A method, system, medium, and device for controlling the dance of a humanoid robot.

By constructing a corresponding set of music beat anchor points and human motion frames, frequency conversion and rhythm condition generation are performed, solving the beat drift problem in robot dance control and achieving stable synchronization between robot movements and music.

CN123103668APending Publication Date: 2026-11-17SHANDONG JIANZHU UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611504194.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-24
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing humanoid robot dance control, the motion acquisition frequency is inconsistent with the robot strategy control frequency, resulting in beat drift. Furthermore, the lack of explicit acquisition of the current music rhythm state and forward information affects the synchronization between motion and music.

Method used

By acquiring the synchronization relationship between the target music and the human motion sequence, music beat anchor points are extracted, a set of key beat-action frame correspondences is constructed, and frequency conversion is performed to generate music rhythm conditions that are synchronized frame by frame with the robot's reference motion. The robot is then controlled to perform dance movements according to the actual music playback time.

Benefits of technology

It effectively reduces beat drift in long-sequence dance control, ensures stable synchronization between robot movements and music, and maintains the correspondence between key movements and music beats.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123103668A_ABST
    Figure CN123103668A_ABST
Patent Text Reader

Abstract

This invention proposes a method, system, medium, and device for controlling humanoid robot dance, relating to the field of robot control technology. The method includes: acquiring target music and a sequence of human movements performed by a dancer following the target music; establishing a synchronization relationship between music playback time and human movement time; extracting music beat anchor points; based on the synchronization relationship and the music beat anchor points, obtaining integer keyframe indices of the music beat anchor points in the human movement sequence, and constructing a key beat-action frame correspondence set; acquiring the human movement state at the robot control frequency, and performing frequency conversion on the human movement state based on the synchronization relationship and the key beat-action frame correspondence set; converting the converted human movements frame by frame into reference movements for the humanoid robot; generating music rhythm conditions that are synchronized frame by frame with the robot reference movements; and synchronously reading the reference movements and rhythm conditions according to the actual music playback time to control the humanoid robot to perform dance movements.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of robot control technology, and in particular to a method, system, medium, and device for controlling humanoid robot dance. Background Technology

[0002] Humanoid robot dance control not only requires the robot to meet constraints on joint range of motion, actuation capability, and contact stability, but also requires that motion events such as foot landing, limb motion extremes, pauses, and key postures maintain a stable time correspondence with the beat, accents, and key points of the target music. Current human dance movements are typically captured during the playback of the target music, but the motion capture frequency, motion editing frequency, and robot strategy control frequency are often inconsistent. For example, human movements may be captured at 60 Hz, while the robot's reference motion and strategy control frequency uses 50 Hz. Simply modifying the file frame rate, directly deleting motion frames, or incorrectly scaling the timeline can easily alter the actual duration of the movements, causing key movements that were originally aligned with the music beat to gradually advance or lag behind.

[0003] Furthermore, after the robot's actions are redirected by the human body, although the robot's reference actions can still maintain a beat correspondence on the timeline, conventional motion tracking strategies usually only receive robot state and reference action information and cannot explicitly obtain the current music rhythm state and the forward information of the next key beat. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes a humanoid robot dance control method, system, medium, and device, which effectively reduces beat drift in long-sequence robot dance control.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for controlling the dance of a humanoid robot, comprising: Acquire the target music and the sequence of human movements formed by the dancers following the target music, and establish the synchronization relationship between the music playback time and the human movement time; Perform beat analysis on the target music and extract the music beat anchor points; Based on the synchronization relationship and music beat anchor points, obtain the integer keyframe index of the music beat anchor point in the human action sequence, and construct the key beat-action frame correspondence set. The human motion state under the robot control frequency is obtained, and the human motion state is frequency-converted according to the synchronization relationship and the key beat-action frame correspondence set; the converted human motion is converted into humanoid robot reference motion frame by frame. Music rhythm conditions based on strategy observation and frame-by-frame synchronization of robot reference actions; The reference action and rhythm conditions are read synchronously according to the actual music playback time to determine the current music playback time and control the humanoid robot to perform dance actions.

