A method, device and robot for autonomous learning of a robot

Abstract

The present application discloses a method for robot adaptive learning, comprising: receiving scene information collected by a sensing device, substituting the information into a current classification model, obtaining a user identity and a user behavior recognition result, and judging whether to interact with the current user; if the judgment result of whether to interact with the current user is yes, selecting and executing an interaction method according to a preset correspondence between the recognition result and the interaction method; judging whether the interaction method is appropriate according to the feedback information of the current user on the interaction method collected by the sensing device; if the judgment result of whether the interaction method is appropriate is yes, using the user information collected by the sensing device and the user identity and the user behavior recognition result as a positive sample; if the judgment result of whether the interaction method is appropriate is no, using the user information collected by the sensing device and the user identity and the user behavior recognition result as a negative sample; and adding the positive sample or the negative sample to a training sample set for training the current classification model.

Description

Technical Field

[0001] The present application relates to the field of intelligent robot technology, and more particularly to a method for autonomous learning of a robot, a device for autonomous learning of a robot, and a robot. Background Art

[0002] With the development of artificial intelligence, intelligent robots are now being applied in various fields. Existing robots often interact with users based on a pre-designed model framework. To make intelligent robots more user-friendly during use, updating the existing classification models is essential. To update the classification models, those skilled in the art often need to compile large amounts of user usage and scenario data, classify and filter this data, and then train it using a deep learning framework. Summary of the Invention

[0003] The present application provides a method for robot adaptive learning, a device for robot adaptive learning, and a robot.

[0004] This application provides a method for robot adaptive learning, comprising:

[0005] Receive scene information collected by the sensing device, substitute it into the current classification model, and obtain user identity and user behavior recognition results;

[0006] Determine whether to interact with the current user based on the user identity and user behavior recognition results;

[0007] If the result of the above determination of whether interaction with the current user occurs is yes, an interaction method is selected and executed according to the preset correspondence between the recognition result and the interaction method;

[0008] Determining whether the interaction mode is appropriate based on feedback information of the current user regarding the interaction mode collected by the sensing device;

[0009] If the result of the above determination of whether the interaction mode is appropriate is yes, the user information, user identity and user behavior recognition results collected by the sensing device are used as positive samples;

[0010] If the result of the above determination of whether the interaction mode is appropriate is no, the user information, user identity and user behavior recognition results collected by the sensing device are used as negative samples;

[0011] The positive samples or negative samples are added to a training sample set for training the current classification model.

[0012] Preferably, the sensing device includes: an image sensor, a sound sensor, and a touch sensor.

[0013] Preferably, the current classification model is pre-trained based on user and user behavior recognition results.

[0014] Preferably, judging whether interaction with the current user occurs based on the user identity and user behavior recognition results includes:

[0015] Determine whether the current user's confidence level is greater than a preset threshold based on the current user's voice information, facial image, distance between the current user and the robot, and the current user's body movements collected by the sensing device;

[0016] When the confidence level of the current user is greater than a preset threshold, interaction occurs with the current user.

[0017] Preferably, the interaction method includes at least one of the following methods: voice interaction, user graphical interface interaction, and body interaction.

[0018] Preferably, judging whether the interaction method is appropriate based on the feedback information of the current user regarding the interaction method collected by the sensing device includes: judging whether the feedback information of the current user regarding the interaction method is the same as the expected feedback result, thereby judging whether the interaction method is appropriate.

[0019] Preferably, the determining whether the feedback information of the current user on the interaction mode is the same as the expected feedback result, thereby determining whether the interaction mode is appropriate, includes:

[0020] Determining whether the current user has made a positive response to the voice interaction information, thereby determining whether the interaction method is appropriate;

[0021] If the current user responds positively to the voice interaction information, the interaction method is considered appropriate;

[0022] If the current user does not respond to the voice interaction information or does not make a positive response, it is considered that the interaction method is inappropriate.

[0023] Preferably, the determining whether the feedback information of the current user on the interaction mode is the same as the expected feedback result, thereby determining whether the interaction mode is appropriate, includes:

[0024] Determining whether the current user operates the currently displayed user graphical interface according to the voice prompt information or text prompt information, thereby determining whether the interaction method is appropriate;

[0025] If the current user operates the currently displayed user graphical interface according to the voice prompt information or text prompt information, it is considered that the interaction mode is appropriate;

[0026] If the current user does not operate the current user graphical interface or exits the current user graphical interface, it is considered that the interaction mode is inappropriate.

[0027] Preferably, the feedback information of the current user regarding the interactive mode collected by the sensing device includes: sound information, operation information of a user graphical interface, location information of the current user, and facial information of the current user.

