Cabin testing method, device and equipment based on manipulator

Abstract

The invention provides a cockpit testing method, device and equipment based on a manipulator, and relates to the technical field of vehicle testing. The method comprises the following steps: receiving a cabin test execution instruction for the manipulator; comparing the execution instruction with data in a database; and under the condition that data matched with the execution instruction exists in a database, taking the execution instruction as input of a target model, and obtaining a test report of the cockpit test based on the manipulator. According to the scheme, the mechanical arm does not need to be reprogrammed and debugged according to the execution instruction, format conversion does not need to be conducted on the execution instruction of a natural language, and the test report corresponding to the execution instruction can be obtained by directly inputting the execution instruction into the target model for simulation. According to the scheme, the problems that the manipulator is complex and diverse in action and low in reuse degree, a large amount of time needs to be spent on reprogramming and debugging for each new test task, and the test efficiency is low are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle testing, and in particular to a cockpit testing method, device and equipment based on a manipulator. Background Art

[0002] Cockpit testing is an important part of ensuring vehicle quality and user experience. The robot has the characteristics of high accuracy, high stability and no environmental influence. It can perform long-term repetitive operations in various environments to ensure the consistency and accuracy of test results.

[0003] However, with the increase in vehicle models and the complexity of cockpit functions, the preparation work before using the manipulator for testing has become increasingly arduous and complicated. First of all, the manipulator's motion debugging and trajectory recording are complicated: the cockpit layout and functions of each vehicle model are different, and the manipulator needs to perform a lot of basic motion debugging and record the corresponding motion trajectory. Because the manipulator's movements are complex and diverse, and the degree of reusability is not high, each new test task requires a lot of time to reprogram and debug, resulting in low test efficiency. Summary of the invention

[0004] The embodiments of the present invention provide a cockpit testing method, device and equipment based on a manipulator, so as to solve the problems in the prior art that the manipulator's movements are complex and diverse, and the reuse rate is not high. Each new test task requires a lot of time to reprogram and debug, resulting in low test efficiency.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention provides a cockpit testing method based on a manipulator, comprising:

[0007] receiving a cockpit test execution instruction for the manipulator, wherein the execution instruction is natural language data;

[0008] comparing the execution instruction with the data in the database;

[0009] In the case that there is data matching the execution instruction in the database, the execution instruction is used as an input of the target model to obtain a test report of the cockpit test based on the manipulator;

[0010] The database includes historical execution instructions for performing model training on the target model and historical test reports corresponding to the historical execution of the instructions.

[0011] Furthermore, before comparing the execution instruction with the data in the database, the following steps are performed:

[0012] Performing data cleaning and standardization processing on the execution instructions;

[0013] The standardization process includes standardization of physical parameter units and standardization of coordinate systems.

[0014] Furthermore, the method further comprises:

[0015] Acquire multiple groups of test instructions for testing the manipulator, and test reports corresponding to the test instructions, and save the test instructions to the database;

[0016] According to the test instruction and the test report, the prediction model is trained to obtain the target model.

[0017] Further, according to the test instruction and the test report, the prediction model is trained, including:

[0018] Extracting features from the test instruction to obtain feature parameters; the test instruction is natural language data;

[0019] Performing model training on the prediction model according to the test instruction, the characteristic parameters and the test report;

[0020] The characteristic parameters include at least one of the following: the starting position, target position, speed, acceleration, angles between joints of the manipulator, relative positions between joints, speed change rate, and acceleration change rate of the end effector of the manipulator.

[0021] Furthermore, according to the test instruction, the characteristic parameter and the test report, before the prediction model is trained, at least one of the following is included:

[0022] Performing time alignment on the characteristic parameters;

[0023] Using a sliding window or statistical analysis to remove outliers and noise data in the characteristic parameters;

[0024] Using the difference method to supplement the missing data in the characteristic parameters;

[0025] The characteristic parameters are standardized or normalized.

[0026] Furthermore, before the prediction model is trained according to the test instruction, the characteristic parameter and the test report, the method further includes:

[0027] Using a sliding window or time series segmentation method to convert the format of the characteristic parameters; and / or,

[0028] The internal principal component analysis method is used to perform dimension reduction processing on the characteristic parameters after the conversion format.

[0029] Furthermore, the method further comprises:

[0030] Verifying the model correctness of the target model;

[0031] Optimizing the hyperparameters of the target model that has passed the verification by using a grid search method, a random search method, or a Bayesian optimization method;

[0032] The hyperparameters include: learning rate, number of hidden layers, and number of nodes.

