CN110866588B - A training and learning method and system for realizing the individualization of the learnable ability model of intelligent virtual digital animals - Google Patents

Abstract

The invention discloses a training and learning method and system for realizing the individualization of the learnable ability model of an intelligent virtual digital animal. The method includes: generating a program instance of an intelligent virtual digital animal on the cloud; constructing and producing an animal robot physical mapping for the cloud instance of the intelligent virtual digital animal, and distributing it to the user, and the animal robot downloads a corresponding learnable ability model from the cloud; the user and the animal robot The physical objects interact with natural scenes, and the animal robot automatically collects and generates personalized training data, trains and updates a copy of the learnable ability model and uploads it to the cloud, and the cloud updates the learnable ability model of the corresponding virtual digital animal. The present invention effectively solves the problem that the existing artificial intelligence model training mode has too high a threshold for ordinary people, cannot exert the value of the public labor force on a large scale, cannot participate in the personalized training of the virtual digital agent ability model, and lacks the real scene of interacting with users. And other issues.

Description

A training and learning method and system for realizing the personalization of the learnable ability model of intelligent virtual digital animals

technical field

The invention belongs to the technical field of machine learning training, and in particular relates to a training and learning method and system for realizing individualized learning ability models of intelligent virtual digital animals.

Background technique

Virtual digital animals are virtual objects described by defining basic attribute lists and behavior lists on the cloud server side through computer program data structures and code fragments. These virtual objects are usually conceptual mappings of one or more real animals in real scenes. And the concept map can even superimpose the attributes of other biological species and fictional species. With the development of artificial intelligence, it is proposed to add a "learnable ability model" intelligent module that can be learned through training to virtual digital animals, which is called intelligent virtual digital animals. The "learnable ability model" intelligent module mentioned here can typically be expressed by constructing a neural network model, and at the same time, by designing multiple intelligent modules for intelligent virtual digital animals that are oriented to different ability learning, the ability to support multiple abilities can be realized. Train to learn.

In recent years, with the rapid development of artificial intelligence technology theory and methods, and the great progress of intelligent manufacturing capabilities (robot dogs, robots, robot bee colonies and other intelligent bodies emerge in endlessly and are updated), users are more interested in traditional pure software interactive intelligent virtual digital Animals, put forward the new expectation of personalized training and learning ability, and innovation of interaction mode (traditionally, the interaction is completely in the virtual scene of the software interface, such as using the mouse or touch screen, and interacting with the virtual pet through the virtual graphics and picture interface drawn by the software) .

The current technical method for training intelligent modules of intelligent virtual digital animals is that professional computer talents, through professional means such as programming, use pre-collected data to train the learnable model to obtain a unified ability model, and then the obtained unified ability The model is copied indiscriminately to the user program object instance of each "intelligent virtual digital animal". The shortcomings and deficiencies of this training technology, its method or mode can be summed up as "lack of individualized learnable training of intelligent virtual digital animals based on real interaction scene data of users", "lack of human and virtual digital animals in real scenes". The physical interaction of animals” is specifically manifested in: users (ordinary people) cannot personally and directly train and adjust their own intelligent virtual digital animals, and users (ordinary people) can only interact in the software scene, not in the physical world. Interact directly with your own smart virtual digital animal. Therefore, it is difficult to meet the requirements such as intelligent digital pet dogs partially or completely replacing biological dogs (owners can selectively train biological dogs to learn personalized skills according to their own needs), so that ordinary people can participate in intelligent virtual digital pets without threshold like domesticated animals. Animal training, as well as intelligent service robot dogs and intelligent service robots that can be trained to learn and grow for specific user behavior habits, have a higher demand for personalized training and learning capabilities.

Contents of the invention

The purpose of the present invention is to overcome the shortcomings of the existing training and learning technology in the method mode, and provide a new training and learning method and system for realizing the individualization of the learnable ability model of intelligent virtual digital animals.