[0006] In a second aspect, the present invention provides a humanoid robot dance control system, comprising: The music-motion synchronization module is configured to acquire target music and a sequence of human movements performed by dancers following the target music, and to establish a synchronization relationship between the music playback time and the human movement time. The beat anchor point extraction module is configured to perform beat analysis on the target music and extract the music beat anchor points; The keyframe acquisition module is configured to acquire the integer keyframe index of the music beat anchor point in the human action sequence based on the synchronization relationship and the music beat anchor point, and construct a key beat-action frame correspondence set. The frequency conversion module is configured to acquire the human motion state under the robot control frequency, perform frequency conversion on the human motion state according to the synchronization relationship and the key beat-action frame correspondence set, and convert the converted human motion into humanoid robot reference motion frame by frame. The rhythm condition generation module is configured to generate music rhythm conditions that are synchronized frame by frame with the robot's reference action based on policy observation. The control module is configured to synchronously read the reference action and rhythm conditions according to the actual music playback time, determine the current music playback time, and control the humanoid robot to perform dance movements.

[0007] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the humanoid robot dance control method described in the first aspect.

[0008] Fourthly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the humanoid robot dance control method described in the first aspect.

[0009] Compared with the prior art, the beneficial effects of the present invention are as follows: In the offline motion editing and frequency conversion stage, this invention maintains the beat relationship by keeping the physical time of the motion and the music time of the key motion constant. In the motion tracking training and actual deployment stage, the music rhythm condition that is synchronized with the robot's reference motion frame by frame is added to the strategy observation, and a unified reference index is determined according to the actual playback time of the music, thereby reducing beat drift in long-sequence dance.

[0010] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute a limitation thereof.

[0011] Figure 1 A main flowchart of a humanoid robot dance control method provided in an embodiment of the present invention; Figure 2 A schematic diagram of beat-keeping motion editing and frequency conversion based on unified physical time provided in an embodiment of the present invention; Figure 3 A schematic diagram illustrating the synchronous construction of robot reference action, music rhythm conditions, and strategy observation provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a unified music time index for the training and deployment phases provided in an embodiment of the present invention; Figure 5 This invention provides a humanoid robot dance control system. Detailed Implementation

[0012] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0013] Example 1 like Figure 1 As shown, this embodiment discloses a method for controlling the dance of a humanoid robot, including the following steps: S1: Obtain the target music and the sequence of human body movements formed by the dancers following the target music, and establish a synchronization relationship between the music playback time and the human body movement time; S2: Perform beat analysis on the target music and extract the music beat anchor points; S3: Based on the synchronization relationship and the music beat anchor point, obtain the integer keyframe index of the music beat anchor point in the human action sequence, and construct the key beat-action frame correspondence set; S4: Obtain the human motion state under the robot control frequency, and perform frequency conversion on the human motion state according to the synchronization relationship and the key beat-action frame correspondence set; convert the converted human motion into humanoid robot reference motion frame by frame. S5: Music rhythm conditions based on policy observation generation and frame-by-frame synchronization of robot reference actions; S6: Read the reference action and rhythm conditions synchronously according to the actual music playback time, determine the current music playback time, and control the humanoid robot to perform dance actions.

[0014] Next, combined Figure 1 This embodiment provides a detailed description of a humanoid robot dance control method.

[0015] (a) Defining synchronized music-human motion data The target music is played, and dancers follow the music to perform dance movements. The human movements are captured using inertial motion capture devices, optical motion capture devices, or image-based human posture acquisition devices. The original human movement sequence is represented as: ,

[0016] In the formula, This represents the original number of human motion frames; among the various state items, Indicates the first The acquisition time of frame motion, This is the location of the root node of the human body. The pose of the human root node. The posture of human skeletal nodes. This represents the optional foot contact state. The human root node is the root of the skeletal tree, without a parent joint, and is typically taken as the center of the pelvis / hip (center point of the hip), serving as the global reference origin for the entire human body in three-dimensional space.

