GRU and DTW combined dance robot action control method and system

Abstract

The invention provides a dance robot action control method and system combining GRU and DTW in the technical field of robot control, and the method comprises the steps: S1, creating a dance action arrangement model, and setting a loss function of the dance action arrangement model; s2, acquiring a large amount of historical music data to construct a data set; s3, dividing the data set into a training set, a verification set and a test set, training the dance movement arrangement model through the training set and the loss function, and verifying and testing the trained dance movement arrangement model in sequence through the verification set and the test set; s4, compressing the dance movement arrangement model passing the test, and deploying the compressed dance movement arrangement model to the dance robot; and S5, the dance robot arranges dance actions based on the deployed dance action arrangement model. The method has the advantages that the flexibility, the matching degree and the fluency of motion control of the dance robot are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of robot control technology, and in particular to a dancing robot motion control method and system combining GRU and DTW. Background Art

[0002] With the deep integration of artificial intelligence and robotics, dancing robots, a prime example of interdisciplinary integrated innovation, have gradually moved from laboratory research to commercial applications. Integrating technologies from multiple fields, including mechanical transmission, motion control, pattern recognition, and deep learning, dancing robots demonstrate tremendous potential in entertainment performances, education, and medical rehabilitation.

[0003] In terms of movement generation mechanisms, current mainstream dancing robots mostly use a pre-programmed movement library combined with a finite state machine control architecture, but this has the following limitations: First, the capacity of the fixed movement library is negatively correlated with the flexibility of the choreography. When encountering complex and changing musical rhythms (such as jazz improvisations or electronic music breakbeats), it is difficult to generate innovative movement combinations that match the beat characteristics in real time. Second, traditional movement matching algorithms often use rule-based beat detection methods, such as using short-time Fourier transforms (STFTs) to extract music spectrum energy features and determine beat points through peak detection. Such methods are suitable for pop music with regular rhythms, but when processing rubato (elastic tempo) performances or polyphonic music, their rhythm tracking errors can reach ±150ms, directly resulting in a significant phase difference between the robot arm joint motion and the musical beat. A deeper technical challenge lies in the cross-modal tempo alignment of music and movement. Traditionally, the tempo alignment of music features and dance movements mainly relies on manual adjustment, which is not only inefficient but also cannot achieve precise synchronization of dance movements with the music beat, affecting the fluency and artistry of the dancing robot performance.

[0004] Therefore, how to provide a dancing robot motion control method and system that combines GRU and DTW to improve the flexibility, matching and smoothness of the dancing robot motion control has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a dancing robot motion control method and system combining GRU and DTW, so as to improve the flexibility, matching and fluency of the dancing robot motion control.

[0006] In a first aspect, the present invention provides a dancing robot motion control method combining GRU and DTW, comprising the following steps:

[0007] Step S1: creating a dance movement choreography model based on the dance movement generation module and the timing alignment module connected in sequence, and setting a loss function of the dance movement choreography model;

[0008] The dance movement generation module is constructed based on a music feature extraction unit, a posture relationship generation unit, and a movement generation unit; the music feature extraction unit is used to extract music features from input music data; the posture relationship generation unit is used to convert music features into posture relationship features through GRU and cross attention mechanism; the movement generation unit is used to generate dance movements based on the posture relationship features;

[0009] The timing alignment module is constructed based on the action rhythm primary alignment unit and the action rhythm secondary alignment unit; the action rhythm primary alignment unit is used to perform preliminary alignment of the generated dance movements and the extracted music features through Transformer; the action rhythm secondary alignment unit is used to perform secondary alignment of the preliminary aligned dance movements and music features through DTW to generate a dance movement choreography result containing dance movements and movement times;

[0010] Step S2: Acquire a large amount of historical music data, pre-process and annotate each of the historical music data to construct a data set;

[0011] Step S3, dividing the data set into a training set, a validation set, and a test set, training the dance choreography model using the training set and the loss function, and verifying and testing the trained dance choreography model using the validation set and the test set in turn;

[0012] Step S4: compressing the dance movement choreography model after passing the test, and deploying the compressed dance movement choreography model to the dancing robot;

[0013] Step S5: The dancing robot choreographs dance movements based on the deployed dance movement choreography model.

[0014] Furthermore, the step S2 is specifically as follows:

[0015] A large amount of historical music data is obtained, and each of the historical music data is preprocessed including at least format conversion, cropping, segmentation, noise reduction and music feature extraction, and each of the preprocessed historical music data is labeled with music features, dance movements and movement timing. A data set is constructed based on the labeled historical music data, and the sample size of the data set is expanded through a generative adversarial network.

[0016] Furthermore, the step S3 is specifically as follows:

[0017] Dividing the dataset into a training set, a validation set, and a test set based on a ratio of 8:1:1, training a dance choreography model using the training set, and continuously optimizing hyperparameters including at least a learning rate, a learning decay rate, a random dropout rate, a batch size, a hidden layer size, and a regularization parameter during the training process until a loss value of the loss function is less than a preset loss threshold or a preset early stopping condition is satisfied;

[0018] The trained dance choreography model is verified using the verification set to determine whether the movement matching degree is greater than a preset matching degree threshold. If not, the verification fails and the training set is expanded to continue training. If so, the verification passes, and:

[0019] The dance choreography model that has passed the verification is tested using the test set to determine whether the confidence level is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training. If so, the test passes and the process ends.

[0020] Furthermore, the step S4 is specifically as follows:

[0021] The dance movement choreography model that passes the test is compressed using knowledge distillation technology, and the compressed dance movement choreography model is deployed to the dance robot using containerization technology.

[0022] Furthermore, the step S5 is specifically as follows:

[0023] The dancing robot obtains input real-time music data through a communication module or a microphone, pre-processes the real-time music data, and then inputs the data into a deployed dance movement choreography model to obtain a dance movement choreography result corresponding to the real-time music data, and executes the dance movement choreography result to perform a dance performance.