The present invention manufactures a corresponding animal robot for each "intelligent virtual digital animal", and the animal robot synchronously downloads the "learnable ability model" intelligent module of the "intelligent virtual digital animal" from the cloud, and stores it as a local copy. In the real scene where each user (ordinary person) directly interacts with his own animal robot, or in the process of the user consciously training (tuning, note: different from the traditional configuration of some fixed parameters) his own animal robot, automatic Generate personalized training data, train a copy of the learnable ability model and update the model parameters, and synchronize to the cloud, thus solving the "lack of personalized learning training for intelligent virtual digital animals based on real user interaction scene data" and "lack of real-world The physical interaction between people in the scene and virtual digital animals" and other issues.

The technical scheme that the present invention adopts is as follows:

In the first aspect, the present invention provides a training and learning method for realizing the individualization of the learnable ability model of an intelligent virtual digital animal, the steps of which include:

The actual animal robot downloads a copy of the learnable ability model from the cloud server; the animal robot is an animal robot manufactured according to the intelligent virtual digital animal instance on the cloud server;

The animal robot realizes the personalized training of the learnable ability model of the intelligent virtual digital animal through scene interaction with the user;

The actual animal robot uploads a copy of the trained learnable ability model to the cloud server, so that the cloud server can update the learnable ability model of the intelligent virtual digital animal according to the uploaded copy of the learnable ability model.

Further, the animal robot downloads a copy of the learnable ability model from the cloud server, including:

The animal robot sends an identity verification request to the cloud server, so that the cloud can perform identity verification after receiving the request, and write and lock the corresponding learnable ability model after the verification is passed;

The animal robot in kind requests to the cloud server to download the learnable ability model;

The data integrity of the learnable ability model downloaded by the physical verification of the animal robot;

The animal robot object will load the downloaded learnable ability model, and set it as the "initial copy of the learnable ability model" of the animal robot object.

Further, the animal robot realizes the personalized training of the learnable ability model of the intelligent virtual digital animal by interacting with the user in the scene, including:

The user selects a training target for the animal robot;

The user interacts with the animal robot in a personalized scene;

During the interaction process, the animal robot automatically collects and generates personalized training data through sensors, and labels the data according to the training objectives;

Based on the collected data, the copy of the learnable ability model is automatically trained and updated.

In a second aspect, the present invention provides a training and learning method for realizing the individualization of the learnable ability model of an intelligent virtual digital animal, the steps of which include:

The cloud server generates intelligent virtual digital animal instances;

According to the request of the real animal robot, the cloud server sends a copy of the learnable ability model to the real animal robot; the real animal robot is an animal robot manufactured according to the intelligent virtual digital animal instance on the cloud server side;

The cloud server end receives the learned ability model copy of the training completed by the animal robot, and updates the learnable ability model of the intelligent virtual digital animal; Interact with and perform personalized training to complete a copy of the learnable ability model.

Further, the cloud server adopts the following steps to generate an intelligent virtual digital animal instance:

register users;

The user selects the type of intelligent virtual digital animal to receive;

Generate intelligent virtual digital animal program instances for users.

Further, the learning ability model of updating the intelligent virtual digital animal includes:

Verify the data integrity of the copy of the learnable capability model;

Unlock the learnable capability model write lock;

Update the learnable ability model of the intelligent virtual digital animal according to the uploaded copy of the learnable ability model.

Further, the cloud server end is provided with a cloud hosting platform for the program object instance of the smart virtual digital animal, and the cloud hosting platform for the program object instance of the smart virtual digital animal runs the program logic of the smart virtual digital animal in the cloud, stores user resources, saves and manages virtual Digital animal types, all attributes, behavior lists, and attribute lists of intelligent virtual digital animals are responsible for interacting with animal robots and providing access pages for users to directly access; the virtual digital animal types define data structures of different animal categories, Including the framework of animal attributes, animal behaviors, and animal learnable ability models; the intelligent virtual digital animal is an instantiation program object of the intelligent virtual digital animal type, which consists of the learnable ability model, attribute list, and behavior list of the intelligent virtual digital animal The attribute list records the attribute values of the intelligent virtual animal instance, and the behavior list defines the basic action behavior of the animal.