[0033] Furthermore, the prediction model is structured as a long short-term memory network or a recurrent neural network;

[0034] According to the test instruction, the characteristic parameter and the test report, the prediction model is trained, including:

[0035] The prediction model is trained using a back propagation algorithm and a gradient descent optimization algorithm.

[0036] In a second aspect, an embodiment of the present invention provides a cockpit testing device based on a manipulator, comprising:

[0037] A receiving module, used for receiving an execution instruction for the robot, wherein the execution instruction is natural language data;

[0038] A comparison module, used for comparing the execution instruction with the data in the database;

[0039] A determination module, configured to, if data matching the execution instruction exists in a database, use the execution instruction as an input of a target model to obtain a test report of the cockpit test based on the manipulator;

[0040] The database includes historical execution instructions for performing model training on the target model and historical test reports corresponding to the historical execution of the instructions.

[0041] In a third aspect, an embodiment of the present invention provides a cockpit testing device based on a manipulator, comprising: a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the cockpit testing method based on the manipulator as described above are implemented.

[0042] The beneficial effects of the present invention are:

[0043] The cockpit test method based on the manipulator of the embodiment of the present invention compares the received cockpit test execution instruction for the manipulator with the data in the database. When the execution instruction matches the data in the database, there is no need to reprogram and debug the manipulator according to the execution instruction, nor is there any need to convert the format of the execution instruction in natural language. The execution instruction can be directly input into the target model simulation to obtain the test report corresponding to the execution instruction. The solution of the present invention solves the problem that the manipulator's movements are complex and diverse, and the reuse rate is not high. Each new test task requires a lot of time to reprogram and debug, resulting in low test efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A schematic diagram showing the steps of a cockpit testing method based on a manipulator according to an embodiment of the present invention;

[0045] Figure 2 A schematic diagram showing a module of a cockpit testing device based on a manipulator according to an embodiment of the present invention;

[0046] Figure 3 A schematic diagram showing the structure of a cockpit testing device based on a manipulator according to an embodiment of the present invention;

[0047] Figure 4 A schematic structural diagram of a cockpit testing device based on a manipulator according to another embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail in conjunction with the accompanying drawings and specific embodiments. In the following description, specific details such as specific configurations and components are provided only to help fully understand the embodiments of the present invention. Therefore, it should be clear to those skilled in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention. In addition, for clarity and brevity, the description of known functions and structures is omitted.

[0049] It should be understood that the references to "one embodiment" or "an embodiment" throughout the specification mean that the specific features, structures, or characteristics associated with the embodiment are included in at least one embodiment of the present invention. Therefore, the references to "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0050] The present invention aims to solve the problems in the prior art that the movements of the manipulator are complex and diverse, and the reusability is not high. Each new test task requires a lot of time to reprogram and debug, resulting in low test efficiency. A cockpit testing method, device and equipment based on a manipulator are provided.

[0051] like Figure 1 As shown, an embodiment of the present invention provides a cockpit testing method based on a manipulator, comprising the following steps:

[0052] Step 101, receiving a cockpit test execution instruction for a manipulator, wherein the execution instruction is natural language data;

[0053] Step 102, comparing the execution instruction with the data in the database;

[0054] Step 103, when there is data matching the execution instruction in the database, the execution instruction is used as an input of the target model to obtain a test report of the cockpit test based on the manipulator;

[0055] The database includes historical execution instructions for performing model training on the target model and historical test reports corresponding to the historical execution of the instructions.

[0056] Optionally, the execution instruction may be data in EXCEL format;

[0057] The execution instructions include: trajectory instructions, action instructions, speed instructions and acceleration instructions of the robot.

[0058] The cockpit test of the manipulator is simulated by the target model, thereby avoiding the need to program and debug the manipulator for each test task.

[0059] The cockpit test method based on the manipulator of the embodiment of the present invention compares the received cockpit test execution instruction for the manipulator with the data in the database. When the execution instruction matches the data in the database, there is no need to reprogram and debug the manipulator according to the execution instruction, nor is there any need to convert the format of the execution instruction in natural language. The execution instruction can be directly input into the target model simulation to obtain the test report corresponding to the execution instruction. The solution of the present invention solves the problem that the manipulator's movements are complex and diverse, and the reuse rate is not high. Each new test task requires a lot of time to reprogram and debug, resulting in low test efficiency.

[0060] Optionally, before comparing the execution instruction with the data in the database, the method includes:

[0061] Performing data cleaning and standardization processing on the execution instructions;

[0062] The standardization process includes standardization of physical parameter units and standardization of coordinate systems.