[0028] Preferably, adding the positive sample or negative sample to a training sample set for training the current classification model includes:

[0029] When the number of the positive samples or negative samples reaches a preset number, the preset number of positive samples or negative samples are added to the training sample set, the current classification model is trained, and the current classification model is updated.

[0030] Preferably, adding the positive sample or negative sample to a training sample set for training the current classification model includes:

[0031] When the recognition result is a positive sample or a negative sample, the positive sample or the negative sample is added to the training sample set, the current classification model is trained, and the current classification model is updated in real time.

[0032] In addition, the present application also provides a robot adaptive learning device, comprising:

[0033] Operation behavior collection unit, used to collect scene information, obtain user identity and user behavior recognition results, and collect feedback information from the current user;

[0034] An interactive behavior judgment unit, configured to judge whether an interaction with the current user occurs based on the user identity and user behavior recognition result;

[0035] an operation behavior judgment unit, configured to judge whether the interaction mode is appropriate based on the feedback information of the current user on the interaction mode;

[0036] a data information marking unit, configured to use the user information, user identity, and user behavior recognition results as positive samples or negative samples based on the judgment result of the operation behavior judgment unit;

[0037] The classification model training unit is used to train the current classification model according to the positive samples or negative samples.

[0038] In addition, the present application also relates to a robot, wherein the robot adopts a robot self-learning device, and the self-adaptive learning device comprises:

[0039] The operation behavior collection unit is used to collect scene information, obtain user identity and user behavior recognition results based on the current classification model, and collect feedback information from the current user.

[0040] The interactive behavior judgment unit is used to judge whether interaction occurs with the current user based on the user identity and user behavior recognition results.

[0041] The interactive mode selection unit is used to select and execute an interactive mode according to a preset correspondence between the recognition result and the interactive mode.

[0042] The operation behavior judgment unit is used to judge whether the interaction mode is appropriate based on the feedback information of the current user on the interaction mode collected by the sensing device.

[0043] a data information marking unit, configured to use the user information, user identity, and user behavior recognition results as positive samples or negative samples based on the judgment result of the operation behavior judgment unit;

[0044] The classification model training unit is used to train the current classification model according to the positive samples or negative samples.

[0045] Compared with the prior art, this application has the following advantages:

[0046] This application presets the current classification model and uses the robot's own sensing device to collect data from multiple aspects, obtain the user identity and user behavior recognition results, and interact with the current user; while interacting, the sensing device will continue to collect the current user's response, thereby knowing whether the user identity and user behavior recognition results obtained by the robot are correct. Through the above method, a set of data consisting of three aspects of data, including the data obtained during an interaction, the recognition results, and whether the recognition results are accurate, is obtained. This set of data can be used as accumulated data for classification model training. In the method provided by this application, each recognition is defined autonomously by the robot, that is, by judging whether the interaction method is appropriate, it is deduced whether the judgment results of user information and user identity are correct, thereby dividing this set of data into positive samples and negative samples. By continuously applying the classification model, sample data can be continuously accumulated, providing a data basis for the continuous improvement of the classification model. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flowchart of a robot adaptive learning method provided by the first embodiment of the present application;

[0048] Figure 2 This is a structural diagram of a robot adaptive learning device provided in the second embodiment of the present application. DETAILED DESCRIPTION

[0049] The following description sets forth many specific details to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.

[0050] The first embodiment of the present application provides a method for robot adaptive learning, which provides a basis for subsequent robot adaptive learning devices and robots. It should be noted that in the following embodiments, the method is described in detail using public service robots as an example, but the scope of application of the method is not limited to public service robots, but also includes self-moving vending robots, cleaning robots, etc. Please refer to Figure 1 , which is a flow chart of the first embodiment of the present application.

[0051] The robot adaptive learning method described in the first embodiment of the present application includes:

[0052] S101, receiving scene information collected by the sensing device, substituting it into the current classification model, and obtaining user identity and user behavior recognition results.

[0053] This step uses the current classification model to identify the user's identity and behavior in the current scene, thereby providing a basis for subsequent interaction. The sensing device collects scene information, which means that the robot in this embodiment is in a specific scene and obtains the surrounding scene information in real time through the sensing device set inside it.

[0054] The sensing device refers to a sensitive device that can obtain information about the surrounding environment; different types of sensing devices can be used according to the different types of information that need to be collected, such as cameras and other devices that collect video information, sound sensors that collect audio information, laser sensing devices that collect obstacle distance information, etc.

[0055] The scene information refers to the image of buildings, people, and their activities at the current robot's location. For example, the scene information can be a hospital, a shopping mall, a bank, a restaurant, etc. In addition, the scene information can be manually set or obtained by the robot's own sensor equipment collecting real-time information about the surrounding environment.

