Development method of multi-mode emotion recognition software for voice interaction of intelligent robot

Abstract

The invention provides a multi-modal emotion recognition software development method for voice interaction of an intelligent robot, and the method achieves the synchronous collection and fusion of multi-modal data through the weighted fusion preprocessing of voice signals, facial expression video streams and text contents. A DCNN and LSTM combined model is adopted to deeply extract emotional features, and an attention mechanism is introduced to strengthen key information capture. And a comprehensive emotion representation vector is calculated by using a cross-modal emotion association network CMAN, so that the emotion distinguishing capability is improved. An emotion classifier combining an SVM and a neural network is constructed, and refined output of emotion categories and intensity is achieved. Through real-time feedback and adaptive optimization steps, the model can continuously learn and adapt to user emotion expression habits, and accuracy and user satisfaction are improved. According to the invention, more humanized and accurate multi-modal emotion recognition technical support is provided for voice interaction of the intelligent robot.

Description

Technical Field

[0001] The present invention belongs to the field of software development methods, and specifically discloses a method for developing multimodal emotion recognition software for intelligent robot voice interaction. Background Art

[0002] As artificial intelligence and robotics are increasingly integrated into our lives, intelligent robots are no longer just simple tools or devices. They are expected to become intelligent partners that can understand human emotions and provide personalized services. In order to achieve this vision, voice interaction technology is particularly important because it provides the most direct and natural way of communication between robots and users.

[0003] However, for robots to truly "understand" human speech, it is not as simple as just recognizing the text content in a sentence. More importantly, robots need to be able to capture the emotional information hidden behind the voice, such as the speaker's joy, anger, sadness, and attitude. For robots, this emotional information is the key to understanding the user's true intentions and making appropriate responses.

[0004] Multimodal emotion recognition technology has emerged, which makes comprehensive use of multiple information modalities such as voice, facial expressions, and text in order to capture and identify the user's emotional state more comprehensively and accurately. The emergence of this technology has undoubtedly brought a qualitative leap in the voice interaction capabilities of intelligent robots.

[0005] However, despite the many theoretical advantages of multimodal emotion recognition technology, it still faces many challenges in practical applications. First, data from different modalities often differ in quality and importance. For example, in some cases, speech signals may be blurred by noise, while facial expressions may not be accurately captured due to lighting problems or occlusion. How to reasonably integrate these different modal data to ensure that each modality can play its due role is an urgent problem to be solved.

[0006] Secondly, the extraction of emotional features is also a major difficulty. Emotion is a complex and subtle psychological phenomenon, which is often expressed through a variety of subtle clues. Existing feature extraction methods often find it difficult to fully capture these subtle changes, resulting in limited accuracy in emotion recognition.

[0007] Furthermore, as time goes by and the user groups change, the way and type of emotional expressions that robots face may also change. This requires the emotion recognition model to have sufficient adaptive capabilities to maintain high performance in a changing environment. However, existing models often perform poorly in this regard, and performance is prone to degradation in long-term use. Summary of the invention

[0008] In view of the above problems, the present invention proposes a method for developing multimodal emotion recognition software for intelligent robot voice interaction.

[0009] The present invention includes the following technical solutions:

[0010] A method for developing multimodal emotion recognition software for intelligent robot voice interaction, characterized in that it includes the following steps:

[0011] S1 multimodal data fusion preprocessing steps:

[0012] The weighted fusion method is used to synchronously collect and preprocess voice signals, facial expression video streams and text content; the weight calculation is based on the quality score of each modal signal and the preset importance adjustment coefficient, which is determined by exponential function and normalization processing. This step enables modal data of different quality and importance to be fused in proportion, providing a reliable basis for subsequent processing;

[0013] S2 emotional feature deep extraction steps:

[0014] Design a combined model that combines the deep convolutional neural network (DCNN) and the long short-term memory network (LSTM) to extract high-level features of speech acoustic features, facial expression geometric features, and text emotional vocabulary. Introduce an attention mechanism to dynamically adjust the importance of different time steps or feature points to enhance the model's ability to capture key information.

