Humanoid Robot Motion Control Method Based on Action Data Analysis

Abstract

The present invention discloses a humanoid robot motion control method based on motion data analysis, belonging to the technical field of robot motion control. The present invention collects multi-modal motion data during the motion of a humanoid robot, extracts key motion features, identifies the motion state of the robot's actions, analyzes the variation laws of key parameters in each motion state, constructs joint motion models under different motion states, controls the switching between different states, combines the current motion state with the target state, generates and optimizes the robot's motion trajectory, calculates the energy consumption of different trajectory schemes based on energy consumption estimation, and optimizes the trajectory using the energy-optimal control method, enabling the robot to balance motion smoothness and energy efficiency when performing tasks. By collecting feedback data during the robot's execution of motion in real time and calculating the motion error deviation, the control signal is dynamically adjusted to ensure that the robot can still maintain an accurate motion trajectory in a complex environment.

Description

Technical Field

[0001] The present invention relates to the technical field of robot motion control, and particularly to a humanoid robot motion control method based on action data analysis. Background Technique

[0002] Humanoid robots are increasingly widely used in fields such as industrial manufacturing, medical rehabilitation, the service industry, and special task execution. Due to their human-like form, humanoid robots can more naturally adapt to the human environment and achieve flexible movement and operation. However, since humanoid robots usually have a multi-degree-of-freedom joint system, their motion control process involves complex dynamic characteristics and high-dimensional state variables, making precise and efficient motion control one of the technical difficulties. Traditional humanoid robot motion control methods mainly rely on predefined motion trajectories and control strategies. In the face of complex environments or dynamic tasks, they have the following deficiencies:

[0003] Since it is difficult to adjust the preset trajectory and control strategy according to real-time environmental changes, the robot is difficult to adapt to complex dynamic environments when performing tasks. Based on static optimization or simple feedback control, it is difficult to accurately capture the key characteristics of the robot in different motion states, resulting in large trajectory errors during the robot's execution process, affecting the quality of task completion. At the same time, humanoid robots usually need to consider energy consumption issues during movement, but traditional methods pay less attention to energy-optimal trajectory planning and joint control optimization, resulting in high energy consumption when the robot performs tasks, affecting the battery life and long-term stable operation. Summary of the Invention

[0004] The purpose of the present invention is to provide a humanoid robot motion control method based on action data analysis to solve the problems raised in the above background technique.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A humanoid robot motion control method based on action data analysis, including:

[0006] Obtain multi-modal action data during the robot's movement, extract key motion features based on the multi-modal action data and identify the motion state of the robot's actions, construct a robot motion model, combine the robot's current motion state and the target state, generate and optimize the robot's motion trajectory, generate corresponding control instructions according to the motion trajectory, compare with the preset trajectory, calculate the error deviation, and dynamically adjust the control instructions based on the error deviation.

[0007] Further, when obtaining the multi-modal action data during the robot's movement, it is obtained through inertial measurement units, force sensors, and angle encoders installed on each joint of the humanoid robot. The multi-modal action data includes joint angles, angular velocities, accelerations, joint torques, and the force on the end effector;

[0008] Perform time alignment and interpolation processing on multi-modal action data to generate a standardized time data sequence.

[0009] Furthermore, when extracting key motion features from multi-modal action data and performing motion state recognition on robot actions, feature extraction is performed on the multi-modal action data, and the key motion features include motion phase, joint angle change rate, and energy consumption estimation;

[0010] Based on the angle, angular velocity, and acceleration data of each joint of the robot, calculate the phase information of the robot during continuous motion, and extract key motion phase parameters;

[0011] Calculate the joint angle change rate of each joint of the robot;

[0012] Based on joint torque and angular velocity data, calculate the energy consumption estimation of each joint per unit time;

[0013] Construct a multi-dimensional feature vector based on key motion features and perform dimensionality reduction on the multi-dimensional feature vector;

[0014] Based on the key motion features after dimensionality reduction, classify and recognize the robot motion state through hierarchical clustering, and divide it into different motion states;

[0015] The motion states include start, acceleration, steady state, deceleration, and stop.