[0024] In a second aspect, the present invention provides a dancing robot motion control system combining GRU and DTW, comprising the following modules:

[0025] a dance movement choreography model creation module, configured to create a dance movement choreography model based on the sequentially connected dance movement generation modules and the timing alignment modules, and set a loss function for the dance movement choreography model;

[0026] The dance movement generation module is constructed based on a music feature extraction unit, a posture relationship generation unit, and a movement generation unit; the music feature extraction unit is used to extract music features from input music data; the posture relationship generation unit is used to convert music features into posture relationship features through GRU and cross attention mechanism; the movement generation unit is used to generate dance movements based on the posture relationship features;

[0027] The timing alignment module is constructed based on the action rhythm primary alignment unit and the action rhythm secondary alignment unit; the action rhythm primary alignment unit is used to perform preliminary alignment of the generated dance movements and the extracted music features through Transformer; the action rhythm secondary alignment unit is used to perform secondary alignment of the preliminary aligned dance movements and music features through DTW to generate a dance movement choreography result containing dance movements and movement times;

[0028] A data set construction module is used to obtain a large amount of historical music data, pre-process and annotate each of the historical music data to construct a data set;

[0029] A dance choreography model training module is used to divide the data set into a training set, a validation set, and a test set, train the dance choreography model using the training set and a loss function, and verify and test the trained dance choreography model using the validation set and the test set in turn;

[0030] A dance movement choreography model deployment module is used to compress the dance movement choreography model after passing the test and deploy the compressed dance movement choreography model to the dancing robot;

[0031] The dancing robot motion control module is used for the dancing robot to choreograph dance movements based on the deployed dance movement choreography model.

[0032] Furthermore, the dataset construction module is specifically used to:

[0033] A large amount of historical music data is obtained, and each of the historical music data is preprocessed including at least format conversion, cropping, segmentation, noise reduction and music feature extraction, and each of the preprocessed historical music data is labeled with music features, dance movements and movement timing. A data set is constructed based on the labeled historical music data, and the sample size of the data set is expanded through a generative adversarial network.

[0034] Furthermore, the dance choreography model training module is specifically used to:

[0035] Dividing the dataset into a training set, a validation set, and a test set based on a ratio of 8:1:1, training a dance choreography model using the training set, and continuously optimizing hyperparameters including at least a learning rate, a learning decay rate, a random dropout rate, a batch size, a hidden layer size, and a regularization parameter during the training process until a loss value of the loss function is less than a preset loss threshold or a preset early stopping condition is satisfied;

[0036] The trained dance choreography model is verified using the verification set to determine whether the movement matching degree is greater than a preset matching degree threshold. If not, the verification fails and the training set is expanded to continue training. If so, the verification passes, and:

[0037] The dance choreography model that has passed the verification is tested using the test set to determine whether the confidence level is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training. If so, the test passes and the process ends.

[0038] Furthermore, the dance choreography model deployment module is specifically used to:

[0039] The dance movement choreography model that passes the test is compressed using knowledge distillation technology, and the compressed dance movement choreography model is deployed to the dance robot using containerization technology.

[0040] Furthermore, the dancing robot motion control module is specifically used to:

[0041] The dancing robot obtains input real-time music data through a communication module or a microphone, pre-processes the real-time music data, and then inputs the data into a deployed dance movement choreography model to obtain a dance movement choreography result corresponding to the real-time music data, and executes the dance movement choreography result to perform a dance performance.

[0042] The advantages of the present invention are:

[0043] 1. Create a dance movement choreography model through the dance movement generation module and the timing alignment module connected in sequence, and set the loss function of the dance movement choreography model; the dance movement generation module is constructed based on the music feature extraction unit, the posture relationship generation unit and the action generation unit; the music feature extraction unit is used to extract music features from the input music data; the posture relationship generation unit is used to convert music features into posture relationship features through GRU and cross-attention mechanism; the action generation unit is used to generate dance movements based on posture relationship features; the timing alignment module is constructed based on the action rhythm first-level alignment unit and the action rhythm second-level alignment unit; the action rhythm first-level alignment unit is used to perform preliminary alignment of the generated dance movements and the extracted music features through Transformer; the action rhythm second-level alignment unit is used to perform secondary alignment of the preliminarily aligned dance movements and music features through DTW to generate dance movement choreography results with dance movements and action times; then obtain a large amount of historical music data, and pre-process each historical music data After processing and labeling, a data set is constructed, and the data set is divided into a training set, a validation set and a test set. The dance movement choreography model is trained by the training set and the loss function, and the trained dance movement choreography model is verified and tested in turn by the validation set and the test set. The dance movement choreography model that passes the test is compressed, and the compressed dance movement choreography model is deployed to the dance robot. Finally, the dance robot choreographs dance movements based on the deployed dance movement choreography model; that is, the dance robot's movements are controlled by the pre-trained dance movement choreography model, and the dance movement choreography model generates dance movements through a dance movement generation module constructed by a music feature extraction unit, a posture relationship generation unit and an action generation unit. Compared with the traditional fixed action library, the dance movement choreography model can more flexibly match complex and changeable music rhythms, and automatically aligns the dance movements and music features with the timing alignment module to ensure the accurate synchronization of dance movements and music beats, ultimately greatly improving the flexibility, matching and fluency of the dance robot's movement control.

[0044] 2. Through the GRU (Gated Recurrent Unit) and cross-attention mechanism, music features are dynamically converted into posture relationship features, enhancing the semantic relevance between music and dance movements; GRU is good at capturing temporal dependencies, and the cross-attention mechanism can accurately align the key features of different modalities (music and movement), improving the logic and artistic expression of movement generation.

[0045] 3. Through the process design of music feature extraction → posture relationship generation → dance movement generation, a layer-by-layer mapping from abstract music features to specific dance movements is achieved, ensuring the consistency and diversity of movement generation.

[0046] 4. In the first stage, global timing alignment is performed through Transformer, and its self-attention mechanism is used to capture long-distance dependencies. In the second stage, local fine-grained alignment is performed through DTW (Dynamic Time Warping) to resolve slight deviations between the music rhythm and the movement rhythm. The two stages complement each other, taking into account both efficiency and accuracy. The elastic time warping characteristics of DTW can effectively cope with fluctuations in music beats or delays in the execution of dance robots, ensuring real-time synchronization of dance movements and music data.

[0047] 5. By expanding the dataset through generative adversarial networks (GANs), the scarcity of dance movement annotation data is alleviated. At the same time, diverse samples are generated through adversarial training, which enhances the model's adaptability to complex music styles and movement changes.