[0015] S3 cross-modal sentiment association learning steps:

[0016] Using the cross-modal emotion association network CMAN, the multimodal attention mechanism is used to fuse the extracted features, calculate the comprehensive emotion representation vector, and use the triple loss function to optimize the model, so that similar emotion samples are close in the feature space and different emotion samples are far away, thus improving the model's ability to distinguish complex emotions;

[0017] S4 sentiment classification and intensity assessment steps:

[0018] Construct a sentiment classifier, combine support vector machine (SVM) and neural network, output sentiment categories and their intensity levels at the same time, introduce sentiment intensity evaluation strategy, calculate sentiment intensity based on comprehensive sentiment representation vector, and realize refined processing of sentiment recognition;

[0019] S5 real-time feedback and adaptive optimization steps:

[0020] Implement a closed-loop feedback mechanism, adjust model parameters based on real-time user feedback, use online learning algorithms and incremental SVM to instantly update model parameters based on new samples and their labels, so that the model can continuously learn and adapt to the emotional expression habits of different users, thereby improving long-term accuracy and user satisfaction.

[0021] Furthermore, the present invention discloses a computer storage device for executing the above-mentioned method for developing multimodal emotion recognition software for intelligent robot voice interaction.

[0022] Furthermore, the present invention discloses a data processing server, characterized in that it comprises the computer storage device as described in claim 7.

[0023] Furthermore, the present invention discloses an intelligent robot capable of executing the software developed by the above development method.

[0024] The present invention has the following beneficial effects:

[0025] The present invention proposes a new multimodal emotion recognition software development method. The core purpose of the present invention is to comprehensively improve the emotion recognition ability of intelligent robots in voice interaction through a series of innovative technical means. Specifically, the present invention aims to build a more efficient, accurate and highly adaptive multimodal emotion recognition system by optimizing the fusion preprocessing process of multimodal data, deepening the extraction method of emotional features, strengthening the cross-modal emotion association learning mechanism, refining the emotion classification and intensity evaluation strategy, and realizing the real-time feedback and adaptive optimization function of the model. This will provide a more accurate emotional understanding foundation for intelligent robots, enabling them to better adapt to and meet the personalized needs of users, thereby significantly improving the user experience and satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a main flow chart of the development method of the present invention;

[0027] Figure 2 It is the main module of the intelligent robot in Example 2. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] Example 1

[0030] A method for developing multimodal emotion recognition software for intelligent robot voice interaction, such as Figure 1 As shown, the following steps are included:

[0031] S1 multimodal data fusion preprocessing steps:

[0032] S11 Synchronous Data Collection:

[0033] Start audio, video and text acquisition devices in real time and synchronously to obtain the user's voice signal, facial expression video stream and text content;

[0034] S12 modal quality assessment:

[0035] Perform a quality assessment on each acquired modal data and determine its quality score, which is determined based on the signal clarity, completeness, signal-to-noise ratio, and user feedback;

[0036] S13 weight calculation:

[0037] The weight of each modality in the fusion process is calculated based on the quality score of each modality data and the preset importance adjustment coefficient; the weight calculation takes into account the quality of each modality data and its importance in practical applications, ensuring that high-quality and important modalities occupy a larger proportion in the fusion;

[0038] S14 multimodal data fusion:

[0039] The calculated weights are used to perform weighted fusion on the synchronously collected modal data. During the fusion process, the contribution of each modal data in the fusion result is adjusted according to the weight, so that modal data of different quality and importance can be fused in proportion.

[0040] S15 data preprocessing:

[0041] The fused multimodal data is preprocessed, including denoising, standardization, and normalization, to achieve data consistency and reliability; the preprocessed data provides high-quality data input for subsequent deep extraction and recognition of emotional features.

[0042] S2 emotional feature deep extraction steps:

[0043] S21 receives and processes the pre-processed voice signal, facial expression video stream and text content;

[0044] S22 uses a deep convolutional neural network (DCNN) to deeply extract the acoustic features of speech signals. Through multi-layer convolution and pooling operations, a high-level feature representation of the speech signal is obtained. The feature extraction formula is expressed as:

[0045] F speech =DCNN(Speechsignal )

[0046] Among them, F speech Represents the extracted speech features, Speech signal represents an input speech signal;

[0047] S23 uses the long short-term memory network (LSTM) to deeply extract the geometric features of facial expression video streams. By capturing the long-term dependencies in the sequence data, it obtains a high-level feature representation of facial expressions. The feature extraction formula is expressed as:

[0048] F face =LSTM(Face stream ) Among them, F face Represents the extracted facial expression features, Face stream represents the input facial expression video stream;

[0049] S24 extracts sentiment words from the text content, converts the text into a vector representation using word embedding technology, and then extracts high-level features through a neural network to obtain a high-level feature representation of the text sentiment. The feature extraction formula is expressed as:

[0050] F text =NN(Embedding(Text content ))

[0051] Among them, F text Represents the extracted text features, Text content Represents the input text content, Embedding represents the word embedding operation, and NN represents the neural network;

[0052] S25 introduces the attention mechanism to dynamically weight the extracted speech, facial expression and text features to highlight key features and improve the accuracy of emotion recognition. The weighted formula is expressed as:

[0053] F fused =α·F speech +β·F face +γ·F text

[0054] Among them, F fused represents the fused features, α, β, and γ represent the attention weights of speech, facial expression, and text features, respectively, satisfying α+β+γ=1, and the weight values ​​are obtained according to the learning during the model training process.

[0055] S3 cross-modal sentiment association learning steps:

[0056] S31 receives and processes the deeply extracted speech features, facial expression features and text emotion features;

[0057] S32 uses the cross-modal sentiment association network (CMAN) to fuse the above multimodal features to capture the sentiment association between different modalities. The fusion formula is expressed as:

[0058] F multimodal =CMAN(F speech , F face , F text )

[0059] Among them, F multimodal represents the fused multimodal features, F speec h、F face and F text Respectively represent the input speech features, facial expression features and text emotion features;

[0060] S33 introduces a multimodal attention mechanism in the cross-modal sentiment association network to dynamically weight the fused multimodal features to highlight the contribution of the key modalities. The weighted formula is expressed as:

[0061]

[0062] Among them, F attention represents the multimodal features after attention weighting, w i represents the attention weight of the i-th modality, F multimodali Represents the fusion feature of the i-th modality, with a weight value of w i Based on what is learned during model training;

[0063] S34 uses a triplet loss function to optimize the model to make samples with similar emotions closer in the feature space and samples with different emotions farther away, thereby improving the model's ability to distinguish complex emotions. The triplet loss function formula is expressed as:

[0064]

[0065] Among them, L triplet represents the triplet loss, a, p and n represent the anchor sample, positive sample (sample with similar emotion to the anchor) and negative sample (sample with different emotion from the anchor), d(a,p) and d(a,n) represent the distance between the anchor and the positive sample, and between the anchor and the negative sample, respectively, and m represents the preset threshold;

[0066] S35 optimizes the model parameters by minimizing the triplet loss function to obtain a multimodal emotion association model with emotion differentiation capability.

[0067] S4 sentiment classification and intensity assessment steps:

[0068] S41 receives and processes the comprehensive emotion representation vector obtained through cross-modal emotion association learning;

[0069] S42 constructs a sentiment classifier, which combines support vector machine (SVM) and neural network to classify the comprehensive sentiment representation vector and output specific sentiment categories. Its classification formula is expressed as:

[0070] Emotion class =Classifier(F emotion )

[0071] Among them, Emotionclass represents the output emotion category, Classifier represents the constructed emotion classifier, and F emotion A comprehensive sentiment representation vector representing the input;

[0072] S43 introduces the emotion intensity evaluation strategy in the emotion classifier, calculates the emotion intensity according to the comprehensive emotion representation vector, and realizes the refined processing of emotion recognition. The emotion intensity calculation formula is expressed as:

[0073] Emotion intensity =f(F emotion ) Among them, Emotion intensity represents the calculated emotion intensity, and f represents the emotion intensity evaluation function, which is obtained through learning during the model training process and is used to map the comprehensive emotion representation vector to a specific emotion intensity value;

[0074] S44 simultaneously outputs the emotion category and its corresponding emotion intensity level, providing users with comprehensive emotion recognition results.

[0075] S5 real-time feedback and adaptive optimization steps:

[0076] During the emotion recognition process, S51 receives real-time feedback from users on the recognition results, including evaluations on the correctness of the emotion category and the accuracy of the emotion intensity.