[0056] The classification model refers to mapping samples of unknown categories to one of the given categories based on the characteristics of the data. The construction process of the classification model is usually implemented by a machine learning process. In order to improve the accuracy of classification, the construction of the classification model can have various forms. For example, a random forest classifier is added on the basis of a decision tree, that is, a classifier composed of multiple decision trees. When the sample to be classified enters the random forest, each decision tree is actually asked to perform classification, and finally the category selected the most times by all decision trees is selected as the final classification result.

[0057] The scene information collected by the perception device described in this application is the sample to be classified used to input the classification model, and the user identity and user behavior recognition results are the classification results output by the classification model.

[0058] The current classification model is trained by pre-collected data samples to obtain an initial classification model. It should be noted that as time goes by and the classification model is updated, the classification model at any time point can be called the current classification model at the current time point.

[0059] The user's identity can be identified in a variety of ways. For example, the sensing device may obtain a facial image of the current user, and the server may then perform identity analysis on the facial image, determining the age, gender, and other information of the owner of the facial image based on the analysis results. Alternatively, the sensing device may obtain a facial image, and based on the server's facial recognition algorithm and facial database analysis, determine that the user is a member of the mall (restaurant). The server may then retrieve all the user's membership application information and commonly used service items from the database.

[0060] The user behavior recognition result refers to the recognition result of the user's current behavior type, calculated based on the current classification model, using the user behavior information collected by the sensing device. For example, if the laser sensing device that collects obstacle distance information detects that the current user is more than 10 meters away from the robot, the robot's recognition result based on the current classification model for the current user's behavior is that the current user's behavior does not require interaction. For another example, if the robot's built-in camera captures the current user standing in front of the robot, the robot's recognition result based on the current classification model for the current user's behavior is that the current user's behavior requires interaction.

[0061] This step provides the robot with scene information in the current environment, as well as the user's facial information, voice information, touch information of the robot, etc. through the perception device. In order to achieve this purpose, the perception device in the first embodiment of the present application should include at least one of an image sensor, a sound sensor, and a touch sensor. Of course, the perception device is not limited to the above-mentioned sensors, and other types of perception devices can also be used; the specific types of perception devices used can include multiple types. The original intention of this step is to enable the robot to "perceive" the surrounding environment and "understand" the user's information through the perception device. Therefore, any perception device that can give the robot this function, or other devices that can enable the robot to have this function can be used. These are simple transformations of this method, do not deviate from the core of this application, and are all within the scope of protection of this application.

[0062] S102: Determine whether to interact with the current user based on the user identity and user behavior recognition results.

[0063] This step is the process of deciding whether to provide services to customers by analyzing the user identity and user behavior recognition results in the current scenario, and also provides the possibility for the current user to provide feedback information.

[0064] The interaction refers to the "communication" initiated by the robot with the user.

[0065] Determining whether to interact with the current user refers to determining whether "communication" has occurred with the current user. To better integrate artificial intelligence into the robot adaptive learning method provided in the first embodiment of this application, the robot must determine whether the current user requires the robot's assistance based on the user identity information and the user behavior recognition results. In other words, it must determine whether the user wishes to have a "conversation" with the robot and the topic of the "conversation" they wish to have. In this step, combining the user identity and user behavior recognition results allows for accurate judgment of whether to interact and the topic of the interaction based on both pieces of information. User identity helps effectively screen "conversation" topics. By identifying the current user behavior, the user's current status can be determined, allowing selection of "conversation" topics that the user may be interested in.

[0066] Therefore, this judgment process is generally a process of judging whether the confidence level of the current user's need to interact is greater than a preset threshold based on the current user's voice information, facial image, distance between the current user and the robot, current user's body movements and other information collected by the perception device.

[0067] The confidence level, also known as reliability, confidence level, or confidence coefficient, is used to determine whether the robot should interact with or provide assistance to the current user. Because the information collected by the sensing device is random, the robot's results are uncertain. Therefore, a metric is needed to help the robot determine whether to interact with or provide assistance to the current user.

[0068] The interactive process is necessary and is also an important condition for the implementation of the robot adaptive learning method described in this application.

[0069] S103: If the result of the above determination of whether interaction occurs with the current user is yes, an interaction method is selected and executed according to a preset correspondence between the recognition result and the interaction method.

[0070] The preset recognition result refers to the user identity and user behavior recognition result.

[0071] The correspondence between the preset recognition results and the interaction methods means that each preset recognition result, that is, the user identity and user behavior recognition results, will have one or several selectable corresponding interaction methods. For example, if the robot recognizes user A, and user A has consulted about a certain financial product, the robot will use voice or a combination of voice and user graphical interface to interact with user A. That is, the robot asks user A, "Hello, Mr. / Ms. A, the financial product you have been interested in has a high rate of return recently. Do you need to know more about it?" At the same time, the user graphical interface will jump to the interface related to the financial product.