[0048] 6. By automatically adjusting hyperparameters such as learning rate and regularization parameters during training, and combining the early stopping mechanism to prevent overfitting, the model convergence speed and generalization performance are significantly improved.

[0049] 7. Through knowledge distillation technology, large dance movement choreography models are compressed into lightweight models, which reduces computing resource consumption while retaining performance, making them suitable for deployment on dancing robots with limited computing power. Containerization technology is used to encapsulate the model, enabling rapid cross-platform migration and version management, improving deployment efficiency and maintainability in practical applications.

[0050] 8. From music input (through a microphone or communication module) to the generation of dance movement choreography results, the entire process supports real-time processing, meeting the dance robot's need for instant response; by setting the dance movement choreography results to include not only dance movements but also the movement time of each dance movement, it ensures that the dance robot's execution is highly matched with the music beat.

[0051] 9. Through phased verification of training sets, validation sets, and test sets, combined with quantitative indicators of action matching and confidence, a closed-loop feedback mechanism is formed to continuously optimize model performance; by automatically expanding training data and retraining when verification or testing fails, manual intervention is reduced and model iteration efficiency is improved.

[0052] 10. By organically integrating advanced technologies such as GRU (time series modeling), Transformer (global alignment), DTW (local alignment), GAN (data enhancement), and knowledge distillation (model compression), a synergistic effect is formed to overcome the limitations of a single technology.

[0053] 11. By supporting two input methods: communication module (such as preset music files) and pickup (real-time audio stream), it adapts to the needs of different scenarios (such as stage performances and impromptu interactions) and expands the application boundaries.

[0054] 12. By decoupling modules such as music feature extraction, posture generation, and timing alignment, it is easy to upgrade or replace them individually (such as replacing them with more efficient feature extraction algorithms), reducing the system iteration cost.

[0055] 13. Global coarse-grained alignment is processed through Transformer (first-level alignment), and DTW (second-level alignment) is used to perform fine-grained optimization only on key segments, reducing computational redundancy and lowering real-time computing load.

[0056] 14. The semantic association between music features and dance movements is achieved through the GRU and cross-attention mechanism. The two-stage temporal alignment strategy of Transformer and DTW is combined to take into account global rhythm matching and local elastic adjustment, significantly improving the synchronization accuracy between movement and music. GAN is used to enhance data diversity and knowledge distillation compression model to reduce computing power requirements while ensuring cross-style adaptability and real-time response. Modular design and containerized deployment achieve flexible expansion and efficient operation and maintenance. Through innovative multi-technology integration, the difficult problem of movement generation and elastic matching of music beats in the field of dance robots is solved, with high robustness, low latency and low power consumption, providing a safe and reliable solution for intelligent performance systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0058] Figure 1 This is a flow chart of a dancing robot motion control method combining GRU and DTW in the present invention.

[0059] Figure 2 This is a structural diagram of a dancing robot motion control system combining GRU and DTW in the present invention.

[0060] Figure 3 It is a schematic diagram of the posture relationship feature of the present invention. DETAILED DESCRIPTION

[0061] The technical solution in the embodiments of the present application has the following overall idea: the dance robot's movements are controlled through a pre-trained dance movement choreography model. The dance movement choreography model generates dance movements through a dance movement generation module constructed by a music feature extraction unit, a posture relationship generation unit, and a movement generation unit. Compared with the traditional fixed movement library, it can more flexibly match complex and changeable music rhythms. Combined with the timing alignment module, the dance movements and music features are automatically aligned twice to ensure the precise synchronization of dance movements and music beats, so as to improve the flexibility, matching and smoothness of the dance robot's movement control.

[0062] Please refer to Figures 1 to 3As shown, a preferred embodiment of the present invention's method for controlling the motion of a dancing robot combining GRU and DTW includes the following steps:

[0063] Step S1: creating a dance movement choreography model based on the dance movement generation module and the timing alignment module connected in sequence, and setting a loss function of the dance movement choreography model;

[0064] The dance movement generation module is constructed based on a music feature extraction unit, a posture relationship generation unit, and a movement generation unit; the music feature extraction unit is used to extract music features from input music data; the posture relationship generation unit is used to convert music features into posture relationship features through GRU and cross attention mechanism; the movement generation unit is used to generate dance movements based on the posture relationship features;

[0065] Through the GRU (Gated Recurrent Unit) and cross-attention mechanism, music features are dynamically converted into posture relationship features, enhancing the semantic correlation between music and dance movements; GRU is good at capturing temporal dependencies, and the cross-attention mechanism can accurately align the key features of different modalities (music and movement), improving the logic and artistic expression of movement generation.

[0066] Through the process design of music feature extraction → posture relationship generation → dance movement generation, a layer-by-layer mapping from abstract music features to specific dance movements is achieved, ensuring the consistency and diversity of movement generation.

[0067] The music feature extraction unit is built on a CNN and is used to extract 128-dimensional music features from music data. The posture relationship generation unit uses GRU to process the temporal dependencies of music features. Using a cross-attention mechanism, the music features are used as keys and values, and the GRU hidden state is used as the query. This dynamically focuses on the parts of the music data most relevant to the current posture generation (e.g., strong beats correspond to high attention weights). After a fully connected layer and softmax, the unit outputs a temporally continuous 7-dimensional one-hot encoded posture relationship sequence (posture relationship features). Compared to traditional pure GRUs, the posture type accuracy is increased from 78% to 92%, effectively alleviating the problem of information attenuation in long sequences. Posture relationship features include symmetrical type, asymmetrical type, symmetrical, asymmetrical, canon, static, and other.

[0068] In the motion generation unit, a rule library based on "posture relationship features - dance movements" was established, relying on a database containing 431 key postures. Dance movements were dynamically selected and combined based on the generated posture relationship features: Symmetrical movement requires that the angle difference between all symmetrical joints be ≤5°, generating mirror-image movements such as swinging the left and right arms at the same angle; Canon movement triggers movement of the other hand after a 0.3-0.5 second interval by changing the joints of one hand (e.g., rotating the right hand 45° and then the left hand performing the same movement 0.4 seconds later); and Static movement maintains the previous posture for 1-2 beats, creating a rhythmic pause. The rule library filters candidate dance movements (retaining 3-5 typical dance movements for each category) and adds time parameters (movement time + static time, with an accuracy of 0.05 seconds). The final output is a dance movement sequence containing timing information (movement time), which takes into account the physical motion limitations of the robot's 14 servos and the artistic diversity of dance.