[0063] Optionally, performing data cleaning on the execution instruction includes:

[0064] Invalid data, erroneous data, abnormal data and noise data in the execution instruction are detected.

[0065] In the embodiment of the present invention, a sliding window method or a statistical analysis algorithm is used to identify and eliminate the invalid data, erroneous data, abnormal data and noise data.

[0066] The manipulator-based cockpit testing method of the embodiment of the present invention performs data cleaning and standardization processing on the execution instructions, so that the execution instructions can maintain consistency in physical units and coordinate systems with the data in the database, thereby ensuring that the execution instructions are successfully matched with the data in the database; and by eliminating abnormal data, the success rate of successfully matching the execution instructions with the data in the database can be improved.

[0067] Optionally, the method further comprises:

[0068] Acquire multiple groups of test instructions for testing the manipulator, and test reports corresponding to the test instructions, and save the test instructions to the database;

[0069] According to the test instruction and the test report, the prediction model is trained to obtain the target model.

[0070] Optionally, the multiple groups of test instructions include multiple groups of operation trajectory data of the manipulator under different tasks and environments such as temperature, load, speed, etc.

[0071] It should be noted that the test report corresponding to the test instruction is obtained by means of the prior art, that is, the manipulator is programmed and debugged according to the test instruction so that the manipulator is tested, and the test report is obtained after the test is successful.

[0072] Optionally, the prediction model includes: a recursive neural network suitable for processing time series data, and a convolutional neural network suitable for processing spatial series data.

[0073] In an embodiment of the present invention, the recursive neural network may be a long short-term memory network (Long Short-Term Memory, LSTM) or a recurrent neural network (Recurrent Neural Network, RNN), etc.

[0074] Optionally, the prediction model is trained using a back propagation algorithm and a gradient descent optimization algorithm;

[0075] The mean square error is selected as the loss function to measure the target model error.

[0076] Optionally, performing model training on the prediction model according to the test instruction and the test report includes:

[0077] Extracting features from the test instruction to obtain feature parameters; the test instruction is natural language data;

[0078] Performing model training on the prediction model according to the test instruction, the characteristic parameters and the test report;

[0079] The characteristic parameters include at least one of the following: the starting position, target position, speed, acceleration, angles between joints of the manipulator, relative positions between joints, speed change rate, and acceleration change rate of the end effector of the manipulator.

[0080] Optionally, the feature extraction of the test instruction can be understood as collecting and recording the feature parameters corresponding to the test instruction through peripherals such as position sensors, force sensors, accelerometers and gyroscopes, cameras, etc. according to the test instruction.

[0081] It should be noted that the characteristic parameters such as the angles between the joints of the manipulator, the relative positions between the joints, the velocity change rate and the acceleration change rate are generated derivative features.

[0082] The cockpit testing method based on a manipulator in an embodiment of the present invention trains a prediction model through multiple groups of the test instructions, the characteristic parameters and the test report, and inputs the target model according to the execution instructions in natural language. A test report for the cockpit test based on the manipulator can be obtained without converting the execution instructions or reprogramming the manipulator, thereby improving the test efficiency.

[0083] Optionally, before the prediction model is trained according to the test instruction, the characteristic parameter and the test report, at least one of the following is included:

[0084] Performing time alignment on the characteristic parameters;

[0085] Using a sliding window or statistical analysis to remove outliers and noise data in the characteristic parameters;

[0086] Using the difference method to supplement the missing data in the characteristic parameters;

[0087] The characteristic parameters are standardized or normalized.

[0088] Optionally, the missing data in the characteristic parameters are supplemented by using a difference method, including:

[0089] Linear interpolation or spline interpolation is used to supplement the missing data in the characteristic parameters.

[0090] Optionally, the normalization method includes:

[0091] Z-score standardization and Min-Max normalization processing methods.

[0092] In the embodiment of the present invention, the time alignment is used to keep the characteristic parameters consistent in time sequence, thereby avoiding the model with errors in training caused by inconsistent time sequence.

[0093] By removing the outliers and noise data in the characteristic parameters, the problem of difficult convergence of model training caused by outliers can be avoided, thereby improving the accuracy and training efficiency of the target model obtained by training;

[0094] By supplementing the confirmed characteristic parameters through difference, the continuity and integrity of the training data can be ensured, and the accuracy and universality of the target model can be improved;

[0095] By standardizing and normalizing the characteristic data, the influence of different dimensions or coordinate systems of the characteristic parameters can be eliminated.