[0016] Furthermore, based on joint torque and angular velocity data, calculating the energy consumption estimation of each joint per unit time includes:

[0017] Real-time monitor the joint torque and angular velocity data of the joint points;

[0018] Retrieve the theoretical value of the joint torque corresponding to the sampling moment of each joint point;

[0019] According to the torque data deviation between the actual value of the joint torque and the theoretical value of the joint torque at the sampling moment of each joint point;

[0020] Retrieve the theoretical value of the angular velocity data corresponding to the sampling moment of each joint point;

[0021] According to the angular velocity data deviation between the actual value of the angular velocity data and the theoretical value of the angular velocity data at the sampling moment of each joint point;

[0022] When there is no torque data deviation and angular velocity data deviation, use the basic energy consumption estimation model to obtain the energy consumption estimation of each joint point;

[0023] Among them, the energy consumption estimation of each joint point is obtained through the following formula:

[0024]

[0025] Among them, E represents the estimated energy consumption of each joint point; n represents the number of sampling times of joint torque and angular velocity data; T i represents the joint torque at the i-th sampling; w i represents the angular velocity data at the i-th sampling; Δt represents the sampling time interval;

[0026] When there are torque data deviations and angular velocity data deviations, the occurring torque data deviations and angular velocity data deviations are retrieved;

[0027] The estimated energy consumption is obtained by combining the torque data deviation and the angular velocity data deviation with the basic energy consumption estimation model.

[0028] Furthermore, obtaining the estimated energy consumption by combining the torque data deviation and the angular velocity data deviation with the basic energy consumption estimation model includes:

[0029] Retrieve all the torque data deviations and angular velocity data deviations obtained from the current joint point monitoring;

[0030] Perform standardization processing on the torque data deviation and the angular velocity data deviation to obtain the standardized torque data deviation and angular velocity data deviation;

[0031] Integrate all the standardized torque data deviations to generate a torque data deviation data set;

[0032] Obtain the average torque deviation ratio according to each standardized torque data deviation in the torque data deviation data set;

[0033] Integrate all the standardized angular velocity data deviations to generate an angular velocity data deviation data set;

[0034] Obtain the average angular velocity deviation ratio according to each standardized angular velocity data deviation in the angular velocity data deviation data set;

[0035] Obtain the energy consumption loss coefficient according to the average torque deviation ratio and the average angular velocity deviation ratio;

[0036] Among them, the energy consumption loss coefficient is obtained through the following formula:

[0037]

[0038] Among them, ξ represents the energy consumption loss coefficient; T p and w prespectively represent the average torque deviation ratio and the average angular velocity deviation ratio; α represents a preset power-law scaling coefficient; β represents the damping coefficient corresponding to the joint point; γ represents the suppression coefficient of the adjustment torque amplitude on the coupling term; λ represents a preset exponential term weight coefficient;

[0039] Use the energy consumption loss coefficient to obtain the energy consumption estimate in combination with the basic energy consumption estimation model;

[0040] Among them, the energy consumption estimate of each joint point is obtained through the following formula:

[0041]

[0042] Among them, E represents the energy consumption estimate of each joint point; n represents the sampling times of joint torque and angular velocity data; T i represents the joint torque at the i-th sampling; w i represents the angular velocity data at the i-th sampling; Δt represents the sampling time interval; ξ represents the energy consumption loss coefficient.

[0043] Furthermore, when constructing the robot motion model, based on the extracted key motion features, analyze the variation laws of key parameters in each motion state;

[0044] Construct the angle, angular velocity, acceleration, and torque variation models of each joint in different motion states, and the variation models are used to describe the dynamic variation trend of the joint over time;

[0045] According to the complete motion process of the robot from startup to stop, based on the time data sequence, analyze the transition relationship between different motion states, and establish a robot motion state transition model through the state machine modeling method. The motion state transition model is used to define the transition conditions and transition modes of different motion states;

[0046] Analyze the mutual influence relationship of joint motions in different motion states. The mutual influence relationship includes the cooperative effect between adjacent joints, inertial influence, and external environment interference factors on the robot motion, and establish a state dependence model based on the multivariate regression method.