[0069] The rule base is as follows:

[0070]

[0071] The timing alignment module is constructed based on the action rhythm primary alignment unit and the action rhythm secondary alignment unit; the action rhythm primary alignment unit is used to perform preliminary alignment of the generated dance movements and the extracted music features through Transformer; the action rhythm secondary alignment unit is used to perform secondary alignment of the preliminary aligned dance movements and music features through DTW to generate a dance movement choreography result containing dance movements and movement times;

[0072] In the first stage, Transformer is used to perform global timing alignment, and its self-attention mechanism is used to capture long-distance dependencies; in the second stage, DTW (Dynamic Time Warping) is used to perform local fine-grained alignment to resolve slight deviations between the music rhythm and the movement rhythm; the two stages complement each other, taking into account both efficiency and accuracy; the elastic time warping characteristics of DTW can effectively cope with fluctuations in music beats or delays in the execution of dance robots, ensuring real-time synchronization between dance movements and music data.

[0073] Global coarse-grained alignment is processed through Transformer (first-level alignment), and DTW (second-level alignment) only performs fine-grained optimization on key segments, reducing computational redundancy and lowering real-time computing load.

[0074] By decoupling modules such as music feature extraction, posture generation, and timing alignment, they can be easily upgraded or replaced individually (such as with more efficient feature extraction algorithms), reducing system iteration costs.

[0075] In the initial alignment stage, a "two-stream cross-modal Transformer architecture" is used to input 128-dimensional music features and dance movements into the encoder and decoder, respectively. Six layers of multi-head self-attention (8 heads) and positional encoding are used to capture the long-range dependencies between music features and dance movements. The decoder uses the dance movement as the query and the music context vector as the key-value. Through cross-attention, it calculates the probability distribution of the musical beat corresponding to each dance movement (for example, the movement density in the chorus is positively correlated with the music speed). The fully connected layer outputs a preliminary aligned movement time series with a time accuracy of 0.1 seconds and an error of ≤150ms. This improves parallel computing efficiency by three times compared to LSTM, effectively handling the global correlation of complex rhythmic patterns such as beat changes and syncopation.

[0076] In response to the local time deviation in the preliminary alignment, the dynamic time warping (DTW) algorithm is introduced for fine adjustment. First, the action sequence composed of 14-dimensional servo angles and the music sequence containing beat position and intensity are extracted, and a weighted Euclidean distance matrix is constructed (core joint weight 1.2, end joint weight 0.8, to strengthen the influence of strong beats). The minimum cumulative distance path is searched through dynamic programming, allowing nonlinear time scaling (such as action acceleration / deceleration). According to the optimal mapping relationship, the action execution time is scaled proportionally (accuracy 0.02 seconds), and finally, precise synchronization with a beat alignment error of ≤89ms is achieved. At the same time, the servo speed limit (such as a maximum speed of 180° / s) is incorporated to ensure the executable nature of the action. This "coarse alignment-fine adjustment" scheme combines the global structure matching of Transformer with the local deviation calibration of DTW, breaking through the accuracy bottleneck of a single algorithm and providing a robust timing alignment solution for dance performances with complex music structures.

[0077] Step S2: Acquire a large amount of historical music data, pre-process and annotate each of the historical music data to construct a data set;

[0078] Step S3, dividing the data set into a training set, a validation set, and a test set, training the dance choreography model using the training set and the loss function, and verifying and testing the trained dance choreography model using the validation set and the test set in turn;

[0079] Step S4: compressing the dance movement choreography model after passing the test, and deploying the compressed dance movement choreography model to the dancing robot;

[0080] Step S5: The dancing robot choreographs dance movements based on the deployed dance movement choreography model.

[0081] Through GRU and cross-attention mechanism, semantic association between music features and dance movements is realized. Combined with the two-stage timing alignment strategy of Transformer and DTW, global rhythm matching and local elastic adjustment are taken into account, which significantly improves the synchronization accuracy between movement and music. GAN is used to enhance data diversity and knowledge distillation compression model, while ensuring cross-style adaptability and real-time response while reducing computing power requirements, and flexible expansion and efficient operation and maintenance are achieved through modular design and containerized deployment. Through innovative multi-technology integration, the difficult problem of movement generation and elastic matching of music beats in the field of dance robots is solved, with high robustness, low latency and low power consumption, providing a safe and reliable solution for intelligent performance systems.

[0082] The step S2 is specifically as follows:

[0083] A large amount of historical music data is obtained, and each of the historical music data is preprocessed including at least format conversion, cropping, segmentation, noise reduction and music feature extraction, and each of the preprocessed historical music data is labeled with music features, dance movements and movement timing. A data set is constructed based on the labeled historical music data, and the sample size of the data set is expanded through a generative adversarial network.

[0084] By expanding the dataset through generative adversarial networks (GANs), the scarcity of dance movement annotation data is alleviated. At the same time, diverse samples are generated through adversarial training, which enhances the model's adaptability to complex music styles and movement changes.

[0085] The step S3 is specifically as follows:

[0086] Dividing the dataset into a training set, a validation set, and a test set based on a ratio of 8:1:1, training a dance choreography model using the training set, and continuously optimizing hyperparameters including at least a learning rate, a learning decay rate, a random dropout rate, a batch size, a hidden layer size, and a regularization parameter during the training process until a loss value of the loss function is less than a preset loss threshold or a preset early stopping condition is satisfied;

[0087] The trained dance choreography model is verified using the verification set to determine whether the movement matching degree is greater than a preset matching degree threshold. If not, the verification fails and the training set is expanded to continue training. If so, the verification passes, and:

[0088] The dance choreography model that has passed the verification is tested using the test set to determine whether the confidence level is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training. If so, the test passes and the process ends.

[0089] By automatically adjusting hyperparameters such as learning rate and regularization parameters during training, and combining the early stopping mechanism to prevent overfitting, the model convergence speed and generalization performance are significantly improved.

[0090] Through phased verification of training sets, validation sets, and test sets, combined with quantitative indicators of action matching and confidence, a closed-loop feedback mechanism is formed to continuously optimize model performance. By automatically expanding the training data and retraining when verification or testing fails, manual intervention is reduced and model iteration efficiency is improved.