[0072] Of course, the premise for the above-mentioned interaction method to occur is that the confidence level of the scene information containing user A collected by the robot is greater than a preset threshold.

[0073] The above description shows that there are many possible interaction methods, such as voice interaction, user interface interaction, and physical interaction. However, these interaction methods do not exist independently; in other implementations, more and more complex interaction methods may be used. Furthermore, these interaction methods may be singular or implemented in a multi-faceted manner. These are simple variations of the steps, do not deviate from the core of this application, and are within the scope of protection of this application.

[0074] S104: Determine whether the interaction mode is appropriate based on the feedback information of the current user regarding the interaction mode collected by the sensing device.

[0075] This step uses the sensing device to collect the current user's response to the robot's interaction as feedback information, thereby determining whether the user identity and user behavior recognition results obtained based on the current classification model are correct. This provides the necessary basis for updating the classification model.

[0076] The feedback information refers to the current user's response to the robot's interactive method collected by the perception device, and is information that can directly reflect whether the user identity and user behavior recognition results obtained based on the scene information collected by the perception device are correct.

[0077] Furthermore, the appropriateness standard is relative, so prior to this step, it's necessary to anticipate some possible feedback information for each interaction method. If this anticipated feedback information is positive for the interaction method, then comparing the actual feedback information with the anticipated feedback information will determine whether it is appropriate. It should be noted that the anticipated feedback information can also be negative relative to the interaction method. This is a simple variation of the method, does not deviate from the core of this application, and is within the scope of protection of this application.

[0078] The perception device is working all the time. After interacting with the current user, the perception device will promptly feed back the user's response to the interaction to the robot, and judge whether the feedback information of the current user to the interaction method is the same as the expected feedback result, so as to judge whether the interaction method is appropriate, and then judge whether the user identity and user behavior recognition results obtained based on the current classification model and the scene information collected by the perception device are correct.

[0079] Here, feedback information that is the same as the expected feedback result is called a positive response. In other words, by judging whether the current user has made a positive response to the voice interaction information, it is judged whether the interaction method is appropriate;

[0080] If the current user makes a positive response to the voice interaction information, the interaction method is considered appropriate; if the current user does not respond to the voice interaction information or does not make a positive response, the interaction method is considered inappropriate.

[0081] For example, if the robot sends a voice message asking "Do you need help?" and receives a voice message saying "Yes" or "Need help," the user is deemed to have responded positively to the voice interaction, and the interaction is appropriate. If the robot receives a voice message saying "No, thank you," or receives feedback indicating that the user has left the room, the user is deemed to have not responded positively to the voice interaction, and the interaction is inappropriate.

[0082] For another example, if the robot sends an interactive message such as "Hello Mr. / Ms. A, the financial product you're interested in has a very high return rate recently. Would you like to learn more about it?" and displays the corresponding GUI of the financial product, and the robot receives feedback such as a voice message saying "OK," or if the user performs an action on the GUI of the financial product according to the prompt on the GUI, then the user is determined to have responded positively to the interactive message, and the interaction is appropriate.

[0083] Enabling the robot to determine whether the interactive information it sends is appropriate is the key to the robot adaptive learning method proposed in the first embodiment of this application, but more importantly, how to enable the robot to have the ability to adaptively learn based on these interactive information judged in real time.

[0084] S105-1, if the result of the above-mentioned determination of whether the interaction mode is appropriate is yes, the user information, user identity and user behavior recognition results collected by the sensing device are used as positive samples.

[0085] S105-2: If the result of the above-mentioned determination of whether the interaction mode is appropriate is no, the user information, user identity, and user behavior recognition results collected by the sensing device are used as negative samples.

[0086] If the judgment result of whether the interaction method is appropriate is yes, then based on the current classification model, the user identity and user behavior recognition results obtained according to the scene information collected by the perception device are accurate. At this time, the user identity and user behavior recognition results are marked as positive samples.

[0087] If the result of determining whether the interaction method is appropriate is no, then based on the current classification model, the user identity and user behavior recognition results obtained according to the scene information collected by the perception device are inaccurate. At this time, the user identity and user behavior recognition results are marked as negative samples.

[0088] Each judgment result is annotated here. That is, if the judgment result of whether the interaction method is appropriate is yes, the user information, user identity, and user behavior recognition results are marked as positive samples; if the judgment result of whether the interaction method is appropriate is no, the user information, user identity, and user behavior recognition results are marked as negative samples. In this way, both the scene information collected by the perception device and the individual user information, user identity, and user behavior recognition results will be stored in the robot's memory.