[0047] Furthermore, when constructing the robot motion model, use the Lagrangian method to establish the robot motion equation. The robot motion equation is used to describe the dynamic characteristics of each joint of the robot in different states. By analyzing the relationship between joint torque, speed, acceleration, and energy consumption, establish the robot motion model of the robot in different motion states;

[0048] Combine the actual operation data of the robot, optimize the parameters of the robot motion model in a data-driven manner, and train the robot motion model through machine learning;

[0049] Combined with the historical motion data of the robot, the constructed robot motion model is subjected to simulation testing and error analysis, the error between the model prediction value and the actual measurement value is calculated, and the model is optimized through parameter adjustment.

[0050] Furthermore, when generating and optimizing the robot motion trajectory, based on the robot motion model, combined with the current motion state and the target state, the key parameters for trajectory generation are determined. The key parameters include joint angle, angular velocity, acceleration, and path point information, and a trajectory parameterization model is established;

[0051] Discrete trajectory points are generated between the current state and the target state of the robot;

[0052] Based on the robot motion state transition model, the motion time allocation for different trajectory segments is calculated. The motion time allocation includes the time ratios of the acceleration, constant velocity, and deceleration phases;

[0053] For the initially generated motion trajectory, combined with the robot structure characteristics and task requirements, constraint conditions are set. The constraint conditions include joint angle range constraints, maximum velocity constraints, maximum acceleration constraints, and joint torque constraints, and the motion trajectory is optimized and adjusted using constraint optimization;

[0054] Based on the energy consumption estimation in the robot motion model, the energy consumption of different motion trajectories is calculated, and the motion trajectory is optimized through the energy optimal control method, and the optimized motion trajectory is smoothed.

[0055] Furthermore, when generating the corresponding control instructions according to the motion trajectory, based on the optimized motion trajectory, the target states of each joint of the robot are extracted. The target states include angle, angular velocity, and acceleration parameters. Combined with the trajectory time allocation information, the expected motion states of each joint at different time steps are determined;

[0056] Based on the robot motion model, the joint torques and control torques required for each joint of the robot to execute the target actions are calculated, and a target joint torque distribution is generated;

[0057] The calculated joint target states, joint torques, and control torques are converted into standardized control signals, and the control signals are sent to the robot joint drive unit through the real-time control bus.

[0058] Furthermore, when calculating the error deviation and dynamically adjusting the control instructions based on the error deviation, the motion state data of each joint during the robot's execution of the motion is collected in real time. The motion state data includes joint angle, angular velocity, acceleration, joint torque, and the force condition of the end effector;

[0059] Based on the optimized motion trajectory, compare the current joint angle data of the robot with the target angles in the preset trajectory, and calculate the angle errors of each joint;

[0060] Combined with the robot motion model, calculate the actual trajectory of the robot end effector, and compare it with the preset trajectory to calculate the trajectory deviation, where the trajectory deviation includes spatial position deviation and direction deviation;

[0061] Based on the joint angle errors and trajectory deviation, generate an error compensation factor in combination with the robot motion model, and adjust the control signals of each joint based on the error compensation factor to real-time correct the target joint angles, angular velocities, and control torques.

[0062] Compared with the prior art, the beneficial effects of the present invention are:

[0063] 1. By collecting multi-modal action data during the movement of the humanoid robot, generating a standardized time data sequence, extracting key motion features, and performing motion state recognition on the robot actions, and subdividing them into motion states such as start, acceleration, steady state, deceleration, and stop, the robot can accurately judge its own motion state. The motion recognition method based on multi-modal data fusion improves the robot's understanding ability of complex motion processes, enabling it to more intelligently adapt to different task scenarios.

[0064] 2. When constructing the robot motion model, by analyzing the variation laws of the key parameters of each motion state, constructing joint motion models under different motion states, and using the state machine modeling method to establish a robot motion state transition model, so as to accurately control the switching between different states. Combining the current motion state and the target state, generate and optimize the robot motion trajectory, and use trajectory parametric modeling and constraint optimization techniques to ensure that the trajectory meets the joint angle, speed, acceleration, and torque constraint conditions. Calculate the energy consumption of different trajectory schemes based on the energy consumption estimation, and use the energy optimal control method to optimize the trajectory, enabling the robot to balance motion smoothness and energy efficiency when performing tasks, and improving the robustness and adaptability of the overall motion control.

