )≈1 and R(T gen )≈0, then the discriminator loss value for:

[0094]

[0095] in, Expressed as expected value;

[0096] Furthermore, through the discriminator loss value Back-propagation optimizes the generator G;

[0097] Furthermore, the discriminator is fixed, the generator is optimized, and the smooth trajectory point sequence T is input gen , since the goal of the generator is to minimize the adversarial loss value, the generator loss value is for:

[0098]

[0099] in, is the path length error, is the motion speed error;

[0100] Specifically, the path length error and motion speed error The formula is:

[0101]

[0102] Among them, L gen is the path length feature vector of the obtained smooth trajectory point sequence, L real is the path length feature vector of the true smooth trajectory point sequence; v avg，gen is the motion velocity feature vector of the obtained smooth trajectory point sequence, v avg，real is the motion velocity feature vector of the real smooth trajectory point sequence;

[0103] Furthermore, through the generator loss value Back-propagation optimizes the discriminator R(T);

[0104] Furthermore, the process of optimizing the generator G and the discriminator R(T) is iterated until the smooth trajectory point sequence reaches the optimal state, and the trajectory feature parameters are output;

[0105] Specifically, the iteration termination condition is that the optimization results in the discriminator and the generator no longer change, or the real smooth trajectory point sequence T real And the obtained smooth trajectory point sequence T gen All the trajectory points in have been processed;

[0106] S3. Filter the optimal trajectory of the robot arm when performing the operation according to the trajectory characteristic parameters, so as to improve the efficiency of the robot arm when performing the operation;

[0107] Further, the screening method is as follows:

[0108] Set the score of each trajectory feature parameter, perform permutations and combinations according to each trajectory feature parameter, obtain the path trajectory after permutations and combinations, calculate the score of the path trajectory, sort the path trajectory in descending order according to the path trajectory score, and select the highest score as the optimal trajectory for the robot arm operation execution;

[0109] Specifically, the scoring method for calculating the path trajectory can be obtained by the existing scoring formula, which will not be described in detail here;

[0110] Specifically, if the path trajectory scores are the same, any one of them is selected as the highest score;

[0111] It should be noted that by calculating the highest value of the path trajectory score, the selected optimal trajectory can have a lower path length and smoother acceleration changes, thereby improving the efficiency of the robot arm when performing tasks.

[0112] Example 2

[0113] Reference Figure 3, which is the second embodiment of the present invention, and this embodiment provides a method for optimizing the operation trajectory of a robot arm based on a generative adversarial network, including: to verify the advantages of the solution of the present invention in optimizing the trajectory of the robot arm, a comparative experiment is conducted with the prior art; the test objects include a robot arm trajectory generation solution optimized based on a generative adversarial network (GAN) (the solution of the present invention) and a traditional trajectory optimization method based on a kinematic model (the prior art);

[0114] Experimental environment configuration: Robot arm model: UR10e six-axis robot arm, end load 3kg, working radius 1300mm;

[0115] The hardware configuration is as follows:

[0116] Controller: NVIDIA Jetson AGX Xavier 32GB; Data acquisition system: Vicon system based on optical motion capture, sampling frequency 100Hz; Sensor: six-axis force / torque sensor (ATI F / TMini45);

[0117] The software environment configuration is as follows:

[0118] Deep learning framework: PyTorch 1.13; operating system: Ubuntu 20.04; GAN model training adopts a dual network architecture of generator and discriminator, with a generator depth of 5 layers, a discriminator depth of 4 layers, and 2000 training rounds;

[0119] The experimental task is to simulate the robot arm to complete the welding operation in a limited space. The content is as follows:

[0120] Task path: The path contains multiple consecutive welding points (50 points in total), and the trajectory is required to smoothly transition between welding points, and the total path length shall not exceed 1.5 meters; Task time: The task completion time is set to no more than 15 seconds; Performance evaluation indicators: including path length, trajectory smoothness, energy consumption and task completion time;

[0121] The experimental process is as follows: First, the task environment is scanned and imported into the robot control system, and based on the existing technology (kinematic model) and the solution of the present invention (GAN generated trajectory optimization), the task trajectories are generated respectively, and then the robot executes the two sets of generated trajectories in turn, and the end force / torque sensor records the task energy consumption in real time. In addition, the path points and smoothness indicators of the trajectory are recorded through the motion capture system, and the path length, task energy consumption, trajectory smoothness (acceleration change standard deviation) and completion time are compared. Finally, the performance differences of the two methods on key performance indicators are calculated, and the data obtained are shown in Table 1;

[0122] Table 1

[0123]

[0124]

[0125] First, it can be seen from Table 1 that in terms of path length optimization, the optimal trajectory path length generated by the present invention is 1.48 meters, while the path length generated by the prior art is 1.62 meters, which is shortened by 0.14 meters (optimized by 8.64%). This shows that the scheme of the present invention can plan the trajectory more efficiently, avoid unnecessary redundant motion, and effectively improve the execution efficiency of the robot arm task; in terms of trajectory smoothness, the trajectory smoothness index (acceleration standard deviation) generated by the scheme of the present invention is 0.018, which is 48.57% better than 0.035 in the prior art; and trajectory smoothness directly affects the stability of the robot arm movement and the quality of task completion. The lower acceleration change shows that the present invention The transition between trajectory points generated by the scheme is more natural, avoiding motion jitter caused by sudden changes in the trajectory; in terms of reducing task energy consumption, the energy consumption of the scheme of the present invention is 10.1 joules, while the energy consumption of the prior art is 12.6 joules, which is a 19.84% reduction in energy consumption; the reduction in energy consumption reflects that the scheme of the present invention better balances the path length and speed distribution in trajectory planning, thereby reducing the energy waste of the robot arm during movement; in terms of shortening the completion time, the task completion time of the scheme of the present invention is 13.6 seconds, which is slightly faster than the 14.2 seconds of the prior art, and is shortened by 4.23%; although the time reduction is small, its comprehensive performance still shows that the scheme of the present invention can complete the task more efficiently;