[0091] The step S4 is specifically as follows:

[0092] The dance movement choreography model that passes the test is compressed using knowledge distillation technology, and the compressed dance movement choreography model is deployed to the dance robot using containerization technology.

[0093] Through knowledge distillation technology, large-scale dance movement choreography models are compressed into lightweight models, reducing computing resource consumption while retaining performance, making them suitable for deployment on dancing robots with limited computing power. Containerization technology is used to encapsulate the model to achieve rapid cross-platform migration and version management, improving deployment efficiency and maintainability in practical applications.

[0094] By organically integrating advanced technologies such as GRU (time series modeling), Transformer (global alignment), DTW (local alignment), GAN (data enhancement), and knowledge distillation (model compression), a synergistic effect is formed to overcome the limitations of a single technology.

[0095] The step S5 is specifically as follows:

[0096] The dancing robot obtains input real-time music data through a communication module or a microphone, pre-processes the real-time music data, and then inputs the data into a deployed dance movement choreography model to obtain a dance movement choreography result corresponding to the real-time music data, and executes the dance movement choreography result to perform a dance performance.

[0097] From music input (through a microphone or communication module) to the generation of dance movement choreography results, the entire process supports real-time processing, meeting the dance robot's need for instant response; by setting the dance movement choreography results to include not only dance movements but also the movement time of each dance movement, it ensures that the dance robot's execution is highly matched with the music beat.

[0098] By supporting two input methods: communication module (such as preset music files) and pickup (real-time audio stream), it adapts to the needs of different scenarios (such as stage performances and impromptu interactions) and expands the application boundaries.

[0099] A preferred embodiment of the present invention's dancing robot motion control system combining GRU and DTW includes the following modules:

[0100] a dance movement choreography model creation module, configured to create a dance movement choreography model based on the sequentially connected dance movement generation modules and the timing alignment modules, and set a loss function for the dance movement choreography model;

[0101] The dance movement generation module is constructed based on a music feature extraction unit, a posture relationship generation unit, and a movement generation unit; the music feature extraction unit is used to extract music features from input music data; the posture relationship generation unit is used to convert music features into posture relationship features through GRU and cross attention mechanism; the movement generation unit is used to generate dance movements based on the posture relationship features;

[0102] Through the GRU (Gated Recurrent Unit) and cross-attention mechanism, music features are dynamically converted into posture relationship features, enhancing the semantic correlation between music and dance movements; GRU is good at capturing temporal dependencies, and the cross-attention mechanism can accurately align the key features of different modalities (music and movement), improving the logic and artistic expression of movement generation.

[0103] Through the process design of music feature extraction → posture relationship generation → dance movement generation, a layer-by-layer mapping from abstract music features to specific dance movements is achieved, ensuring the consistency and diversity of movement generation.

[0104] The music feature extraction unit is built on a CNN and is used to extract 128-dimensional music features from music data. The posture relationship generation unit uses GRU to process the temporal dependencies of music features. Using a cross-attention mechanism, the music features are used as keys and values, and the GRU hidden state is used as the query. This dynamically focuses on the parts of the music data most relevant to the current posture generation (e.g., strong beats correspond to high attention weights). After a fully connected layer and softmax, the unit outputs a temporally continuous 7-dimensional one-hot encoded posture relationship sequence (posture relationship features). Compared to traditional pure GRUs, the posture type accuracy is increased from 78% to 92%, effectively alleviating the problem of information attenuation in long sequences. Posture relationship features include symmetrical type, asymmetrical type, symmetrical, asymmetrical, canon, static, and other.

[0105] In the motion generation unit, a rule library based on "posture relationship features - dance movements" was established, relying on a database containing 431 key postures. Dance movements were dynamically selected and combined based on the generated posture relationship features: Symmetrical movement requires that the angle difference between all symmetrical joints be ≤5°, generating mirror-image movements such as swinging the left and right arms at the same angle; Canon movement triggers movement of the other hand after a 0.3-0.5 second interval by changing the joints of one hand (e.g., rotating the right hand 45° and then the left hand performing the same movement 0.4 seconds later); and Static movement maintains the previous posture for 1-2 beats, creating a rhythmic pause. The rule library filters candidate dance movements (retaining 3-5 typical dance movements for each category) and adds time parameters (movement time + static time, with an accuracy of 0.05 seconds). The final output is a dance movement sequence containing timing information (movement time), which takes into account the physical motion limitations of the robot's 14 servos and the artistic diversity of dance.

[0106] The rule base is as follows:

[0107]

[0108] The timing alignment module is constructed based on the action rhythm primary alignment unit and the action rhythm secondary alignment unit; the action rhythm primary alignment unit is used to perform preliminary alignment of the generated dance movements and the extracted music features through Transformer; the action rhythm secondary alignment unit is used to perform secondary alignment of the preliminary aligned dance movements and music features through DTW to generate a dance movement choreography result containing dance movements and movement times;

[0109] In the first stage, Transformer is used to perform global timing alignment, and its self-attention mechanism is used to capture long-distance dependencies; in the second stage, DTW (Dynamic Time Warping) is used to perform local fine-grained alignment to resolve slight deviations between the music rhythm and the movement rhythm; the two stages complement each other, taking into account both efficiency and accuracy; the elastic time warping characteristics of DTW can effectively cope with fluctuations in music beats or delays in the execution of dance robots, ensuring real-time synchronization between dance movements and music data.

[0110] Global coarse-grained alignment is processed through Transformer (first-level alignment), and DTW (second-level alignment) only performs fine-grained optimization on key segments, reducing computational redundancy and lowering real-time computing load.

[0111] By decoupling modules such as music feature extraction, posture generation, and timing alignment, they can be easily upgraded or replaced individually (such as with more efficient feature extraction algorithms), reducing system iteration costs.

[0112] In the initial alignment stage, a "two-stream cross-modal Transformer architecture" is used to input 128-dimensional music features and dance movements into the encoder and decoder, respectively. Six layers of multi-head self-attention (8 heads) and positional encoding are used to capture the long-range dependencies between music features and dance movements. The decoder uses the dance movement as the query and the music context vector as the key-value. Through cross-attention, it calculates the probability distribution of the musical beat corresponding to each dance movement (for example, the movement density in the chorus is positively correlated with the music speed). The fully connected layer outputs a preliminary aligned movement time series with a time accuracy of 0.1 seconds and an error of ≤150ms. This improves parallel computing efficiency by three times compared to LSTM, effectively handling the global correlation of complex rhythmic patterns such as beat changes and syncopation.