[0065] 3. By real-time collecting the feedback data during the robot's execution of the motion, comparing it with the preset trajectory, calculating the motion error deviation, calculating the error compensation factor based on the robot motion model, and combining the current motion state of the robot, dynamically adjust the control signal. By continuously correcting the target joint angles, angular velocities, and control torques, achieve high-precision error compensation, ensure that the robot can still maintain an accurate motion trajectory in a complex environment, dynamically adjust the control parameters according to the real-time feedback data, enable the robot to maintain stable motion under different loads, different terrains, and different task conditions, improve the adaptability of the robot in the real environment, and improve the reliability and execution efficiency of the overall motion control. Description of the Drawings

[0066] Figure 1 It is a schematic flow chart of the motion control method for the humanoid robot of the present invention. Detailed Embodiments

[0067] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0068] Please refer to Figure 1 , the present invention provides the following technical solutions:

[0069] A motion control method for a humanoid robot based on action data analysis, including:

[0070] Obtain multi-modal action data when the robot moves, extract key motion features based on the multi-modal action data and identify the motion state of the robot's actions, construct a robot motion model, combine the current motion state and the target state of the robot, generate and optimize the robot motion trajectory, generate corresponding control instructions according to the motion trajectory, compare with the preset trajectory, calculate the error deviation and dynamically adjust the control instructions based on the error deviation.

[0071] When obtaining the multi-modal action data when the robot moves, it is obtained through the inertial measurement units, force sensors and angle encoders installed on each joint of the humanoid robot. The multi-modal action data includes joint angles, angular velocities, accelerations, joint torques and the forces on the end effector;

[0072] Perform time alignment and interpolation processing on the multi-modal action data to generate a standardized time data sequence.

[0073] In the above embodiments, by acquiring multi-modal action data during the movement of the robot and performing time alignment and interpolation processing on the data, high-precision data acquisition and standardized preprocessing are achieved. The acquisition method of multi-modal action data fully considers various key parameters that the robot may involve during the movement process, such as joint angles, angular velocities, accelerations, joint torques, and the force conditions of the end effector, which truly reflects the current movement state of the robot. Time alignment and interpolation processing ensure the synchronization of different sensor data. Through the timestamp alignment method, all data can be adjusted to a unified time reference, while interpolation processing can fill in the missing data points, making the finally obtained time data sequence have the characteristic of equal time intervals, effectively improving the data quality, providing more accurate and stable input data for the robot motion control system, and thus enhancing the reliability and adaptability of the entire control method.

[0074] When extracting key motion features based on multi-modal action data and performing motion state recognition on the robot's actions, feature extraction is performed on the multi-modal action data, and the key motion features include motion phase, joint angle change rate, and energy consumption estimation value;

[0075] Based on the angle, angular velocity, and acceleration data of each joint of the robot, calculate the phase information during the continuous movement process of the robot and extract key motion phase parameters;

[0076] Calculate the joint angle change rate of each joint of the robot;

[0077] Based on the joint torque and angular velocity data, calculate the energy consumption estimation value of each joint per unit time;

[0078] Construct a multi-dimensional feature vector based on the key motion features and perform dimensionality reduction on the multi-dimensional feature vector;

[0079] Based on the key motion features after dimensionality reduction, classify and identify the robot's motion state through hierarchical clustering and divide it into different motion states;

[0080] The motion states include start, acceleration, steady state, deceleration, and stop.

[0081] In the above embodiments, through feature extraction technology, the most representative key motion features during the robot's motion are refined, such as motion phase, joint angle change rate, and energy consumption estimation value, making the analysis of the motion state more accurate. Among them, the motion phase is used to describe the coordination of each joint of the robot during continuous motion, the joint angle change rate reflects the change trend of the robot's actions, and the energy consumption estimation value helps to optimize the robot's energy utilization efficiency. Through the construction of multi-dimensional feature vectors and dimensionality reduction processing, the data dimension is reduced, thereby reducing the computational complexity and improving the processing efficiency. In the feature space after dimensionality reduction, the distribution of the robot's motion state is clearer, which helps with subsequent motion state classification. By using the hierarchical clustering method, different motion states of the robot can be accurately classified, improving the accuracy and robustness of state recognition, being able to analyze the robot's current motion state in real time, and also being able to perform trend prediction based on historical data, providing an important basis for subsequent trajectory optimization and control instruction adjustment.