[0089] S106: Add the positive sample or negative sample to a training sample set for training the current classification model.

[0090] The training sample refers to a data set consisting of both samples to be classified and classification results. Each sample has a similar expression as follows: (X1, X2: C). In this application, X1 represents the scene information collected by the perception device, X2 represents the user identity and user behavior recognition results, and C represents the judgment result of whether the user identity and user behavior recognition results obtained based on the scene information collected by the perception device are correct. If the judgment result is correct, let C=1, indicating that the sample is a positive sample. If the judgment result is wrong, let C=0, indicating that the sample is a negative sample.

[0091] The ratio of these erroneous samples to the total number of samples is usually called the "error rate." For example, if there are a negative samples among m samples, the error rate is E = a / m. Correspondingly, 1-a / m is called the "precision," or precision = 1-error rate. The purpose of training the current classification model is to improve its "precision," making the recognition results based on the classification model more accurate.

[0092] The training sample set refers to a data set consisting of positive samples and negative samples used to mine the current classification model framework.

[0093] When various samples are already available, in order to achieve adaptive learning of the robot, it is only necessary to perform data training on the user information, user identity and user behavior recognition results under these samples based on the current classification model, and update the classification model based on the training results.

[0094] The training method can be implemented in a variety of ways. For example, when the number of positive samples or negative samples reaches a preset number, the preset number of positive samples or negative samples are added to the training sample set to train the current classification model and update the current classification model. Alternatively, when the recognition result is a positive sample or a negative sample, the positive sample or negative sample is added to the training sample set to train the current classification model and update the current classification model in real time.

[0095] The two training methods differ, but each has its own advantages. The former reduces the robot's computational workload, but the classification model updates less frequently and takes longer to update. The latter, on the other hand, updates data more frequently and takes less time. Therefore, the training method can be selected based on the actual situation.

[0096] In order to make it easier for readers to understand the robot autonomous learning method described in the first embodiment of the present application, the method is described in detail in combination with specific usage scenarios:

[0097] In scenario 1, a robot with autonomous learning capabilities is placed in a hospital. User A has registered online before. If the robot recognizes user A, it is considered that user A has a high confidence level, and the robot chooses an interactive method to interact with user A.

[0098] Based on the scene information and the information of user A collected by the perception device, the robot asks user A, "Hello, Mr. / Ms. A, do you need to register?" and at the same time jumps the user graphical interface to the service interface for customer registration.

[0099] If the sound perception device collects user A's affirmative voice information such as answering "yes", or the touch perception device collects the user operating the service interface step by step according to the prompt information on the user graphical interface, then the information of user A obtained based on the current classification model is considered correct. At this time, the robot marks the information of user A as a positive sample.

[0100] If the sound perception device detects that user A answers with a negative voice message such as "No, thank you", or if the touch perception device detects that the user has exited the service interface for customer registration, or if user A does not operate the service interface according to the prompts, then the information about user A obtained based on the current classification model is considered to be incorrect. At this time, the robot labels the information about user A as a negative sample.

[0101] When the robot collects enough samples, it will use the sample data set containing all positive and negative samples for data training and update the current classification model.

[0102] In scenario 2, a robot with self-learning capabilities is placed in a bank. User B comes to the bank multiple times to make deposits. If the robot recognizes User B, it considers User B to have a high confidence level, and the robot chooses an interactive method to interact with User B.

[0103] The robot asks user B, "Hello, Mr. / Ms. B, do you need to handle a deposit transaction?" and at the same time redirects the user graphical interface to the deposit service interface.

[0104] If the voice sensing device detects user B's affirmative response, such as "yes," or if the touch sensing device detects the user following the prompts on the graphical user interface to make a deposit, the robot deems user B's information based on the current classification model correct and labels it as a positive example. Each time a positive or negative example is labeled, it is added to the sample dataset for training and updates the current classification model.

[0105] In scenario 3, a robot with autonomous learning capabilities is placed in a shopping mall. If the robot recognizes user C multiple times, or if the robot's voice sensing device picks up information from the user such as "Excuse me, how do I get to where the clothes are sold?", the robot considers user C to be highly trusted, and chooses an interactive method to interact with user C.

[0106] The robot sends a voice message to user C such as "Do you need a map of this mall?" or the user graphical interface automatically jumps to the interface for selecting a service.

[0107] If the sound perception device collects user C's affirmative voice information such as "yes" and opens the mall map according to the prompt information, it is considered that the information of user C obtained based on the current classification model is correct. At this time, the robot marks user C's information as a positive sample.