[0113] In response to the local time deviation in the preliminary alignment, the dynamic time warping (DTW) algorithm is introduced for fine adjustment. First, the action sequence composed of 14-dimensional servo angles and the music sequence containing beat position and intensity are extracted, and a weighted Euclidean distance matrix is constructed (core joint weight 1.2, end joint weight 0.8, to strengthen the influence of strong beats). The minimum cumulative distance path is searched through dynamic programming, allowing nonlinear time scaling (such as action acceleration / deceleration). According to the optimal mapping relationship, the action execution time is scaled proportionally (accuracy 0.02 seconds), and finally, precise synchronization with a beat alignment error of ≤89ms is achieved. At the same time, the servo speed limit (such as a maximum speed of 180° / s) is incorporated to ensure the executable nature of the action. This "coarse alignment-fine adjustment" scheme combines the global structure matching of Transformer with the local deviation calibration of DTW, breaking through the accuracy bottleneck of a single algorithm and providing a robust timing alignment solution for dance performances with complex music structures.

[0114] A data set construction module is used to obtain a large amount of historical music data, pre-process and annotate each of the historical music data to construct a data set;

[0115] A dance choreography model training module is used to divide the data set into a training set, a validation set, and a test set, train the dance choreography model using the training set and a loss function, and verify and test the trained dance choreography model using the validation set and the test set in turn;

[0116] A dance movement choreography model deployment module is used to compress the dance movement choreography model after passing the test and deploy the compressed dance movement choreography model to the dancing robot;

[0117] The dancing robot motion control module is used for the dancing robot to choreograph dance movements based on the deployed dance movement choreography model.

[0118] Through GRU and cross-attention mechanism, semantic association between music features and dance movements is realized. Combined with the two-stage timing alignment strategy of Transformer and DTW, global rhythm matching and local elastic adjustment are taken into account, which significantly improves the synchronization accuracy between movement and music. GAN is used to enhance data diversity and knowledge distillation compression model, while ensuring cross-style adaptability and real-time response while reducing computing power requirements, and flexible expansion and efficient operation and maintenance are achieved through modular design and containerized deployment. Through innovative multi-technology integration, the difficult problem of movement generation and elastic matching of music beats in the field of dance robots is solved, with high robustness, low latency and low power consumption, providing a safe and reliable solution for intelligent performance systems.

[0119] The dataset construction module is specifically used for:

[0120] A large amount of historical music data is obtained, and each of the historical music data is preprocessed including at least format conversion, cropping, segmentation, noise reduction and music feature extraction, and each of the preprocessed historical music data is labeled with music features, dance movements and movement timing. A data set is constructed based on the labeled historical music data, and the sample size of the data set is expanded through a generative adversarial network.

[0121] By expanding the dataset through generative adversarial networks (GANs), the scarcity of dance movement annotation data is alleviated. At the same time, diverse samples are generated through adversarial training, which enhances the model's adaptability to complex music styles and movement changes.

[0122] The dance choreography model training module is specifically used for:

[0123] Dividing the dataset into a training set, a validation set, and a test set based on a ratio of 8:1:1, training a dance choreography model using the training set, and continuously optimizing hyperparameters including at least a learning rate, a learning decay rate, a random dropout rate, a batch size, a hidden layer size, and a regularization parameter during the training process until a loss value of the loss function is less than a preset loss threshold or a preset early stopping condition is satisfied;

[0124] The trained dance choreography model is verified using the verification set to determine whether the movement matching degree is greater than a preset matching degree threshold. If not, the verification fails and the training set is expanded to continue training. If so, the verification passes, and:

[0125] The dance choreography model that has passed the verification is tested using the test set to determine whether the confidence level is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training. If so, the test passes and the process ends.

[0126] By automatically adjusting hyperparameters such as learning rate and regularization parameters during training, and combining the early stopping mechanism to prevent overfitting, the model convergence speed and generalization performance are significantly improved.

[0127] Through phased verification of training sets, validation sets, and test sets, combined with quantitative indicators of action matching and confidence, a closed-loop feedback mechanism is formed to continuously optimize model performance. By automatically expanding the training data and retraining when verification or testing fails, manual intervention is reduced and model iteration efficiency is improved.

[0128] The dance choreography model deployment module is specifically used to:

[0129] The dance movement choreography model that passes the test is compressed using knowledge distillation technology, and the compressed dance movement choreography model is deployed to the dance robot using containerization technology.

[0130] Through knowledge distillation technology, large-scale dance movement choreography models are compressed into lightweight models, reducing computing resource consumption while retaining performance, making them suitable for deployment on dancing robots with limited computing power. Containerization technology is used to encapsulate the model to achieve rapid cross-platform migration and version management, improving deployment efficiency and maintainability in practical applications.

[0131] By organically integrating advanced technologies such as GRU (time series modeling), Transformer (global alignment), DTW (local alignment), GAN (data enhancement), and knowledge distillation (model compression), a synergistic effect is formed to overcome the limitations of a single technology.

[0132] The dancing robot motion control module is specifically used for:

[0133] The dancing robot obtains input real-time music data through a communication module or a microphone, pre-processes the real-time music data, and then inputs the data into a deployed dance movement choreography model to obtain a dance movement choreography result corresponding to the real-time music data, and executes the dance movement choreography result to perform a dance performance.

[0134] From music input (through a microphone or communication module) to the generation of dance movement choreography results, the entire process supports real-time processing, meeting the dance robot's need for instant response; by setting the dance movement choreography results to include not only dance movements but also the movement time of each dance movement, it ensures that the dance robot's execution is highly matched with the music beat.

[0135] By supporting two input methods: communication module (such as preset music files) and pickup (real-time audio stream), it adapts to the needs of different scenarios (such as stage performances and impromptu interactions) and expands the application boundaries.