[0082] Specifically, based on the joint torque and angular velocity data, calculate the energy consumption estimation value of each joint per unit time, including:

[0083] Real-time monitor the joint torque and angular velocity data of the joint points;

[0084] Retrieve the theoretical value of the joint torque corresponding to the sampling moment of each joint point;

[0085] According to the torque data deviation between the actual value of the joint torque and the theoretical value of the joint torque at the sampling moment of each joint point;

[0086] Retrieve the theoretical value of the angular velocity data corresponding to the sampling moment of each joint point;

[0087] According to the angular velocity data deviation between the actual value of the angular velocity data and the theoretical value of the angular velocity data at the sampling moment of each joint point;

[0088] When there is no torque data deviation and angular velocity data deviation, use the basic energy consumption estimation model to obtain the energy consumption estimation value of each joint point;

[0089] Among them, the energy consumption estimation value of each joint point is obtained through the following formula:

[0090]

[0091] Among them, E represents the energy consumption estimation value of each joint point; n represents the sampling times of the joint torque and angular velocity data; T i represents the joint torque at the i-th sampling; w i represents the angular velocity data at the i-th sampling; Δt represents the sampling time interval;

[0092] When there are torque data deviations and angular velocity data deviations, the occurring torque data deviations and angular velocity data deviations are retrieved.

[0093] The torque data deviation and the angular velocity data deviation are used in combination with a basic energy consumption estimation model to obtain an energy consumption estimation.

[0094] The technical effects of the above technical solution are as follows: In the above technical solution, represents the mechanical power at the current moment, the integral of the absolute values of all instantaneous powers per unit time, reflecting the total mechanical energy consumption; the power is accumulated within a unit time window Δt (such as 1 second) and multiplied by the time interval to obtain the work done by the joint within the time interval Δt of the i-th sampling, that is, the energy consumption of the joint during this time period.

[0095] By real-time monitoring of the joint torque and angular velocity data of the joint point and comparing them with the theoretical values, data deviations can be discovered and corrected in a timely manner. This real-time monitoring and correction mechanism ensures that the data used for energy consumption calculation is more accurate, thereby improving the accuracy of the energy consumption estimation. When there are no data deviations, directly using the basic energy consumption estimation model for calculation can ensure the most accurate energy consumption estimation under ideal conditions. This technical solution takes into account the existence of data deviations and provides corresponding processing mechanisms. When there are torque data deviations and angular velocity data deviations, instead of simply ignoring these deviations, these deviation information is retrieved and used to correct the energy consumption estimation. This design enables the system to adapt to different working environments and conditions, improving the flexibility and adaptability of the system. Through real-time monitoring and data processing, abnormal data can be discovered and corrected in a timely manner, avoiding calculation errors and resource waste caused by inaccurate data. This technical solution ensures the reliability of the energy consumption estimation through multiple links such as real-time monitoring, data correction, and model application. Even in the presence of data deviations, the estimation can be corrected by retrieving and using this deviation information, thereby improving the reliability of the entire system.

[0096] Specifically, using the torque data deviation and the angular velocity data deviation in combination with a basic energy consumption estimation model to obtain an energy consumption estimation includes:

[0097] Retrieve all the torque data deviations and angular velocity data deviations obtained by monitoring the current joint point.

[0098] Perform standardization processing on the torque data deviation and the angular velocity data deviation to obtain the standardized torque data deviation and angular velocity data deviation.

[0099] Integrate all the standardized torque data deviations to generate a torque data deviation data set.