[0017] Record the start time of the target music playback and the start time of human motion capture Define the synchronization time offset as: ; When the original action sampling frequency is At that time, the first The music playback time corresponding to each frame of human motion is: ; By using the aforementioned time offset, each frame in the human motion sequence is uniformly mapped to the target music timeline, enabling subsequent motion editing, frequency conversion, and robot control to all be based on the same physical time.

[0018] (ii) Constructing a time sequence of musical beats The target music is analyzed for its beats or pre-annotated beat files are read to obtain one or more musical beat anchor points, including ordinary beats, accented beats, measure boundaries, phrase boundaries, or key points in dance. These musical beat anchor points serve as discrete time references on the target music's timeline, clarifying the temporal correspondence between musical beat moments and key events or keyframes of human movement. They also provide a temporal basis for subsequent beat verification after movement frequency conversion, frame-by-frame construction of music rhythm conditions, and a unified reference index for the actual machine control stage.

[0019] The time sequence of musical beats is represented as: ,

[0020] In the formula, For the number of anchor points for the music beat; For the first The time of a musical beat anchor point relative to the start of the music; To normalize beat saliency, this is used to characterize the prominence of the corresponding musical beat anchor point in the musical rhythm structure or its importance to key dance points. A higher value indicates a more significant musical beat anchor point. The beat saliency can be determined based on one or more of the following: audio onset intensity, accent / measure / phrase level, and manually labeled weights, and then normalized to... . This refers to the beat level or beat type. The anchor points for each musical beat are arranged in strict ascending order of time:

[0021] When a downbeat, measure boundary, and phrase boundary occur at the same time, they are merged into a single time-unique musical beat anchor point, and their beat type is represented by a composite label or the highest-level label. The musical beat time sequence can be obtained by audio beat detection algorithms, music analysis software, MIDI beat information, manual annotation, or a combination of the above methods. This embodiment does not require a specific beat detection algorithm as a necessary limitation.

[0022] (III) Establishing the correspondence between key beats and action frames Since the human dance movements are captured synchronously during the playback of the target music, there is no need to re-execute the complex music-movement cross-modal automatic matching. The calculation is based on the music beat anchor time and synchronization time offset. The consecutive frame coordinates and integer keyframe indices of each music beat anchor point in the original human motion sequence: ,

[0023] For dances that do not require setting keyframes for every regular beat, key points can be selected from the music beat anchors, corresponding to foot landing, extreme limb movements, pauses, the end of turns, or the arrival of key poses, forming a set of key beat-action frames:

[0024] This corresponding set is used to verify whether the music timing of key actions has changed after the action editing and frequency conversion are completed, and to provide a time basis for generating frame-by-frame synchronized music rhythm conditions during the training phase.

[0025] (iv) Editing beat-maintaining movements and converting target frequencies The target frequency used by the humanoid robot's reference action or motion tracking strategy. Taking the start time of human motion capture as zero point, the first... The physical time corresponding to each target action frame is: ,

[0026] Query the coordinates of consecutive frames in the original human motion sequence based on the same physical time:

[0027] Based on continuous frame coordinates Interpolate the original human movements to generate the human movement state at the target frequency:

[0028] The root node position can be determined using linear interpolation, cubic spline interpolation, or Hermite interpolation; the root node pose and skeleton node rotation are preferably determined using unit quaternion spherical linear interpolation. This method establishes the action correspondence before and after frequency conversion using the same physical time rather than the same frame number.

[0029] When performing uniform frequency conversion in 3D motion editing software with frame number as the horizontal axis of the time axis, let the synchronization start frame be... The original keyframe coordinates are The transformed keyframe coordinates are:

[0030] In an embodiment where the original motion sampling frequency is 60 Hz and the target frequency is 50 Hz:

[0031] Therefore, scaling the keyframe coordinates after the synchronization start frame by a 5 / 6 ratio and setting the motion playback frame rate to 50 Hz ensures that the actual duration of the motion remains consistent before and after the frequency conversion. After motion editing and frequency conversion are complete, the first... The actual frame index of each key motion event in the target motion sequence is calculated, along with its music timing and beat error. , ,

[0032] when At that time, it is determined that the key action event meets the beat maintenance requirement; It can be set to one or more control cycles according to the target control frequency. When At this time, only local keyframes near the critical action event are translated, compressed, or stretched to ensure they fall back into the allowable time range of the corresponding beat anchor point, while keeping the positions of adjacent unadjusted key beats unchanged or limiting their changes to a preset range. The above beat-preserving motion editing and frequency conversion process is as follows: Figure 2 As shown.