[0136] In summary, the advantages of the present invention are:

[0137] 1. Create a dance movement choreography model through the dance movement generation module and the timing alignment module connected in sequence, and set the loss function of the dance movement choreography model; the dance movement generation module is constructed based on the music feature extraction unit, the posture relationship generation unit and the action generation unit; the music feature extraction unit is used to extract music features from the input music data; the posture relationship generation unit is used to convert music features into posture relationship features through GRU and cross-attention mechanism; the action generation unit is used to generate dance movements based on posture relationship features; the timing alignment module is constructed based on the action rhythm first-level alignment unit and the action rhythm second-level alignment unit; the action rhythm first-level alignment unit is used to perform preliminary alignment of the generated dance movements and the extracted music features through Transformer; the action rhythm second-level alignment unit is used to perform secondary alignment of the preliminarily aligned dance movements and music features through DTW to generate dance movement choreography results with dance movements and action times; then obtain a large amount of historical music data, and pre-process each historical music data After processing and labeling, a data set is constructed, and the data set is divided into a training set, a validation set and a test set. The dance movement choreography model is trained by the training set and the loss function, and the trained dance movement choreography model is verified and tested in turn by the validation set and the test set. The dance movement choreography model that passes the test is compressed, and the compressed dance movement choreography model is deployed to the dance robot. Finally, the dance robot choreographs dance movements based on the deployed dance movement choreography model; that is, the dance robot's movements are controlled by the pre-trained dance movement choreography model, and the dance movement choreography model generates dance movements through a dance movement generation module constructed by a music feature extraction unit, a posture relationship generation unit and an action generation unit. Compared with the traditional fixed action library, the dance movement choreography model can more flexibly match complex and changeable music rhythms, and automatically aligns the dance movements and music features with the timing alignment module to ensure the accurate synchronization of dance movements and music beats, ultimately greatly improving the flexibility, matching and fluency of the dance robot's movement control.

[0138] 2. Through the GRU (Gated Recurrent Unit) and cross-attention mechanism, music features are dynamically converted into posture relationship features, enhancing the semantic relevance between music and dance movements; GRU is good at capturing temporal dependencies, and the cross-attention mechanism can accurately align the key features of different modalities (music and movement), improving the logic and artistic expression of movement generation.

[0139] 3. Through the process design of music feature extraction → posture relationship generation → dance movement generation, a layer-by-layer mapping from abstract music features to specific dance movements is achieved, ensuring the consistency and diversity of movement generation.

[0140] 4. In the first stage, global timing alignment is performed through Transformer, and its self-attention mechanism is used to capture long-distance dependencies. In the second stage, local fine-grained alignment is performed through DTW (Dynamic Time Warping) to resolve slight deviations between the music rhythm and the movement rhythm. The two stages complement each other, taking into account both efficiency and accuracy. The elastic time warping characteristics of DTW can effectively cope with fluctuations in music beats or delays in the execution of dance robots, ensuring real-time synchronization of dance movements and music data.

[0141] 5. By expanding the dataset through generative adversarial networks (GANs), the scarcity of dance movement annotation data is alleviated. At the same time, diverse samples are generated through adversarial training, which enhances the model's adaptability to complex music styles and movement changes.

[0142] 6. By automatically adjusting hyperparameters such as learning rate and regularization parameters during training, and combining the early stopping mechanism to prevent overfitting, the model convergence speed and generalization performance are significantly improved.

[0143] 7. Through knowledge distillation technology, large dance movement choreography models are compressed into lightweight models, which reduces computing resource consumption while retaining performance, making them suitable for deployment on dancing robots with limited computing power. Containerization technology is used to encapsulate the model, enabling rapid cross-platform migration and version management, improving deployment efficiency and maintainability in practical applications.

[0144] 8. From music input (through a microphone or communication module) to the generation of dance movement choreography results, the entire process supports real-time processing, meeting the dance robot's need for instant response; by setting the dance movement choreography results to include not only dance movements but also the movement time of each dance movement, it ensures that the dance robot's execution is highly matched with the music beat.

[0145] 9. Through phased verification of training sets, validation sets, and test sets, combined with quantitative indicators of action matching and confidence, a closed-loop feedback mechanism is formed to continuously optimize model performance; by automatically expanding training data and retraining when verification or testing fails, manual intervention is reduced and model iteration efficiency is improved.

[0146] 10. By organically integrating advanced technologies such as GRU (time series modeling), Transformer (global alignment), DTW (local alignment), GAN (data enhancement), and knowledge distillation (model compression), a synergistic effect is formed to overcome the limitations of a single technology.

[0147] 11. By supporting two input methods: communication module (such as preset music files) and pickup (real-time audio stream), it adapts to the needs of different scenarios (such as stage performances and impromptu interactions) and expands the application boundaries.

[0148] 12. By decoupling modules such as music feature extraction, posture generation, and timing alignment, it is easy to upgrade or replace them individually (such as replacing them with more efficient feature extraction algorithms), reducing the system iteration cost.

[0149] 13. Global coarse-grained alignment is processed through Transformer (first-level alignment), and DTW (second-level alignment) is used to perform fine-grained optimization only on key segments, reducing computational redundancy and lowering real-time computing load.

[0150] 14. The semantic association between music features and dance movements is achieved through the GRU and cross-attention mechanism. The two-stage temporal alignment strategy of Transformer and DTW is combined to take into account global rhythm matching and local elastic adjustment, significantly improving the synchronization accuracy between movement and music. GAN is used to enhance data diversity and knowledge distillation compression model to reduce computing power requirements while ensuring cross-style adaptability and real-time response. Modular design and containerized deployment achieve flexible expansion and efficient operation and maintenance. Through innovative multi-technology integration, the difficult problem of movement generation and elastic matching of music beats in the field of dance robots is solved, with high robustness, low latency and low power consumption, providing a safe and reliable solution for intelligent performance systems.