[0108] If the sound sensing device detects that user C answers with a negative voice message such as "No, thank you," or the touch sensing device detects that the user has exited the service selection interface, or the distance sensing device detects that the user has left, the information about user C obtained based on the current classification model is considered to be incorrect. In this case, the robot labels user C's information as a negative sample.

[0109] When the robot has collected enough samples, it will train the sample dataset containing all positive and negative samples and update the current classification model. Alternatively, each time the robot labels a positive or negative sample, it will add the sample to the sample dataset for training and update the current classification model.

[0110] It should be noted that in the process of training using a sample data set consisting of positive samples and negative samples, the size of the sample data set can be freely set according to actual conditions. Changing the size of the sample data set according to actual conditions is a simple transformation of this method and does not deviate from the core of this application.

[0111] Through the above introduction to the first embodiment of the present application and the use of the adaptive learning method proposed in the first embodiment of the present application by the robot in various usage scenarios, it can be seen that the present application presets the current classification model and uses the perception device set by the robot itself to collect multiple data to obtain the user identity and user behavior recognition results, and interact with the current user; while interacting, the perception device will continue to collect the response made by the current user, thereby knowing whether the user identity and user behavior recognition results obtained by the robot are correct. Through the above method, a set of data consisting of three aspects of data, namely, the data obtained during an interaction, the recognition results, and whether the recognition results are accurate, is obtained. This set of data can be used as accumulated data for classification model training. In the method provided by the present application, each recognition is defined by the robot autonomously, that is, by judging whether the interaction method is appropriate, it is deduced whether the judgment results of the user information and user identity are correct, thereby dividing this set of data into positive samples and negative samples. By continuously applying the classification model, sample data can be continuously accumulated, providing a data basis for the continuous improvement of the classification model.

[0112] In the above embodiment, a method for robot adaptive learning is provided. Correspondingly, the present application also provides a device for robot adaptive learning. Figure 2 , which is a schematic diagram of the structure of a robot adaptive learning device provided in the second embodiment of this application. Since the device embodiment is basically similar to the method embodiment, the description is relatively simple. For relevant details, please refer to the partial description of the system embodiment. The device embodiment described below is only illustrative.

[0113] A robot adaptive learning device 101 of this embodiment includes:

[0114] The operation behavior collection unit 101-1 is used to receive scene information collected by the sensing device, obtain the user identity and user behavior recognition results according to the current classification model, and collect feedback information of the current user.

[0115] Among them, the perception device includes: an image sensor, a sound sensor, and a touch sensor; the current classification model is pre-trained based on the results of user and user behavior recognition.

[0116] The interactive behavior judgment unit 101-2 is used to judge whether an interaction has occurred with the current user based on the user identity and user behavior recognition results, including: judging whether the current user's confidence level is greater than a preset threshold based on the current user's voice information, facial image, distance between the current user and the robot, and current user's body movements collected by the sensing device;

[0117] When the confidence level of the current user is greater than a preset threshold, interaction occurs with the current user. The interaction mode selection unit 101-3 is configured to select and execute an interaction mode based on a preset correspondence between the recognition result and the interaction mode.

[0118] The interaction method includes at least one of the following methods: voice interaction, user graphical interface interaction, and body interaction.

[0119] The operation behavior judgment unit 101-4 is used to judge whether the interaction method is appropriate based on the feedback information of the current user on the interaction method collected by the perception device, including: judging whether the feedback information of the current user on the interaction method is the same as the expected feedback result, thereby judging whether the interaction method is appropriate.

[0120] For example, when the interaction method is voice interaction, it is determined whether the current user responds positively to the voice interaction information, thereby determining whether the interaction method is appropriate;

[0121] If the current user responds positively to the voice interaction information, the interaction method is considered appropriate;

[0122] If the current user does not respond to the voice interaction information or does not make a positive response, it is considered that the interaction method is inappropriate.

[0123] When the interactive mode is a user graphical interface interaction, determining whether the current user operates the currently displayed user graphical interface according to the voice prompt information or text prompt information, thereby determining whether the interactive mode is appropriate;

[0124] If the current user operates the currently displayed user graphical interface according to the voice prompt information or text prompt information, it is considered that the interaction mode is appropriate;

[0125] If the current user does not operate the current user graphical interface or exits the current user graphical interface, the interaction method is considered inappropriate. The feedback information of the current user regarding the interaction method collected by the sensing device includes: sound information, operation information of the user graphical interface, location information of the current user, and facial information of the current user.

[0126] The data information marking unit 101-5 is used to use the user information, user identity and user behavior recognition results as positive samples or negative samples according to the judgment result of the operation behavior judgment unit;

[0127] The classification model training unit 101-6 is used to train the current classification model according to the positive samples or negative samples.