Further, the learnable capability model is composed of a learnable capability model structure definition, model parameters, and model parameter integrity check values; a set of parameter sets, namely model parameters, are obtained through training and learning, and for each set of stable Model parameters, generate integrity check values.

In a third aspect, the present invention provides an animal robot, which includes a physical body and a computing system; the computing system downloads a copy of the learnable ability model of an intelligent virtual digital animal from a cloud server, and realizes the learnable ability model copy of the intelligent virtual digital animal through scene interaction with the user. Personalized training of the ability model, and upload the copy of the trained learnable ability model to the cloud server, so that the cloud server can update the learnable ability model of the intelligent virtual digital animal according to the uploaded copy of the learnable ability model.

Further, the animal robot is an animal robot manufactured according to the structure information of the intelligent virtual digital animal instance on the cloud server side, which is a physical mapping of the intelligent virtual digital animal instance, and realizes the extension of the virtual digital world to the physical world; The physical body includes the physical presentation of the attributes and behaviors of intelligent virtual digital animals, including color, texture, programmable control of manipulators, manipulators, and visual sensors. The computing system includes an instance management agent module, a learnable ability model copy module, NPU module, control logic integrated circuit, storage module.

Further, the instance management agent module communicates with the instance management module on the cloud server side through the network, and is responsible for submitting an identity authentication request to the cloud hosting platform, downloading the learnable capability model, verifying the downloaded learnable capability model, and installing the learnable capability model A copy of the local model of the animal robot, a program for scheduling training and updating the copy of the learnable ability model, uploading the copy of the learned ability model after training and updating to the cloud; the NPU module is a special processor for intelligent computing, used to accelerate neural network reasoning and training performance, the control logic integrated circuit is a processor that realizes the programmable control of the animal robot, and the storage module provides a fast storage function for the copy of the learnable ability model, training data, and program operation.

In the fourth aspect, the present invention provides a cloud server, which is provided with an intelligent virtual digital animal program object instance cloud hosting platform;

The intelligent virtual digital animal program object instance cloud hosting platform generates an intelligent virtual digital animal instance, and according to the request of the animal robot, sends a copy of the learnable ability model to the animal robot; Animal robots manufactured by virtual digital animal instances;

The intelligent virtual digital animal program object instance cloud hosting platform receives the copy of the trained learnable ability model uploaded by the animal robot, and updates the learnable ability model of the intelligent virtual digital animal; the trained copy of the learnable ability model, It is a copy of the learnable ability model completed by the user through scene interaction with the animal robot and personalized training.

Further, the cloud server includes an instance management module, a virtual digital animal type module and an intelligent virtual animal instance module;

The instance management module manages the whole life cycle of the intelligent virtual digital animal instance, manages the learnable ability model of the instance, and is responsible for interactive communication with the instance management agent module running in the animal robot object through the network;

The virtual digital animal type module includes different types of animal types, and defines data structures of different animal types, including animal attributes, animal behaviors, and the framework of animal learnable ability models;

The intelligent virtual animal instance module is a specific instantiation of the intelligent virtual digital animal type, including a learnable ability model, an attribute list, and a behavior list.

In the fifth aspect, the present invention provides a training and learning system for realizing the personalization of the learnable ability model of an intelligent virtual digital animal, which includes the above-mentioned animal robot and a cloud server.

The advantage of the present invention compared with prior art is:

(1) At present, the training of the artificial intelligence ability model (such as RNN, CNN network, deep reinforcement model, etc.) of intelligent virtual digital animals requires professional computer engineers, and the threshold is high; and the present invention proposes online intelligent virtual digital animals The physical mapping of animal robots under the production line, the method and mode for ordinary people to train ability models for intelligent virtual digital animals in the natural scene interaction with offline animal robots, significantly reduces the threshold for training artificial intelligence ability models of virtual digital animals, This new technology model provides a convenient way for the general labor force to participate in the ability model training of virtual digital agents.