[0151] Although the specific embodiments of the present invention are described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A dancing robot motion control method combining GRU and DTW, characterized by: The steps include: Step S1: creating a dance movement choreography model based on the dance movement generation module and the timing alignment module connected in sequence, and setting a loss function of the dance movement choreography model; The dance movement generation module is constructed based on a music feature extraction unit, a posture relationship generation unit, and a movement generation unit; the music feature extraction unit is used to extract music features from input music data; the posture relationship generation unit is used to convert music features into posture relationship features through GRU and cross attention mechanism; the movement generation unit is used to generate dance movements based on the posture relationship features; The timing alignment module is constructed based on the action rhythm primary alignment unit and the action rhythm secondary alignment unit; the action rhythm primary alignment unit is used to perform preliminary alignment of the generated dance movements and the extracted music features through Transformer; the action rhythm secondary alignment unit is used to perform secondary alignment of the preliminary aligned dance movements and music features through DTW to generate a dance movement choreography result containing dance movements and movement times; Step S2: Acquire a large amount of historical music data, pre-process and annotate each of the historical music data to construct a data set; Step S3, dividing the data set into a training set, a validation set, and a test set, training the dance choreography model using the training set and the loss function, and verifying and testing the trained dance choreography model using the validation set and the test set in turn; Step S4: compressing the dance movement choreography model after passing the test, and deploying the compressed dance movement choreography model to the dancing robot; Step S5: The dancing robot choreographs dance movements based on the deployed dance movement choreography model.

2. The dancing robot motion control method combining GRU and DTW as claimed in claim 1, characterized in that: The step S2 is specifically as follows: A large amount of historical music data is obtained, and each of the historical music data is preprocessed including at least format conversion, cropping, segmentation, noise reduction and music feature extraction, and each of the preprocessed historical music data is labeled with music features, dance movements and movement timing. A data set is constructed based on the labeled historical music data, and the sample size of the data set is expanded through a generative adversarial network.

3. The dancing robot motion control method combining GRU and DTW as claimed in claim 1, characterized in that: The step S3 is specifically as follows: Dividing the dataset into a training set, a validation set, and a test set based on a ratio of 8:1:1, training a dance choreography model using the training set, and continuously optimizing hyperparameters including at least a learning rate, a learning decay rate, a random dropout rate, a batch size, a hidden layer size, and a regularization parameter during the training process until a loss value of the loss function is less than a preset loss threshold or a preset early stopping condition is satisfied; The trained dance choreography model is verified using the verification set to determine whether the movement matching degree is greater than a preset matching degree threshold. If not, the verification fails and the training set is expanded to continue training. If so, the verification passes, and: The dance choreography model that has passed the verification is tested using the test set to determine whether the confidence level is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training. If so, the test passes and the process ends.

4. The dancing robot motion control method combining GRU and DTW as claimed in claim 1, characterized in that: The step S4 is specifically as follows: The dance movement choreography model that passes the test is compressed using knowledge distillation technology, and the compressed dance movement choreography model is deployed to the dance robot using containerization technology.

5. The dancing robot motion control method combining GRU and DTW as claimed in claim 1, characterized in that: The step S5 is specifically as follows: The dancing robot obtains input real-time music data through a communication module or a microphone, pre-processes the real-time music data, and then inputs the data into a deployed dance movement choreography model to obtain a dance movement choreography result corresponding to the real-time music data, and executes the dance movement choreography result to perform a dance performance.

6. A dancing robot motion control system combining GRU and DTW, characterized by: Includes the following modules: a dance movement choreography model creation module, configured to create a dance movement choreography model based on the sequentially connected dance movement generation modules and the timing alignment modules, and set a loss function for the dance movement choreography model; The dance movement generation module is constructed based on a music feature extraction unit, a posture relationship generation unit, and a movement generation unit; the music feature extraction unit is used to extract music features from input music data; the posture relationship generation unit is used to convert music features into posture relationship features through GRU and cross attention mechanism; the movement generation unit is used to generate dance movements based on the posture relationship features; The timing alignment module is constructed based on the action rhythm primary alignment unit and the action rhythm secondary alignment unit; the action rhythm primary alignment unit is used to perform preliminary alignment of the generated dance movements and the extracted music features through Transformer; the action rhythm secondary alignment unit is used to perform secondary alignment of the preliminary aligned dance movements and music features through DTW to generate a dance movement choreography result containing dance movements and movement times; A data set construction module is used to obtain a large amount of historical music data, pre-process and annotate each of the historical music data to construct a data set; A dance choreography model training module is used to divide the data set into a training set, a validation set, and a test set, train the dance choreography model using the training set and a loss function, and verify and test the trained dance choreography model using the validation set and the test set in turn; A dance movement choreography model deployment module is used to compress the dance movement choreography model after passing the test and deploy the compressed dance movement choreography model to the dancing robot; The dancing robot motion control module is used for the dancing robot to choreograph dance movements based on the deployed dance movement choreography model.

7. The dancing robot motion control system combining GRU and DTW as claimed in claim 6, characterized in that: The dataset construction module is specifically used for: A large amount of historical music data is obtained, and each of the historical music data is preprocessed including at least format conversion, cropping, segmentation, noise reduction and music feature extraction, and each of the preprocessed historical music data is labeled with music features, dance movements and movement timing. A data set is constructed based on the labeled historical music data, and the sample size of the data set is expanded through a generative adversarial network.

8. The dancing robot motion control system combining GRU and DTW as claimed in claim 6, characterized in that: The dance choreography model training module is specifically used for: Dividing the dataset into a training set, a validation set, and a test set based on a ratio of 8:1:1, training a dance choreography model using the training set, and continuously optimizing hyperparameters including at least a learning rate, a learning decay rate, a random dropout rate, a batch size, a hidden layer size, and a regularization parameter during the training process until a loss value of the loss function is less than a preset loss threshold or a preset early stopping condition is satisfied; The trained dance choreography model is verified using the verification set to determine whether the movement matching degree is greater than a preset matching degree threshold. If not, the verification fails and the training set is expanded to continue training. If so, the verification passes, and: The dance choreography model that has passed the verification is tested using the test set to determine whether the confidence level is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training. If so, the test passes and the process ends.

9. The dancing robot motion control system combining GRU and DTW as claimed in claim 6, characterized in that: The dance choreography model deployment module is specifically used to: The dance movement choreography model that passes the test is compressed using knowledge distillation technology, and the compressed dance movement choreography model is deployed to the dance robot using containerization technology.

10. The dancing robot motion control system combining GRU and DTW as claimed in claim 6, characterized in that: The dancing robot motion control module is specifically used for: The dancing robot obtains input real-time music data through a communication module or a microphone, pre-processes the real-time music data, and then inputs the data into a deployed dance movement choreography model to obtain a dance movement choreography result corresponding to the real-time music data, and executes the dance movement choreography result to perform a dance performance.

Images