[0128] The training method can be in various forms, for example:

[0129] When the number of the positive samples or negative samples reaches a preset number, the preset number of positive samples or negative samples are added to the training sample set, the current classification model is trained, and the current classification model is updated.

[0130] When the recognition result is a positive sample or a negative sample, the positive sample or the negative sample is added to the training sample set, the current classification model is trained, and the current classification model is updated in real time.

[0131] Preferably, the method for determining whether interaction with the current user occurs based on the user identity and user behavior recognition results includes:

[0132] Determine whether the current user's confidence level is greater than a preset threshold based on the current user's voice information, facial image, distance between the current user and the robot, and the current user's body movements collected by the sensing device;

[0133] When the confidence level of the current user is greater than a preset threshold, interaction occurs with the current user.

[0134] The interaction method includes at least one of the following methods: voice interaction, user graphical interface interaction, and body interaction.

[0135] Preferably, judging whether the interaction mode is appropriate based on the feedback information of the current user regarding the interaction mode collected by the sensing device is achieved by judging whether the feedback information of the current user regarding the interaction mode is the same as the expected feedback result, including:

[0136] Determining whether the current user has made a positive response to the voice interaction information, thereby determining whether the interaction method is appropriate;

[0137] If the current user responds positively to the voice interaction information, the interaction method is considered appropriate;

[0138] If the current user does not respond to the voice interaction information or does not make a positive response, it is considered that the interaction method is inappropriate.

[0139] Preferably, it is determined whether the current user operates the currently displayed user graphical interface according to the voice prompt information or the text prompt information, thereby determining whether the interaction mode is appropriate;

[0140] If the current user operates the currently displayed user graphical interface according to the voice prompt information or text prompt information, it is considered that the interaction mode is appropriate;

[0141] If the current user does not operate the current user graphical interface or exits the current user graphical interface, it is considered that the interaction mode is inappropriate.

[0142] The feedback information of the current user regarding the interaction mode collected by the sensing device includes: sound information, operation information of a user graphical interface, location information of the current user, and face information of the current user.

[0143] The method for training the current classification model according to the positive sample or the negative sample includes:

[0144] When the number of the positive samples or negative samples reaches a preset number, the preset number of positive samples or negative samples are added to the training sample set, the current classification model is trained, and the current classification model is updated.

[0145] When the recognition result is a positive sample or a negative sample, the positive sample or the negative sample is added to the training sample set, the current classification model is trained, and the current classification model is updated in real time.

[0146] In addition, the third embodiment of the present application further provides a robot that uses a robot adaptive learning device. Since the robot embodiment is basically similar to the device embodiment, the description is relatively simple. For relevant details, please refer to the partial description of the system embodiment. The device embodiment described below is only illustrative.

[0147] A robot employing a robot adaptive learning device, wherein the adaptive learning device comprises:

[0148] The operation behavior collection unit is used to collect scene information, obtain user identity and user behavior recognition results based on the current classification model, and collect feedback information from the current user.

[0149] The interactive behavior judgment unit is used to judge whether interaction occurs with the current user based on the user identity and user behavior recognition results.

[0150] The interactive mode selection unit is used to select and execute an interactive mode according to a preset correspondence between the recognition result and the interactive mode.

[0151] The operation behavior judgment unit is used to judge whether the interaction mode is appropriate based on the feedback information of the current user on the interaction mode collected by the sensing device.

[0152] a data information marking unit, configured to use the user information, user identity, and user behavior recognition results as positive samples or negative samples based on the judgment result of the operation behavior judgment unit;

[0153] The classification model training unit is used to train the current classification model according to the positive samples or negative samples.

[0154] Although the present application is disclosed as above with the preferred embodiments, it is not intended to limit the present application. Any person skilled in the art may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

[0155] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0156] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0157] 1. Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include non-transitory media such as modulated data signals and carrier waves.

[0158] 2. Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

Claims

1. A robot adaptive learning method, characterized in that: include: Receive scene information collected by the sensing device, substitute it into the current classification model, and obtain user identity and user behavior recognition results; Determine whether to interact with the current user based on the user identity and user behavior recognition results; If the result of the above determination of whether interaction with the current user occurs is yes, an interaction method is selected and executed according to the preset correspondence between the recognition result and the interaction method; Determining whether the interaction mode is appropriate based on feedback information of the current user regarding the interaction mode collected by the sensing device; If the result of the above determination of whether the interaction mode is appropriate is yes, the user information, user identity and user behavior recognition results collected by the sensing device are used as positive samples; If the result of the above determination of whether the interaction mode is appropriate is no, the user information, user identity and user behavior recognition results collected by the sensing device are used as negative samples; The positive samples or negative samples are added to a training sample set for training the current classification model.