(2) At present, the learning ability model training of intelligent virtual digital animals adopts the method of unified data collection, unified training, and unified deployment online or before going online, and there are problems of lack of personalized intelligence, flexibility, and sense of participation in real scenes The training and learning method and system proposed by the present invention are very flexible, and solve the problems such as the personalization of the learnable ability model of intelligent virtual digital animals for a large number of users, and the real scene physical interaction between people and virtual digital animals. This combination of online intelligent virtual digital instances and offline physical robot mapping training and interaction methods, in order to achieve, such as "under the training" of the "master", personalized abilities (such as learning to call, recognize colors, pick up and throw objects, etc.) Applications such as digital pets that continuously "learn and grow" provide technical methods and models.

Description of drawings

Fig. 1 is the overall structural representation of the present invention;

Fig. 2 is the process of generating an animal robot mapping for an intelligent virtual digital animal structure of the present invention;

Fig. 3 is the training and learning process of realizing the personalization of the learnable ability model of the intelligent virtual digital animal according to the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and examples, but the scope of the present invention is not limited in any way.

The individualized training and learning system for realizing the learnable ability model of intelligent virtual digital animals in this embodiment, as shown in FIG. . The intelligent virtual digital animal program object instance cloud hosting platform 1 is mainly composed of an instance management module 101, a virtual digital animal type module 102, and an intelligent virtual digital animal instance module 103 (multiple different instances form an instance set). The intelligent virtual digital animal instance module 103 includes a learnable ability model 1031 , an attribute list 1032 and a behavior list 1033 . The animal robot object 2 is composed of an instance management agent module 201, a learnable ability model copy module 202, an NPU (neural network processor) module 203, a storage module 205, a control logic integrated circuit 204, and a physical body. The modules indicated by the dotted boxes in the figure are not the core components of the present invention. Person 4 is the user, who is the owner of a certain intelligent virtual digital animal instance and the corresponding physical animal robot, and interacts directly with the corresponding physical animal robot in the physical scene. Sensors, fiber optic sensors, etc. collect and generate training data in the interaction, and the training updates the copy of the learnable ability model. The intelligent virtual digital animal program object instance cloud hosting platform 1 runs on the cloud, the real animal robot 2 is on the user end (ordinary person), and the network 3 connects the cloud and the real animal robot to provide a data transmission link.

The intelligent virtual digital animal program object instance cloud hosting platform 1 runs the intelligent virtual digital animal program logic in the cloud, stores user resources, saves and manages virtual digital animal types, all attributes and behavior attribute lists of intelligent virtual digital animals, and is responsible for tracking animals The robot interacts with physical objects, and also provides access pages for users (humans) to directly access, etc. Among them, the instance management module 101 manages the whole life cycle of the intelligent virtual digital animal instance, manages the learnable ability model of the instance (locking the learnable ability model, unlocking the learnable ability model, updating the learnable ability model, etc.), and is responsible for tracking through the network 3 The instance management agent module 201 running in the animal robot object 2 conducts interactive communication, and the communication content mainly includes verifying user and instance identities, providing learnable ability model downloads, receiving learnable ability model uploads, and verifying learnable ability models.

The virtual digital animal type module 102 is different types of animal types managed by the platform, and defines the data structure of different animal types, such as the framework of animal attributes, animal behaviors, and animal learnable ability models.

The intelligent virtual animal instance module 103 is the specific instantiation of intelligent virtual digital animal types. Each animal type usually has multiple instances, and each instance has an independently saved learnable ability model. All instances form an instance set. An instance of an intelligent virtual animal is composed of a learnable ability model 1031 , an attribute list 1032 , a behavior list 1033 and so on. The attribute list defines the attributes of the intelligent virtual animal instance (such as unique ID, color, age, gender, etc.), and the behavior list defines the basic behavior of the animal (such as running, walking, turning, jumping, screaming, etc.).