[0033] (v) Redirecting human body movements to humanoid robot reference movements Obtain the kinematic model of the target humanoid robot, and based on the robot's skeletal structure, joint degrees of freedom, limb proportions, and joint range of motion, redirect the human body movements at the target frequency frame by frame to the robot's reference movements: , ,

[0034] In the formula, For the first Frame robot reference joint states, This is a function for redirecting human actions to robot actions. The state of human movement at the target frequency. The kinematic parameters of the target humanoid robot, For the first Frame robot target action frame, For the first The physical time corresponding to each target action frame. For the first Frame robot target frequency, This refers to the target frequency used as a reference action or action tracking strategy for the humanoid robot. The retargeting phase primarily addresses the spatial structural differences between the human and the robot, while maintaining the frame correspondence between input and output actions, the target frequency, and the time index.

[0035] In one embodiment, a GMR-type motion redirection method is used to convert 50 Hz human motion into reference motion for the accelerated evolution Booster K1 22-DOF humanoid robot. During the redirection process, spatial constraints are applied to the robot's joint limits, motion continuity, root node state, foot contact, and foot sliding. When the human posture exceeds the robot's executable range, the motion amplitude is reduced, the root node height is adjusted, or the local joint posture is modified, without changing the music timing and frame index corresponding to the key motion.

[0036] (vi) Constructing a six-dimensional observation of musical rhythm conditions and strategies Based on the time sequence of musical beats, construct musical rhythm conditions for each reference moment in the robot's reference action. Let the... The frame robot references the music timing corresponding to the action. Located at adjacent music beat anchor points and Between these points, the current rhythm phase is: ,

[0037] To avoid rhythm phase from A jump to 0 causes discontinuity in the policy input, using and They jointly represent the periodic phase. A continuous beat salience function is constructed based on the musical beat anchor time and salience:

[0038] In the formula, For the first The influence of a musical beat anchor point on the time scale To prevent positive numbers with a denominator of zero; This describes how close the current moment is to an important beat or key point. The normalized actual remaining time from the current moment to the next musical beat anchor point is defined as:

[0039] In the formula, To preset the maximum look-ahead time, This indicates that the values ​​are limited to between 0 and 1. The salience of the next musical beat anchor point is normalized, and the next beat level is encoded as follows: Finally, a six-dimensional lookahead conditional vector for musical rhythm is constructed:

[0040] The first three dimensions of this six-dimensional vector describe the current rhythmic state, while the last three dimensions describe the look-ahead information for the next musical beat anchor point. This information originates from offline parsing of known target music and does not constitute a prediction of unknown future beats. The rhythmic conditional vector with the same time index as the current reference action is appended to the basic action tracking observation, forming a rhythmic conditionalized strategy observation:

[0041] The frame-by-frame synchronous construction relationship between robot reference actions, music rhythm conditions, and policy observations is as follows: Figure 3 As shown.

[0042] (vii) Training rhythm conditioned motion tracking strategy Import the robot's reference motion sequence into the robot dynamics simulation environment. Basic motion tracking observations can include one or more of the following: robot base state, joint position, joint velocity, historical motion, reference joint state, and reference end effector state. After the basic observations, a six-dimensional music rhythm condition is added, and the input dimension of the policy network is increased by 6 accordingly.

[0043] When the training environment is randomly initialized from any position in the robot's reference action sequence, the reference actions and musical rhythm conditions use the same start index and advance index. The unified reference index can be represented as: ,

[0044] The motion tracking strategy can be trained using PPO-type reinforcement learning algorithms, imitation learning algorithms, or other robot motion tracking algorithms. The strategy outputs one or more of the following: target joint position, target joint position residual, target joint velocity, or target joint torque. In the Booster K1 embodiment, the strategy outputs a 22-dimensional target joint position or target joint position residual, and the underlying joint controller generates the driving torque.