2. The robot adaptive learning method according to claim 1, characterized in that: The sensing device includes: an image sensor, a sound sensor, and a touch sensor.

3. The robot adaptive learning method according to claim 1, characterized in that: The current classification model is pre-trained based on user and user behavior recognition results.

4. The method for robot adaptive learning according to claim 1, characterized in that: The determining whether interaction with the current user occurs based on the user identity and user behavior recognition results includes: Determine whether the current user's confidence level is greater than a preset threshold based on the current user's voice information, facial image, distance between the current user and the robot, and the current user's body movements collected by the sensing device; When the confidence level of the current user is greater than a preset threshold, interaction occurs with the current user.

5. The robot adaptive learning method according to claim 1, characterized in that: The interaction method includes at least one of the following methods: voice interaction, user graphical interface interaction, and body interaction.

6. The robot adaptive learning method according to claim 1, characterized in that: The judging whether the interaction method is appropriate based on the feedback information of the current user regarding the interaction method collected by the sensing device includes: judging whether the feedback information of the current user regarding the interaction method is the same as the expected feedback result, thereby judging whether the interaction method is appropriate.

7. The robot adaptive learning method according to claim 6, characterized in that: The determining whether the feedback information of the current user regarding the interaction mode is the same as the expected feedback result, thereby determining whether the interaction mode is appropriate, includes: Determining whether the current user has made a positive response to the voice interaction information, thereby determining whether the interaction method is appropriate; If the current user responds positively to the voice interaction information, the interaction method is considered appropriate; If the current user does not respond to the voice interaction information or does not make a positive response, it is considered that the interaction method is inappropriate.

8. The robot adaptive learning method according to claim 6, characterized in that: The determining whether the feedback information of the current user regarding the interaction mode is the same as the expected feedback result, thereby determining whether the interaction mode is appropriate, includes: Determining whether the current user operates the currently displayed user graphical interface according to the voice prompt information or text prompt information, thereby determining whether the interaction method is appropriate; If the current user operates the currently displayed user graphical interface according to the voice prompt information or text prompt information, it is considered that the interaction mode is appropriate; If the current user does not operate the current user graphical interface or exits the current user graphical interface, it is considered that the interaction mode is inappropriate.

9. The robot adaptive learning method according to claim 6, characterized in that: The feedback information of the current user regarding the interaction mode collected by the sensing device includes: sound information, operation information of a user graphical interface, location information of the current user, and face information of the current user.

10. The robot adaptive learning method according to claim 1, characterized in that: The adding the positive sample or the negative sample to a training sample set for training the current classification model includes: When the number of the positive samples or negative samples reaches a preset number, the preset number of positive samples or negative samples are added to the training sample set, the current classification model is trained, and the current classification model is updated.

11. The robot adaptive learning method according to claim 1, characterized in that: The adding the positive sample or the negative sample to a training sample set for training the current classification model includes: When the recognition result is a positive sample or a negative sample, the positive sample or the negative sample is added to the training sample set, the current classification model is trained, and the current classification model is updated in real time.

12. A robot adaptive learning device, characterized in that: include: The operation behavior collection unit is used to receive scene information collected by the sensing device, obtain the user identity and user behavior recognition results based on the current classification model, and collect feedback information from the current user; An interactive behavior judgment unit, configured to judge whether an interaction with the current user occurs based on the user identity and user behavior recognition result; An interactive mode selection unit, configured to select and execute an interactive mode based on a preset correspondence between the recognition result and the interactive mode; an operation behavior judgment unit, configured to judge whether the interaction mode is appropriate based on the feedback information of the current user regarding the interaction mode collected by the sensing device; A data information marking unit, configured to use the user information, user identity, and user behavior recognition results as positive samples or negative samples based on the judgment result of the operation behavior judgment unit; The classification model training unit is used to train the current classification model according to the positive samples or negative samples.

13. A robot, characterized in that: The robot adopts a robot adaptive learning device, and the adaptive learning device includes: The operation behavior collection unit is used to collect scene information, obtain user identity and user behavior recognition results based on the current classification model, and collect feedback information from the current user; An interactive behavior judgment unit, configured to judge whether an interaction with the current user occurs based on the user identity and user behavior recognition result; An interactive mode selection unit, configured to select and execute an interactive mode based on a preset correspondence between the recognition result and the interactive mode; an operation behavior judgment unit, configured to judge whether the interaction mode is appropriate based on the feedback information of the current user on the interaction mode collected by the sensing device; A data information marking unit, configured to use the user information, user identity, and user behavior recognition results as positive samples or negative samples based on the judgment result of the operation behavior judgment unit; The classification model training unit is used to train the current classification model according to the positive samples or negative samples.

Images