The learnable capability model 1031 is composed of learnable model structure definitions, model parameters, and model parameter integrity check values. The learnable model structure can be artificial neural network (ANN), deep reinforcement learning (Reinforcement Learning), pulse Neural network (SNN), etc., different selections of learnable model structures do not affect the rights of the present invention. A set of parameter sets, that is, model parameters, is obtained through training and learning. For each set of stable model parameters, a hash value will be generated as an integrity check value (the integrity check value generation algorithm is optional, and the hash algorithm is selected Algorithms other than that do not affect the rights of the present invention).

The animal robot object 2 is an animal robot manufactured according to the structure information of the intelligent virtual digital animal instance module 103, and is a physical mapping of the intelligent virtual digital animal instance, realizing the extension of the virtual digital world to the physical world. This method and mode of training intelligent virtual digital animals is proposed for the first time by the method of the present invention. It is to realize that ordinary public users (rather than professional computer talents) can purposefully generate data and train models in combination with personal wishes in real interactive scenes, so as to achieve A critical step towards personalized ability training for intelligent virtual digital animals. The animal robot is composed of a physical body and a computing system. Among them, the physical body includes the physical presentation of the attributes and behaviors of intelligent virtual digital animals, such as color, texture, programmable control of manipulators, mechanical arms, visual sensors, etc.; the design, realization and manufacturing methods of these physical bodies are not part of the present invention. Important components, different selections of their manufacturing methods, technologies, processes, etc. do not affect the rights of the present invention. The computing system is the focus, mainly composed of instance management agent module 201 , learnable capability model copy module 202 , NPU module 203 , control logic integrated circuit 204 , and storage module 205 .

The instance management agent module 201 communicates with the instance management module 101 through the network 3, and is responsible for submitting an identity authentication request to the cloud hosting platform, downloading the learnable capability model 1031, verifying the downloaded learnable capability model 1031, and installing the learnable capability model 1031 is a local copy, uploading the updated learnable ability model copy 202 after client training to the cloud. The NPU module 203 is a dedicated processor for intelligent computing, which is used to accelerate neural network reasoning and training performance, the control logic integrated circuit 204 is a processor for realizing programmable control of animal robots, and the storage module 205 is a learnable Model copy, training data, program operation, etc. provide fast storage, and different types of modules 203, 204, 205 do not affect the rights of the present invention.

Below in conjunction with Fig. 2 and Fig. 3, the present invention will be further described through the embodiment in which two ordinary users train their own intelligent virtual digital dog to learn different skills respectively. Fig. 2 illustrates the process of generating physical mapping of animal robots for the construction of intelligent virtual digital animals, and Fig. 3 illustrates the training and learning process of realizing the individualized learning ability model of intelligent virtual digital animals.

(1) Generating animal-robot physical mappings for intelligent virtual digital animal construction. As shown in Figure 2, it mainly includes the following steps:

(1.1) User A registers the cloud hosting platform 1 for the program object instance of the intelligent virtual digital animal, and receives the instance of the intelligent virtual digital animal; , gray color, age 3 months, etc.; the learnable ability model of No. 10001 intelligent virtual digital dog integrates image recognition, speech understanding, motion trajectory tracking and other model frameworks based on deep learning, so it has the ability to obtain recognition through training and learning Images, understanding speech, automatic adjustment of motion trajectory and other capabilities. The image recognition model is realized based on CNN network (convolutional neural network);

(1.2) The physical mapping of the animal robot that produces the virtual digital dog No. 10001;

(1.3) Distribute No. 10001 animal robot to user A;

(1.4) User A activates the animal robot No. 10001, and the robot is connected to the Internet;

(1.5) No. 10001 animal robot sends an identity verification request to the cloud hosting platform;

(1.6) The cloud instance management module receives the request, performs identity verification, and identifies it as the real animal robot No. 10001 (if the identity verification fails, feedback the authentication failure information to the real animal robot, and the real animal robot can resend the identity verification request), Then write-lock the learnable ability model of No. 10001 intelligent virtual digital animal (to avoid model inconsistency or write conflict caused by parallel writing);

(1.7) No. 10001 animal robot requests to download the learnable ability model of No. 10001 intelligent virtual digital animal hosted in the cloud. The download is completed and passed the model integrity verification (if the integrity verification fails, it means that the model download is incomplete. need to re-download);

(1.8) Load the downloaded model and initialize it as a copy of the learnable ability model of the No. 10001 animal robot.