[0045] Preferably, the reward for key postures, end-effector positions, or foot contact can be dynamically weighted using the salience of the current beat:

[0046] In the formula, Based on basic motion tracking rewards, For key action tracking rewards, The corresponding weights are used. This dynamic reward weighting is an optional enhancement method and is not a necessary step for implementing six-dimensional rhythmic conditional observation.

[0047] (viii) Deploy the system in accordance with the unified music time index. After the policy training is complete, the policy model is deployed to the target humanoid robot. When playing target music, the actual start time of the music is recorded, and the current music playback time is calculated based on a monotonic clock.

[0048] A unified reference index is calculated based on the current music playback time.

[0049] In the formula, This is the current music playback time. This is the current monotonic clock time read by the robot control system. The monotonic clock time corresponding to when the target music player begins playing; This is the amount of audio playback link latency compensation. For the robot's reference motion frequency, The total number of reference frames for the robot's motion. For the purpose of unifying the reference index.

[0050] According to the Uniform Reference Index Synchronous reading of the first Frame robot reference action and the first The frame music rhythm conditions are used to construct policy observations that are consistent with the training phase, and the robot joint control target is output through the trained policy.

[0051] When short-term fluctuations occur in the live control loop, the reference action is not advanced according to the cumulative control loop count. Instead, the unified reference index is recalculated based on the current music playback time, thus preventing local periodic fluctuations from evolving into long-term cumulative beat errors. The relationship between the unified music time index during training and live deployment is as follows: Figure 4 As shown.

[0052] Example 2 like Figure 5 As shown, this embodiment provides a humanoid robot dance control system, including: The music-motion synchronization module is configured to acquire target music and a sequence of human movements performed by dancers following the target music, and to establish a synchronization relationship between the music playback time and the human movement time. The beat anchor point extraction module is configured to perform beat analysis on the target music and extract the music beat anchor points; The keyframe acquisition module is configured to acquire the integer keyframe index of the music beat anchor point in the human action sequence based on the synchronization relationship and the music beat anchor point, and construct a key beat-action frame correspondence set. The frequency conversion module is configured to acquire the human motion state under the robot control frequency, perform frequency conversion on the human motion state according to the synchronization relationship and the key beat-action frame correspondence set, and convert the converted human motion into humanoid robot reference motion frame by frame. The rhythm condition generation module is configured to generate music rhythm conditions that are synchronized frame by frame with the robot's reference action based on policy observation. The control module is configured to synchronously read the reference action and rhythm conditions according to the actual music playback time, determine the current music playback time, and control the humanoid robot to perform dance movements.

[0053] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of a humanoid robot dance control method as described in Embodiment 1 above.

[0054] Example 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the humanoid robot dance control method described in Embodiment 1 above.

[0055] The steps or modules involved in Embodiments 2 to 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0056] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A humanoid robot dance control method characterized by, include: Acquire the target music and the sequence of human movements formed by the dancers following the target music, and establish the synchronization relationship between the music playback time and the human movement time; Perform beat analysis on the target music and extract the music beat anchor points; Based on the synchronization relationship and music beat anchor points, obtain the integer keyframe index of the music beat anchor point in the human action sequence, and construct the key beat-action frame correspondence set. The human motion state under the robot control frequency is obtained, and the human motion state is frequency-converted according to the synchronization relationship and the key beat-action frame correspondence set; the converted human motion is converted into humanoid robot reference motion frame by frame. Music rhythm conditions based on strategy observation generation and frame-by-frame synchronization of robot reference actions; The reference action and rhythm conditions are read synchronously according to the actual music playback time to determine the current music playback time and control the humanoid robot to perform dance actions.

2. The humanoid robot dance control method of claim 1, wherein, The establishment of the synchronization relationship between music playback time and human motion time is specifically as follows: a synchronization time offset is defined based on the start time of target music playback and the start time of human motion acquisition; based on the time offset, each frame in the human motion sequence is mapped to the target music time axis, so that the music playback time is aligned with the human motion time.