[0100] Obtain the average torque deviation ratio based on each of the torque data deviations after normalization in the torque data deviation dataset;

[0101] Integrate all the angular velocity data deviations after normalization to generate an angular velocity data deviation dataset;

[0102] Obtain the average angular velocity deviation ratio based on each of the angular velocity data deviations after normalization in the angular velocity data deviation dataset;

[0103] Obtain the energy consumption loss coefficient based on the average torque deviation ratio and the average angular velocity deviation ratio;

[0104] Among them, the energy consumption loss coefficient is obtained through the following formula:

[0105]

[0106] Among them, ξ represents the energy consumption loss coefficient; T p and w p respectively represent the average torque deviation ratio and the average angular velocity deviation ratio; α represents a preset power-law scaling coefficient; β represents the damping coefficient corresponding to the joint point; γ represents the suppression coefficient of the adjustment torque amplitude on the coupling term; λ represents a preset exponential term weight coefficient;

[0107] Use the energy consumption loss coefficient to combine with the basic energy consumption estimation model to obtain the energy consumption estimation;

[0108] Among them, the energy consumption estimation of each joint point is obtained through the following formula:

[0109]

[0110] Among them, E represents the energy consumption estimation of each joint point; n represents the sampling times of the joint torque and angular velocity data; T i represents the joint torque at the i-th sampling; w i represents the angular velocity data at the i-th sampling; Δt represents the sampling time interval; ξ represents the energy consumption loss coefficient.

[0111] The technical effect of the above technical solution is: In the above technical solution Dynamically adjust the influence of the torque-angular velocity product on energy consumption through the power-law scaling coefficient, the damping coefficient, and the suppression coefficient of the adjustment torque amplitude on the coupling term, avoiding the problem of excessive dependence on the torque amplitude in the direct form of the traditional linear product (P = τω). At the same time, When the torque and the angular velocity deviation are in the same direction Enhance the energy consumption penalty, and when the deviation is in the opposite direction Reduce the penalty when reflecting the actual system energy feedback characteristics. Meanwhile, the exponential function amplifies the contribution of large deviations, forcing the model to focus on abnormal working conditions.

[0112] By introducing torque data deviation and angular velocity data deviation and normalizing them, various influencing factors in actual operation can be considered more comprehensively, thus improving the accuracy of energy consumption estimation. Calculating the energy consumption loss coefficient using the average torque deviation ratio and the average angular velocity deviation ratio further considers the impact of data deviation on energy consumption, making the estimation closer to the actual situation. This technical solution can dynamically adjust the energy consumption loss coefficient according to different working conditions and the characteristics of joints (such as damping coefficient, suppression coefficient of the coupling term by adjusting the torque amplitude, etc.), thereby enhancing the adaptability of the system. By adjusting parameters such as the power-law scaling coefficient and the exponential term weight coefficient, the calculation of the energy consumption loss coefficient can be further optimized to adapt to different application scenarios and requirements. This technical solution can detect and correct data deviation in a timely manner through real-time monitoring and data processing, thus avoiding estimation errors caused by inaccurate data. At the same time, by introducing the energy consumption loss coefficient, the energy consumption situation in actual operation can be more accurately reflected, improving the reliability of the estimation.

[0113] In summary, this technical solution shows significant technical effects in terms of estimation accuracy, adaptability, calculation efficiency, reliability, and scalability, providing strong technical support for the accurate assessment of joint energy consumption.

[0114] When constructing a robot motion model, based on the extracted key motion features, analyze the variation laws of key parameters in each motion state;

[0115] Construct the angle, angular velocity, acceleration, and torque variation models of each joint in different motion states. The variation models are used to describe the dynamic variation trend of the joint over time;

[0116] Based on the complete motion process of the robot from startup to stop, analyze the transition relationship between different motion states based on time data series. Through the state machine modeling method, establish a robot motion state transition model. The motion state transition model is used to define the transition conditions and transition modes of different motion states;

[0117] Analyze the mutual influence relationship of joint motions in different motion states. The mutual influence relationship includes the synergistic effect between adjacent joints, inertial influence, and interference factors of the external environment on the robot motion. And establish a state dependence model based on the multivariate regression method.