[0126] Secondly, through Figure 3 It can be seen that the trajectory of the scheme of the present invention is closer to the ideal trajectory, and the transition between the trajectory points is more natural, the jitter is relatively reduced, and in addition, the standard deviation of the acceleration change is also significantly reduced, making the path trajectory of the robot arm more stable during operation.

[0127] Finally, compared with the traditional trajectory optimization method based on kinematic models, the solution of the present invention demonstrates higher adaptability and practical value, especially in multi-objective optimization and complex task scenarios, and has significant technical advantages and application potential.

[0128] Those skilled in the art will appreciate that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program codes. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0129] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0130] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0131] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0132] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0133] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for optimizing the operation trajectory of a robotic arm based on a generative adversarial network, characterized in that: include: Collecting trajectory data of the robot arm during operation execution, predicting the current moment of the trajectory point in the trajectory data by using the Kalman filter method and making corrections to obtain a set of smooth trajectory points, normalizing the set of smooth trajectory points, and considering the eigenvector of the trajectory data; Create a generative adversarial network model, take the feature vector of the trajectory data and the normalized smooth trajectory point set as an input set, input them into the generative adversarial network model, process the input set by the generator and the discriminator in the generative adversarial network model, design an alternating optimization mechanism for the generator and the discriminator, train the generative adversarial network model, and obtain trajectory characteristic parameters; According to the trajectory characteristic parameters, the optimal trajectory of the robot arm when performing operation is selected, so as to improve the efficiency of the robot arm when performing operation.

2. The method for optimizing the robot arm operation trajectory based on a generative adversarial network according to claim 1, characterized in that: The Kalman filter method is used to predict the current moment of the trajectory point in the trajectory data and make corrections to obtain a set of smooth trajectory points, including: The Kalman filter method is used to predict the current time of the trajectory point in the trajectory data, and at the same time, the observation value of the trajectory data is collected to correct the predicted value obtained at the current time of the trajectory point in the predicted trajectory data; The predicted values ​​are taken as a set of smooth trajectory points.

3. The method for optimizing the robot arm operation trajectory based on a generative adversarial network according to claim 1, characterized in that: The smooth trajectory point set is normalized while considering the feature vector of the trajectory data, including: Each smooth trajectory point in the smooth trajectory point set is normalized, and the path length and average movement speed of the robot arm during operation execution are calculated from the collected trajectory data to obtain a path length feature vector and a movement speed feature vector.

4. The method for optimizing the robot arm operation trajectory based on a generative adversarial network according to claim 2 or 3, characterized in that: Create a generative adversarial network model, taking the feature vector of the trajectory data and the normalized smooth trajectory point set as the input set, including: Combining the motion speed feature vector and the path length feature vector to obtain a task condition feature vector; The normalized smooth trajectory point set and the task condition feature vector are used as the input set of the generative adversarial network model.

5. The method for optimizing the robot arm operation trajectory based on a generative adversarial network according to claim 4, characterized in that: The input set is processed by the generator and the discriminator in the generative adversarial network model, and an alternating optimization mechanism of the generator and the discriminator is designed, including: Introducing random noise into the generator, taking the task condition feature vector and the random noise as inputs of the generator, generating a path trajectory through a deep neural network and gradually outputting a smooth trajectory point sequence; The obtained smooth trajectory point sequence is input into the discriminator. The discriminator judges the obtained smooth trajectory point sequence by collecting the real smooth trajectory point sequence to obtain the loss value of the generator. According to the loss value of the generator, the authenticity of the input smooth trajectory point sequence is judged.

6. The method for optimizing the robot arm operation trajectory based on a generative adversarial network according to claim 5, characterized in that: An alternating optimization mechanism for the generator and the discriminator is designed to train the generative adversarial network model to obtain trajectory characteristic parameters, including: The generator is fixed, the discriminator is optimized, the real smooth trajectory point sequence and the obtained smooth trajectory point sequence are input, the discriminator loss value is obtained, and the generator is optimized by back-propagating the discriminator loss value; The discriminator is fixed, the generator is optimized, the smooth trajectory point sequence is input, the generator loss value is obtained, and the discriminator is optimized by backpropagation of the generator loss value; The process of optimizing the generator and the discriminator is iterated until the obtained smooth trajectory point sequence reaches the optimal state, and the trajectory feature parameters are output.

7. The method for optimizing the robot arm operation trajectory based on a generative adversarial network according to claim 6, characterized in that: According to the trajectory characteristic parameters, the optimal trajectory when the robot arm is operating is selected, including: A score is set for each trajectory feature parameter, and each trajectory feature parameter is arranged and combined to obtain a path trajectory after the arrangement and combination, and a score of the path trajectory is calculated. The path trajectory scores are arranged in descending order, and the highest score is selected as the optimal trajectory when the robot arm is operated.

Images