(2) Carry out personalized training on the learnable ability model of the intelligent virtual digital animal. As shown in Figure 3, it mainly includes the following steps:

(2.1) User A chooses a training target through the physical animal robot No. 10001 (it can be realized through voice wake-up, menu selection and other technologies, the method is not limited): learn to recognize apples; then the real robot enters the training interaction mode;

(2.2) User A shows different apples and different sides of apples in front of No. 10001 animal robot. The animal robot collects photos of apples through visual sensors. Every movement and angle change of the apple will trigger the collection of new photos;

(2.3) Combined with the training target set in step 2.1), these collected photos are marked as "apple", and the No. 10001 animal robot uses NPU acceleration to train and update the image classification subsystem model that can learn a copy of the ability model until convergence. Here the image classification subsystem is realized based on technologies such as CNN convolutional neural network and multiple classifiers (but other technical choices do not affect the rights of the present invention);

(2.4) No. 10001 animal robot uploads a copy of the learned ability model after training to the cloud;

(2.5) The cloud instance management module receives the upload and verifies the integrity of the model data (if the integrity verification fails, it means that the model upload is incomplete and needs to be re-uploaded);

(2.6) The cloud instance management module releases the write lock of the learnable ability model of the 10001 intelligent virtual digital animal instance;

(2.7) The cloud updates the learnable ability model of the virtual digital animal No. 10001 according to the uploaded copy of the ability model. Therefore, user A's intelligent virtual digital animal has a personalized ability to identify apples. During the training process, user A helped the model generate personalized training data through scene interaction, which is suitable for the general public.

The unique ID number of the smart virtual digital dog received by user B is 10002. Through a process similar to that of user A, user B can train his own smart virtual digital dog No. 10002 to acquire the ability to recognize football, basketball, oranges, etc. Ordinary users can also train their own intelligent virtual digital animals to obtain a specific form of trajectory tracking ability and other learnable abilities. In this low-tech threshold training method and mode, the intelligent virtual digital dog of users A and B presents a differentiated and personalized "growth" of trainable learning ability.

The new model training technology mode proposed in the present invention "a training and learning method and system for realizing the individualization of the learnable ability model of intelligent virtual digital animals" provides for the general public labor force to participate in the ability model training of virtual digital agents. A simple, convenient and efficient way.

In the specific implementation of the solution of the present invention, during the training process, after the user helps the animal robot to generate or collect personalized training data through scene interaction, if the system does not immediately perform model training on the animal robot, but uploads the data first Going to the cloud, postponing the training and updating the learnable ability model of the user's corresponding virtual digital animal on the cloud, and realizing the personalization of the learnable ability model are regarded as the deformation mode of the present invention.

The above embodiments are only used to illustrate the technical solution of the present invention and not to limit it. Those skilled in the art can modify or equivalently replace the technical solution of the present invention without departing from the principle and scope of the present invention. The scope of protection should be determined by the claims.

Claims (12)