[0118] The Lagrangian method is used to establish the robot motion equation, which is used to describe the dynamic characteristics of each joint of the robot under different states. By analyzing the relationships among joint torque, speed, acceleration, and energy consumption, a robot motion model under different motion states is established;

[0119] Combined with the actual operation data of the robot, the parameters of the robot motion model are optimized in a data-driven manner, and the robot motion model is trained through machine learning;

[0120] Combined with the historical motion data of the robot, the constructed robot motion model is subjected to simulation testing and error analysis, the error between the model prediction value and the actual measurement value is calculated, and the model is optimized through parameter adjustment.

[0121] In the above embodiments, for different motion states, a detailed joint motion change model is established, including the change trends of angle, angular velocity, acceleration, and torque, so that the dynamic characteristics of the robot motion are accurately described, the motion laws of the robot under different states can be reflected, and it helps to improve the predictability and stability of the motion trajectory. The state machine modeling method can effectively describe the transition relationships between states such as startup, acceleration, steady state, deceleration, and stop of the robot, and define the transition modes between different states, which enables the robot to automatically adjust to the optimal motion mode according to the current state during the actual execution process, avoid the impact caused by sudden changes, and improve the smoothness and coordination of the motion. The state-dependent modeling method considers the interaction relationships among the joints of the robot, including cooperative effects, inertial influences, and external interference factors, making the motion model more adaptable and robust, and accurately predicting the dynamic responses of the robot under different states.

[0122] When generating and optimizing the robot motion trajectory, based on the robot motion model, combined with the current motion state and the target state, the key parameters for trajectory generation are determined. The key parameters include joint angle, angular velocity, acceleration, and path point information, and a trajectory parameterization model is established;

[0123] Discrete trajectory points are generated between the current state and the target state of the robot;

[0124] Based on the robot motion state transition model, the motion time allocation for different trajectory segments is calculated. The motion time allocation includes the time ratios of the acceleration, uniform motion, and deceleration stages;

[0125] For the initially generated motion trajectory, combined with the robot structure characteristics and task requirements, constraint conditions are set. The constraint conditions include joint angle range constraints, maximum speed constraints, maximum acceleration constraints, and joint torque constraints, and the motion trajectory is optimized and adjusted using constraint optimization;

[0126] Based on the energy consumption estimation in the robot motion model, calculate the energy consumption of different motion trajectories, optimize the motion trajectories through the energy optimal control method, and smooth the optimized motion trajectories.

[0127] In the above embodiment, based on the robot motion model, combining the current state and the target state, a trajectory that conforms to the robot motion characteristics is generated, and the smoothness and rationality of the trajectory are ensured. The trajectory parameterization model is adopted to make the degree of freedom of trajectory optimization higher, and it can be flexibly adjusted according to different application scenarios. In the process of trajectory optimization, the structural characteristics and task requirements of the robot are fully considered, and a variety of constraint conditions are set, such as joint angle range constraints, maximum speed constraints, maximum acceleration constraints, and joint torque constraints, so that the robot can meet the physical limitations and safety requirements during motion execution, and avoid overloading or instability. The application of the energy optimal control method enables the robot to minimize energy consumption to the greatest extent and improve the endurance when performing tasks. By optimizing the energy consumption of different motion trajectories, the optimal motion plan can be selected to further improve the economy and sustainability of the robot motion.

[0128] When generating the corresponding control commands according to the motion trajectory, based on the optimized motion trajectory, extract the target states of each joint of the robot. The target states include angle, angular velocity, and acceleration parameters. Combine the trajectory time allocation information to determine the expected motion states of each joint at different time steps.

[0129] Based on the robot motion model, calculate the joint torques and control torques required for each joint of the robot to execute the target actions, and generate the target joint torque distribution.

[0130] Convert the calculated joint target states, joint torques, and control torques into standardized control signals, and send the control signals to the robot joint drive unit through the real-time control bus.