[0077] S52 determines whether the current emotion recognition model has misjudgment or deviation based on user feedback, and determines the model part that needs to be adjusted and optimized;

[0078] S53 uses an online learning algorithm: Incremental Support Vector Machine (Incremental SVM), which incorporates newly collected sample data with user feedback labels into the model training set;

[0079] S54 uses the newly added sample data and its labels to incrementally update the emotion recognition model, and adjusts the model parameters to reflect user feedback, so that the model can gradually adapt to the emotional expression habits of different users;

[0080] S55 repeats steps S51 to S54, continuously collects user feedback and updates the model, realizes continuous learning and optimization of the emotion recognition model, and improves the accuracy and user satisfaction of long-term use.

[0081] Example 2

[0082] The software is designed using the method of Example 1, and the software is installed in the intelligent robot for operation. The main modules of the intelligent robot (such as Figure 2 As shown) as follows:

[0083] 1) Multimodal data acquisition module:

[0084] Equipped with a high-definition camera to capture a video stream of the user's facial expressions.

[0085] Integrated high-sensitivity microphone array to collect user's voice signals.

[0086] Through natural language processing technology, the text content entered by the user can be obtained and analyzed in real time.

[0087] 2) Data preprocessing and fusion module:

[0088] The collected multimodal data are preprocessed using the weighted fusion method to ensure data quality.

[0089] The fusion weight is dynamically adjusted according to the quality score and importance adjustment coefficient of each modal signal.

[0090] 3) Emotional feature extraction module:

[0091] A combined model of deep convolutional neural network (DCNN) and long short-term memory network (LSTM) is used to deeply extract emotional features.

[0092] The attention mechanism is introduced to automatically focus on key emotional information and improve the accuracy of feature extraction.

[0093] 4) Cross-modal sentiment association learning module:

[0094] A cross-modal sentiment association network (CMAN) is used to fuse feature information from different modalities.

[0095] The model is optimized through the triplet loss function to enhance the ability to distinguish similar emotions.

[0096] 5) Sentiment classification and intensity assessment module:

[0097] Combine support vector machine (SVM) with neural network to build an efficient sentiment classifier.

[0098] Introduce an emotion intensity assessment strategy to provide users with refined emotion feedback.

[0099] 6) Real-time feedback and adaptive optimization module:

[0100] Implement a closed-loop feedback mechanism to adjust model parameters based on real-time user feedback.

[0101] Adopting online learning algorithms, such as incremental SVM, enables the robot to continuously learn and evolve.

[0102] 7) Interaction and response module:

[0103] Adjust the robot's voice response, facial expressions, and action expressions based on the emotion recognition results.

[0104] Provides natural language generation capabilities to ensure that communication with users is natural and smooth.

[0105] 8) Central control and coordination module:

[0106] Responsible for the coordination and communication between modules to ensure the stable operation of the overall system.

[0107] Integrate advanced decision-making algorithms to address challenges in complex interaction scenarios.

[0108] In summary, our invention firstly realizes the comprehensive collection and accurate preprocessing of speech, facial expressions and text content through the multimodal data fusion preprocessing step, significantly improves the quality and reliability of the data, and provides a solid foundation for subsequent emotion recognition. Secondly, the deep extraction step of emotional features uses advanced deep convolutional neural networks and long short-term memory networks, combined with the attention mechanism, to accurately capture key emotional information and effectively improve the accuracy of emotion recognition. Furthermore, the cross-modal emotion association learning step optimizes the cross-modal emotion association network and the triple loss function, so that the model can better distinguish complex emotions, further improving the precision and robustness of emotion recognition. Finally, the implementation of real-time feedback and adaptive optimization steps enables the model to continuously learn and adapt to the user's emotional expression habits, which not only improves the accuracy of long-term use, but also significantly enhances user satisfaction, providing a more natural and smooth emotional communication experience for the voice interaction of intelligent robots.