1. A method of training and learning that realizes the personalization of the learnable ability model of an intelligent virtual digital animal, the steps of which include: The actual animal robot downloads a copy of the learnable ability model from the cloud server; the animal robot is an animal robot manufactured according to the intelligent virtual digital animal instance on the cloud server; The animal robot realizes the personalized training of the learnable ability model of the intelligent virtual digital animal through scene interaction with the user; The animal robot uploads a copy of the trained learnable ability model to the cloud server, so that the cloud server can update the learnable ability model of the intelligent virtual digital animal according to the uploaded copy of the learnable ability model; The animal robot is an animal robot manufactured according to the structure information of the intelligent virtual digital animal instance on the cloud server side, and is a physical mapping of the intelligent virtual digital animal instance, realizing the extension of the virtual digital world to the physical world; the animal robot includes The physical body and the computing system; the computing system downloads a copy of the learnable ability model of the intelligent virtual digital animal from the cloud server, realizes the personalized training of the learnable ability model through scene interaction with the user, and transfers the learned learnable ability model after the training A copy of the ability model is uploaded to the cloud server, so that the cloud server can update the learnable ability model of the intelligent virtual digital animal according to the uploaded copy of the learnable ability model; present; The cloud server end is provided with an intelligent virtual digital animal program object instance cloud hosting platform, and the intelligent virtual digital animal program object instance cloud hosting platform runs the program logic of the intelligent virtual digital animal in the cloud, stores user resources, saves and manages the virtual digital animal type , a list of all attributes and behaviors of intelligent virtual digital animals, and a list of attributes, which are responsible for interacting with animal robots and providing access pages for users to directly access; the types of virtual digital animals define data structures of different animal categories, including animal attributes, The frame of animal behavior and the learnable ability model of animals; the intelligent virtual digital animal is an instantiation program object of the intelligent virtual digital animal type, which is composed of the learnable ability model, attribute list, and behavior list of the intelligent virtual digital animal, wherein the attribute The list records the attribute values of the intelligent virtual animal instance, and the behavior list defines the basic action behavior of the animal. 2. The method according to claim 1, wherein the animal robot downloads a learnable ability model copy from the cloud server in kind, including: The animal robot sends an identity verification request to the cloud server, so that the cloud can perform identity verification after receiving the request, and write and lock the corresponding learnable ability model after the verification is passed; The animal robot in kind requests to the cloud server to download the learnable ability model; The data integrity of the learnable ability model downloaded by the physical verification of the animal robot; The animal robot object will load the downloaded learnable ability model, and set it as the "initial copy of the learnable ability model" of the animal robot object. 3. The method according to claim 1, characterized in that, the physical object of the animal robot realizes the personalized training of the learnable ability model of the intelligent virtual digital animal by interacting with the user, including: The user selects a training target for the animal robot; The user interacts with the animal robot in a personalized scene; During the interaction process, the animal robot automatically collects and generates personalized training data through sensors, and labels the data according to the training objectives; Based on the collected data, the copy of the learnable ability model is automatically trained and updated. 4. A method of training and learning that realizes the individualization of the learnable ability model of an intelligent virtual digital animal, the steps of which include: The cloud server generates intelligent virtual digital animal instances; According to the request of the real animal robot, the cloud server sends a copy of the learnable ability model to the real animal robot; the real animal robot is an animal robot manufactured according to the intelligent virtual digital animal instance on the cloud server side; The cloud server end receives the learned ability model copy of the training completed by the animal robot, and updates the learnable ability model of the intelligent virtual digital animal; A copy of the learnable ability model completed by interaction and personalized training; The animal robot is an animal robot manufactured according to the structure information of the intelligent virtual digital animal instance on the cloud server side, and is a physical mapping of the intelligent virtual digital animal instance, realizing the extension of the virtual digital world to the physical world; the animal robot includes The physical body and the computing system; the computing system downloads a copy of the learnable ability model of the intelligent virtual digital animal from the cloud server, realizes the personalized training of the learnable ability model through scene interaction with the user, and transfers the learned learnable ability model after the training A copy of the ability model is uploaded to the cloud server, so that the cloud server can update the learnable ability model of the intelligent virtual digital animal according to the uploaded copy of the learnable ability model; present; The cloud server end is provided with an intelligent virtual digital animal program object instance cloud hosting platform, and the intelligent virtual digital animal program object instance cloud hosting platform runs the program logic of the intelligent virtual digital animal in the cloud, stores user resources, saves and manages the virtual digital animal type , a list of all attributes and behaviors of intelligent virtual digital animals, and a list of attributes, which are responsible for interacting with animal robots and providing access pages for users to directly access; the types of virtual digital animals define data structures of different animal categories, including animal attributes, The frame of animal behavior and the learnable ability model of animals; the intelligent virtual digital animal is an instantiated program object of the intelligent virtual digital animal type, which is composed of the learnable ability model, attribute list, and behavior list of the intelligent virtual digital animal, wherein the attribute The list records the attribute values of the intelligent virtual animal instance, and the behavior list defines the basic action behavior of the animal. 