[0096] Optionally, before performing model training on the prediction model according to the test instruction, the characteristic parameter and the test report, the method further includes:

[0097] Using a sliding window or time series segmentation method to convert the format of the characteristic parameters; and / or,

[0098] The internal principal component analysis method is used to perform dimension reduction processing on the characteristic parameters after the conversion format.

[0099] In the embodiment of the present invention, the characteristic parameters are converted into a format by using a sliding window or a time series segmentation method, so that the characteristic parameters are suitable for input and the prediction model is trained;

[0100] The high-dimensional features in the feature parameters are reduced in dimension by using methods such as principal component analysis, thereby reducing feature redundancy and improving the efficiency of model training.

[0101] Optionally, the method further comprises:

[0102] Verifying the model correctness of the target model;

[0103] Optimizing the hyperparameters of the target model that has passed the verification by using a grid search method, a random search method, or a Bayesian optimization method;

[0104] The hyperparameters include: learning rate, number of hidden layers, and number of nodes.

[0105] In the embodiment of the present invention, K-fold cross validation is used to analyze and verify the performance indicators, errors and other aspects of the target model to reduce the contingency caused by data division;

[0106] Regularization methods are used to prevent model overfitting and improve the generalization ability of the model;

[0107] The amount of training data is increased through data enhancement techniques (such as data noise addition, data expansion, etc.), thereby improving the robustness of the target model.

[0108] It should be noted that the target model is deployed in the cloud server, and the robot's motion trajectory and actions are controlled by calling the interface protocol. The security of the information interaction protocol ensures that the target model can respond and process tasks in real time. The operation of the model is monitored in real time, new data generated during the operation is collected, and the target model is updated and optimized online.

[0109] Specifically, the method of using the target model includes:

[0110] The target model is loaded, and the relevant data of the target model is stored in the Hadoop distributed file system. The Hadoop distributed file system is a reliable framework for storing big data, which can store a large amount of unstructured and structured data. Its core components support loading big data into memory for processing. Spark's spark.read method can read data from a variety of data sources (such as HDFS, S3, JDBC, etc.) and load it into memory.

[0111] Generate a prediction result (the test report), use the target model to simulate the newly input execution instruction, generate the trajectory of the manipulator and the test report; for example, use the prediction function (predict) of the Keras library to generate the prediction result, and decode the predicted action.

[0112] Optionally, the prediction model is structured as a long short-term memory network or a recurrent neural network;

[0113] According to the test instruction, the characteristic parameter and the test report, the prediction model is trained, including:

[0114] The prediction model is trained using a back propagation algorithm and a gradient descent optimization algorithm.

[0115] like Figure 2 As shown, an embodiment of the present invention provides a cockpit testing device 200 based on a manipulator, comprising:

[0116] A receiving module 201 is used to receive an execution instruction for the robot, wherein the execution instruction is natural language data;

[0117] A comparison module 202, used to compare the execution instruction with the data in the database;

[0118] A determination module 203 is used to obtain a test report of the cockpit test based on the manipulator by using the execution instruction as an input of the target model when there is data matching the execution instruction in the database;

[0119] The database includes historical execution instructions for performing model training on the target model and historical test reports corresponding to the historical execution of the instructions.

[0120] The cockpit test device based on the manipulator of the embodiment of the present invention compares the received cockpit test execution instruction for the manipulator with the data in the database. When the execution instruction matches the data in the database, there is no need to reprogram and debug the manipulator according to the execution instruction, nor is there any need to convert the format of the execution instruction in natural language. The execution instruction can be directly input into the target model simulation to obtain the test report corresponding to the execution instruction. The solution of the present invention solves the problem that the manipulator's movements are complex and diverse, and the reuse rate is not high. Each new test task requires a lot of time to reprogram and debug, resulting in low test efficiency.

[0121] like Figure 3 As shown, an embodiment of the present invention further provides a cockpit testing device based on a manipulator, comprising: a transceiver 301, a processor 302, a memory 303, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the cockpit testing method based on the manipulator as described above are implemented.

[0122] In the embodiment of the present invention, Figure 4 As shown, the cockpit test equipment based on the manipulator includes:

[0123] Input and output module, used to input the execution instructions, and output test reports and operation logs, etc.;

[0124] A data preprocessing module, used for performing data processing and format conversion on the input execution instruction, and sending the converted execution instruction to the manipulator control module or the model training module, or directly sending the execution instruction to the model reasoning module;

[0125] A model reasoning module is used to store the target model, perform simulation calculations according to the received execution instructions, obtain the test report, and send the test report to the model training module;

[0126] A model training module, used to perform online training and update the target model according to the execution instruction and the test report, and send the updated target model to the model reasoning module;

[0127] The robot control module is used to generate a control command for the robot arm according to the received execution instruction after the conversion format, and send the control command to the robot to make the robot perform the corresponding action.