3. The humanoid robot dance control method according to claim 2, wherein The musical beat anchor points include the start time, salience, and beat level of the musical beat; the beat level includes ordinary beats, accented beats, measure boundaries, phrase boundaries, or key points in dance.

4. The humanoid robot dance control method of claim 1, wherein, The process of acquiring the human motion state under the robot control frequency, and performing frequency conversion on the human motion state based on the synchronization relationship and the key beat-action frame correspondence set, specifically includes: Based on the start time of human motion acquisition, the physical time of the target motion frame under the target control frequency is determined, and the position of consecutive frames in the original human motion sequence is queried according to the physical time. Human motion state under target control frequency is generated based on the interpolation of the consecutive frame positions; On the motion editing timeline, based on the synchronization start frame, the key frame position after the synchronization start frame is scaled according to the ratio of the original motion sampling frequency to the target control frequency, and the playback frame rate is set according to the target control frequency to make the actual duration of the motion consistent before and after the conversion. Based on the key beat-action frame correspondence set, detect the actual frame index of the key action event in the target action sequence and its time error relative to the music beat anchor point; if the error is within the preset range, it is determined that the beat preservation requirement is met; otherwise, only adjust the local key frames near the key action event to fall within the allowed time range of the corresponding beat anchor point, and keep the adjacent unadjusted key beat positions unchanged or limit the changes to the preset range.

5. The humanoid robot dance control method of claim 1, wherein, The process of converting the transformed human motion into humanoid robot reference motion frame by frame specifically includes: , , wherein, is the frame robot reference joint state, is the human-to-robot motion retargeting function, is the human motion state at the target frequency, is the kinematics parameters of the target humanoid robot, is the frame robot target motion frame, is the physical time corresponding to the target motion frame, is the frame robot target frequency, is the target frequency employed by the humanoid robot reference motion or motion tracking policy.

6. The humanoid robot dance control method as described in claim 1, characterized in that, The music rhythm conditions generated based on policy observation and synchronized frame-by-frame with the robot's reference actions specifically include: Based on the music beat time sequence, determine the music time corresponding to each frame of the robot's reference action, locate the two adjacent music beat anchor points, and calculate the current rhythm phase; Based on the time position and salience of the music beat anchor point, a continuous beat salience function is constructed, and the normalized remaining time from the current moment to the next music beat anchor point is defined. Construct a six-dimensional rhythm condition vector, where the first three dimensions describe the current rhythm state and the last three dimensions provide forward information for the next beat; By combining the six-dimensional rhythm conditional vector with the basic motion tracking observations at the same time index, rhythm-conditionalized policy observations are formed to guide robot motion generation.

7. The humanoid robot dance control method as described in claim 1, characterized in that, Determining the current music playback time specifically includes: in, This is the current music playback time. This is the current monotonic clock time read by the robot control system. The monotonic clock time corresponding to when the target music player begins playing; This is the amount of audio playback link latency compensation.

8. A humanoid robot dance control system, characterized in that, include: The music-motion synchronization module is configured to acquire target music and a sequence of human movements performed by dancers following the target music, and to establish a synchronization relationship between the music playback time and the human movement time. The beat anchor point extraction module is configured to perform beat analysis on the target music and extract the music beat anchor points; The keyframe acquisition module is configured to acquire the integer keyframe index of the music beat anchor point in the human action sequence based on the synchronization relationship and the music beat anchor point, and construct a key beat-action frame correspondence set. The frequency conversion module is configured to acquire the human motion state under the robot control frequency, perform frequency conversion on the human motion state according to the synchronization relationship and the key beat-action frame correspondence set, and convert the converted human motion into humanoid robot reference motion frame by frame. The rhythm condition generation module is configured to generate music rhythm conditions that are synchronized frame by frame with the robot's reference action based on policy observation. The control module is configured to synchronously read the reference action and rhythm conditions according to the actual music playback time, determine the current music playback time, and control the humanoid robot to perform dance movements.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the humanoid robot dance control method as described in any one of claims 1-7.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the humanoid robot dance control method as described in any one of claims 1-7.