5. The method according to claim 4, wherein the cloud server uses the following steps to generate an intelligent virtual digital animal instance: register users; The user selects the type of intelligent virtual digital animal to receive; Generate intelligent virtual digital animal program instances for users. 6. The method according to claim 4, wherein said updating the learnable ability model of the intelligent virtual digital animal comprises: Verify the data integrity of the copy of the learnable capability model; Unlock the learnable capability model write lock; Update the learnable ability model of the intelligent virtual digital animal according to the uploaded copy of the learnable ability model. 7. The method according to claim 4, wherein the learnable capability model is composed of a learnable capability model structure definition, model parameters, and model parameter integrity check values; a set of parameters is obtained through training and learning The set, that is, the model parameters, generates an integrity check value for each set of stable model parameters. 8. An animal robot realizing the method of claim 1, comprising a physical body and a computing system; the computing system downloads a copy of the learnable ability model of an intelligent virtual digital animal from the cloud server, and communicates with the user Scene interaction realizes personalized training of the learnable ability model, and uploads a copy of the trained learnable ability model to the cloud server, so that the cloud server can update the ability of the intelligent virtual digital animal according to the uploaded copy of the learnable ability model. learning ability model; the animal robot is an animal robot manufactured according to the structure information of the intelligent virtual digital animal instance on the cloud server side, and is a physical mapping of the intelligent virtual digital animal instance, realizing the extension of the virtual digital world to the physical world; The physical body includes the physical presentation of the attributes and behaviors of intelligent virtual digital animals. 9. The animal robot according to claim 8, wherein the physical body includes color, texture, programmable manipulator, mechanical arm, and visual sensor; the computing system includes an instance management agent module, a learnable ability Model copy module, NPU module, control logic integrated circuit, storage module. 10. The animal robot according to claim 9, wherein the instance management agent module communicates with the instance management module on the cloud server side through the network, and is responsible for submitting an identity authentication request, downloading a learnable ability model, and verifying the ability to the cloud hosting platform. Download the learnable ability model, install the learnable ability model as a local model copy of the animal robot, schedule the program for training and update the learnable ability model copy, and upload the updated learnable ability model copy to the cloud; the NPU module It is a special processor for intelligent computing, which is used to accelerate the performance of neural network reasoning and training. The control logic integrated circuit is a processor that realizes programmable control of animal robots. The storage module is a copy of the learnable ability model, training data, program Run provides fast storage capabilities. 11. A cloud server realizing the method according to claim 4, characterized in that, it is provided with an intelligent virtual digital animal program object instance cloud hosting platform; The intelligent virtual digital animal program object instance cloud hosting platform generates an intelligent virtual digital animal instance, and sends a copy of the learnable ability model to the animal robot according to the request of the animal robot; Animal robots manufactured by virtual digital animal instances; The intelligent virtual digital animal program object instance cloud hosting platform receives the copy of the trained learnable ability model uploaded by the animal robot, and updates the learnable ability model of the intelligent virtual digital animal; the trained copy of the learnable ability model, It is a copy of the learnable ability model completed by the user through scene interaction with the animal robot and personalized training; The cloud server includes an instance management module, a virtual digital animal type module and an intelligent virtual animal instance module; The instance management module manages the entire life cycle of the intelligent virtual digital animal instance, manages the learnable ability model of the instance, and is responsible for interactive communication with the instance management agent module running in the animal robot object through the network; The virtual digital animal type module includes different types of animal types, and defines data structures of different animal types, including animal attributes, animal behaviors, and the framework of animal learnable ability models; The intelligent virtual animal instance module is a concrete instantiation of the type of intelligent virtual digital animal, including a learnable ability model, an attribute list, and a behavior list. 12. A training and learning system that realizes the individualization of the learnable ability model of an intelligent virtual digital animal, characterized in that it includes the animal robot according to any one of claims 8 to 10, and the cloud robot according to claim 11. server.

Images