[0128] It should be noted that the input and output module uses Python's built-in os and io modules to read the execution instructions into the memory; uses the unittest, pytest and other test frameworks to generate test reports, uses the Jinja2 template engine to customize the report format, uses Python's built-in logging module to record and manage run logs, defines a unified data storage solution, and uses abstract base classes to define interfaces; in order to improve data access speed, memory cache and distributed cache can be used to reduce repeated read operations.

[0129] In addition, a specific embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of the method in the first embodiment described above are implemented. The same technical effect can be achieved, and to avoid repetition, it will not be described here.

[0130] The present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the above Figure 1 The various processes of the method embodiment shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0131] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary personnel in this technical field, several improvements and modifications can be made without departing from the principles described in the present invention. These improvements and modifications are also within the scope of protection of the present invention.

Claims

1. A cockpit testing method based on a manipulator, characterized in that: include: receiving a cockpit test execution instruction for the manipulator, wherein the execution instruction is natural language data; comparing the execution instruction with the data in the database; In the case that there is data matching the execution instruction in the database, the execution instruction is used as an input of the target model to obtain a test report of the cockpit test based on the manipulator; The database includes historical execution instructions for performing model training on the target model and historical test reports corresponding to the historical execution of the instructions.

2. The cockpit testing method based on a manipulator according to claim 1, characterized in that: Before comparing the execution instruction with the data in the database, the following steps are included: Performing data cleaning and standardization processing on the execution instructions; The standardization process includes standardization of physical parameter units and standardization of coordinate systems.

3. The cockpit testing method based on a manipulator according to claim 1, characterized in that: The method further comprises: Acquire multiple groups of test instructions for testing the manipulator, and test reports corresponding to the test instructions, and save the test instructions to the database; According to the test instruction and the test report, the prediction model is trained to obtain the target model.

4. The cockpit testing method based on a manipulator according to claim 3 is characterized in that: According to the test instruction and the test report, the prediction model is trained, including: Extracting features from the test instruction to obtain feature parameters; the test instruction is natural language data; Performing model training on the prediction model according to the test instruction, the characteristic parameters and the test report; The characteristic parameters include at least one of the following: the starting position, target position, speed, acceleration, angles between joints of the manipulator, relative positions between joints, speed change rate, and acceleration change rate of the end effector of the manipulator.

5. The cockpit testing method based on a manipulator according to claim 4, characterized in that: Before performing model training on the prediction model according to the test instruction, the characteristic parameter and the test report, at least one of the following is included: Performing time alignment on the characteristic parameters; Using a sliding window or statistical analysis to remove outliers and noise data in the characteristic parameters; Using the difference method to supplement the missing data in the characteristic parameters; The characteristic parameters are standardized or normalized.

6. The cockpit testing method based on a manipulator according to claim 4, characterized in that: Before performing model training on the prediction model according to the test instruction, the characteristic parameter and the test report, the method further includes: Using a sliding window or time series segmentation method to convert the format of the characteristic parameters; and / or, The internal principal component analysis method is used to perform dimension reduction processing on the characteristic parameters after the conversion format.

7. The cockpit testing method based on a manipulator according to claim 3, characterized in that: The method further comprises: Verifying the model correctness of the target model; Optimizing the hyperparameters of the target model that has passed the verification by using a grid search method, a random search method, or a Bayesian optimization method; The hyperparameters include: learning rate, number of hidden layers, and number of nodes.

8. The cockpit testing method based on a manipulator according to claim 4, characterized in that: The prediction model is structured as a long short-term memory network or a recurrent neural network; According to the test instruction, the characteristic parameter and the test report, the prediction model is trained, including: The prediction model is trained using a back propagation algorithm and a gradient descent optimization algorithm.

9. A cockpit testing device based on a manipulator, characterized in that: include: A receiving module, used for receiving an execution instruction for the robot, wherein the execution instruction is natural language data; A comparison module, used for comparing the execution instruction with the data in the database; A determination module, configured to, if data matching the execution instruction exists in a database, use the execution instruction as an input of a target model to obtain a test report of the cockpit test based on the manipulator; The database includes historical execution instructions for performing model training on the target model and historical test reports corresponding to the historical execution of the instructions.

10. A cockpit test device based on a manipulator, comprising: A transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the manipulator-based cockpit testing method as described in any one of claims 1 to 8 when executing the computer program.

Images