[0109] The above are only a few preferred embodiments of the present invention, which are described in a more specific and detailed manner, but they cannot be understood as limiting the scope of the present invention. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for developing multimodal emotion recognition software for intelligent robot voice interaction, characterized in that: The following steps are involved: S1. Multimodal data fusion preprocessing steps: The weighted fusion method is used to synchronously collect and preprocess voice signals, facial expression video streams and text content; the weight calculation is based on the quality score of each modal signal and the preset importance adjustment coefficient, which is determined by exponential function and normalization processing. This step enables modal data of different quality and importance to be fused in proportion, providing a reliable basis for subsequent processing; S2 emotional feature deep extraction steps: Design a combined model that combines the deep convolutional neural network (DCNN) and the long short-term memory network (LSTM) to extract high-level features of speech acoustic features, facial expression geometric features, and text emotional vocabulary. Introduce an attention mechanism to dynamically adjust the importance of different time steps or feature points to enhance the model's ability to capture key information. S3 cross-modal sentiment association learning steps: Using the cross-modal emotion association network CMAN, the multimodal attention mechanism is used to fuse the extracted features, calculate the comprehensive emotion representation vector, and use the triple loss function to optimize the model, so that similar emotion samples are close in the feature space and different emotion samples are far away, thus improving the model's ability to distinguish complex emotions; S4 sentiment classification and intensity assessment steps: Construct a sentiment classifier, combine support vector machine (SVM) and neural network, output sentiment categories and their intensity levels at the same time, introduce sentiment intensity evaluation strategy, calculate sentiment intensity based on comprehensive sentiment representation vector, and realize refined processing of sentiment recognition; S5 real-time feedback and adaptive optimization steps: Implement a closed-loop feedback mechanism, adjust model parameters based on real-time user feedback, use online learning algorithms and incremental SVM to instantly update model parameters based on new samples and their labels, so that the model can continuously learn and adapt to the emotional expression habits of different users, thereby improving long-term accuracy and user satisfaction.

2. The method for developing multimodal emotion recognition software for intelligent robot voice interaction according to claim 1, characterized in that: The step S1 comprises the following specific steps: S11 Synchronous Data Collection: Start audio, video and text acquisition devices in real time and synchronously to obtain the user's voice signal, facial expression video stream and text content; S12 modal quality assessment: Perform a quality assessment on each acquired modal data and determine its quality score, which is determined based on the signal clarity, completeness, signal-to-noise ratio, and user feedback; S13 weight calculation: The weight of each modality in the fusion process is calculated based on the quality score of each modality data and the preset importance adjustment coefficient; the weight calculation takes into account the quality of each modality data and its importance in practical applications, ensuring that high-quality and important modalities occupy a larger proportion in the fusion; S14 multimodal data fusion: The calculated weights are used to perform weighted fusion on the synchronously collected modal data. During the fusion process, the contribution of each modal data in the fusion result is adjusted according to the weight, so that modal data of different quality and importance can be fused in proportion. S15 data preprocessing: The fused multimodal data is preprocessed, including denoising, standardization, and normalization, to achieve data consistency and reliability; the preprocessed data provides high-quality data input for subsequent deep extraction and recognition of emotional features.

3. The method for developing multimodal emotion recognition software for intelligent robot voice interaction according to claim 1, characterized in that: The step S2 comprises the following specific steps: S21 receives and processes the pre-processed voice signal, facial expression video stream and text content; S22 uses deep convolutional neural networks to deeply extract the acoustic features of speech signals, and obtains high-level feature representations of speech signals through multi-layer convolution and pooling operations. Its feature extraction formula is expressed as: F speech =DCNN(Speech signal ) Among them, F speech Represents the extracted speech features, Speech signal represents an input speech signal; S23 uses a long short-term memory network to deeply extract the geometric features of facial expression video streams. By capturing the long-term dependencies in the sequence data, it obtains a high-level feature representation of facial expressions. Its feature extraction formula is expressed as: F face =LSTM(Face stream ) Among them, F face Represents the extracted facial expression features, Face stream represents the input facial expression video stream; S24 extracts sentiment words from the text content, converts the text into a vector representation using word embedding technology, and then extracts high-level features through a neural network to obtain a high-level feature representation of the text sentiment. The feature extraction formula is expressed as: F text =NN(Embedding(Text content )) Among them, F text Represents the extracted text features, Text content Represents the text content of the input. Embedding means word embedding operation, NN means neural network; S25 introduces the attention mechanism to dynamically weight the extracted speech, facial expression and text features to highlight key features and improve the accuracy of emotion recognition. The weighted formula is expressed as: F fused =α·F speech +β·F face +γ·F text Among them, F fused represents the fused features, α, β, and γ represent the attention weights of speech, facial expression, and text features, respectively, satisfying α+β+γ=1, and the weight values ​​are obtained according to the learning during the model training process.

4. The method for developing multimodal emotion recognition software for intelligent robot voice interaction according to claim 1, characterized in that: The step S3 comprises the following specific steps: S31 receives and processes the deeply extracted speech features, facial expression features and text emotion features; S32 uses a cross-modal sentiment association network to fuse the above multimodal features to capture the sentiment association between different modalities. Its fusion formula is expressed as: F multimodal =CMAN(F speech ,F face ,F text ) Among them, F multimodal represents the fused multimodal features, F speech 、F face and F text Respectively represent the input speech features, facial expression features and text emotion features; S33 introduces a multimodal attention mechanism in the cross-modal sentiment association network to dynamically weight the fused multimodal features to highlight the contribution of the key modalities. The weighted formula is expressed as: Among them, F attention represents the multimodal features after attention weighting, w i represents the attention weight of the i-th modality, F multimodali Represents the fusion feature of the i-th modality, with a weight value of w i Based on what is learned during model training; S34 uses a triplet loss function to optimize the model to make samples with similar emotions closer in the feature space and samples with different emotions farther away, thereby improving the model's ability to distinguish complex emotions. The triplet loss function formula is expressed as: Among them, L triplet represents the triplet loss, a, p and n represent the anchor sample, positive sample and negative sample respectively, d(a,p) and d(a,n) represent the distance between the anchor and the positive sample, and the anchor and the negative sample respectively, and m represents the preset threshold; S35 optimizes the model parameters by minimizing the triplet loss function to obtain a multimodal emotion association model with emotion differentiation capability.

5. The method for developing multimodal emotion recognition software for intelligent robot voice interaction according to claim 1, characterized in that: The step S4 comprises the following specific steps: S41 receives and processes the comprehensive emotion representation vector obtained through cross-modal emotion association learning; S42 constructs a sentiment classifier, which combines support vector machines and neural networks to classify comprehensive sentiment representation vectors and output specific sentiment categories. Its classification formula is expressed as: Emotion class =Classifier(F emotion ) Among them, Emotionclass represents the output emotion category, Classifier represents the constructed emotion classifier, and F emotion A comprehensive sentiment representation vector representing the input; S43 introduces the emotion intensity evaluation strategy in the emotion classifier, calculates the emotion intensity according to the comprehensive emotion representation vector, and realizes the refined processing of emotion recognition. The emotion intensity calculation formula is expressed as: Emotion intensity =f(F emotion ) Among them, Emotion intensity represents the calculated emotion intensity, and f represents the emotion intensity evaluation function, which is obtained through learning during the model training process and is used to map the comprehensive emotion representation vector to a specific emotion intensity value; S44 simultaneously outputs the emotion category and its corresponding emotion intensity level, providing users with comprehensive emotion recognition results.

6. The method for developing multimodal emotion recognition software for intelligent robot voice interaction according to claim 1, characterized in that: The step S5 comprises the following specific steps: During the emotion recognition process, S51 receives real-time feedback from users on the recognition results, including evaluations on the correctness of the emotion category and the accuracy of the emotion intensity. S52 determines whether the current emotion recognition model has misjudgment or deviation based on user feedback, and determines the model part that needs to be adjusted and optimized; S53 uses an online learning algorithm: incremental support vector machine, which incorporates newly collected sample data with user feedback labels into the model training set; S54 uses the newly added sample data and its labels to incrementally update the emotion recognition model, and adjusts the model parameters to reflect user feedback, so that the model can gradually adapt to the emotional expression habits of different users; S55 repeats steps S51 to S54, continuously collects user feedback and updates the model, realizes continuous learning and optimization of the emotion recognition model, and improves the accuracy and user satisfaction of long-term use.

7. A computer storage device, characterized in that: A method for developing multimodal emotion recognition software for intelligent robot voice interaction as described in any one of claims 1-6.

8. A data processing server, characterized in that: Comprising the computer storage device of claim 7.

9. An intelligent robot, characterized in that: Capable of executing software developed by the development method according to any one of claims 1 to 6.

Images