[0131] In the above embodiment, based on the optimized motion trajectory, the target states of each joint of the robot are extracted, and combined with the trajectory time allocation information, it is ensured that the control commands can be executed strictly according to the time steps, so that the motion states of the robot at different stages can accurately match the trajectory planning results. Calculate the joint torques and control torques required to execute the target actions through the robot motion model, and generate the target joint torque distribution, so that the robot can maintain a stable torque output in different motion states, and avoid motion deviation or oscillation caused by unreasonable torque distribution. Convert the calculated joint target states, joint torques, and control torques into standardized control signals, and send them to the robot joint drive unit through the real-time control bus, making the transmission of the control signals more stable and efficient, and ensuring that the robot can accurately execute the motion tasks according to the planned trajectory.

[0132] When calculating the error deviation and dynamically adjusting the control instruction based on the error deviation, the motion state data of each joint during the robot's execution of motion is collected in real time. The motion state data includes joint angle, angular velocity, acceleration, joint torque, and the force condition of the end effector.

[0133] Based on the optimized motion trajectory, the current joint angle data of the robot is compared with the target angle in the preset trajectory to calculate the angle error of each joint.

[0134] Combined with the robot motion model, the actual trajectory of the robot's end effector is calculated and compared with the preset trajectory to calculate the trajectory deviation. The trajectory deviation includes spatial position deviation and direction deviation.

[0135] Based on the joint angle error and the trajectory deviation, an error compensation factor is generated in combination with the robot motion model. Based on the error compensation factor, the control signals of each joint are adjusted to correct the target joint angle, angular velocity, and control torque in real time.

[0136] In the above embodiment, the joint motion state data during the robot's execution process is collected in real time, and compared with the preset trajectory to calculate the joint angle error and the trajectory deviation, enabling the robot control system to accurately identify the deviations existing in the motion process. Based on the robot motion model, the error compensation factor is calculated, and the control signal is dynamically adjusted based on the error compensation factor to ensure that the robot can continuously correct the target joint angle, angular velocity, and control torque during the execution process, minimizing the trajectory error, establishing a complete error closed-loop compensation mechanism, enabling the robot to continuously adjust its own state during the motion process, improving the adaptability and stability of the motion, and ensuring that the robot can operate stably in a complex environment.

[0137] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, shall be covered by the protection scope of the present invention.

Claims

1. A humanoid robot motion control method based on motion data analysis, characterized in that: include: Acquire multimodal motion data of the robot during motion, extract key motion features based on the multimodal motion data and identify the motion state of the robot, build a robot motion model, combine the robot's current motion state with the target state, generate and optimize the robot's motion trajectory, generate corresponding control instructions based on the motion trajectory, compare with the preset trajectory, calculate the error deviation and dynamically adjust the control instructions based on the error deviation; When extracting key motion features based on multimodal motion data and performing motion state recognition on the robot motion, feature extraction is performed on the multimodal motion data, wherein the key motion features include motion phase, joint angle change rate, and energy consumption estimation; Based on the angle, angular velocity and acceleration data of each joint of the robot, calculate the phase information of the robot during continuous motion and extract key motion phase parameters; Calculate the change rate of the joint angle of each joint of the robot; Based on the joint torque and angular velocity data, the estimated energy consumption per unit time of each joint is calculated; Construct a multi-dimensional feature vector based on key motion features and reduce the dimension of the multi-dimensional feature vector; Based on the key motion features after dimensionality reduction, the robot's motion state is classified and identified through hierarchical clustering, and divided into different motion states; The motion state includes starting, accelerating, steady state, decelerating and stopping; Based on the joint torque and angular velocity data, the energy consumption of each joint per unit time is estimated, including: Real-time monitoring of joint torque and angular velocity data of joint points; Retrieve the theoretical value of the joint torque corresponding to the sampling moment of each joint point; According to the torque data deviation between the actual value of the joint torque at the sampling moment of each joint point and the theoretical value of the joint torque; Retrieve the theoretical value of the angular velocity data corresponding to the sampling moment of each joint point; According to the angular velocity data deviation between the actual value of the angular velocity data at the sampling moment of each joint point and the theoretical value of the angular velocity data; When there is no torque data deviation and angular velocity data deviation, the basic energy consumption estimation model is used to obtain the energy consumption estimation of each joint point; When there is a torque data deviation and an angular velocity data deviation, the torque data deviation and the angular velocity data deviation are retrieved; Obtaining energy consumption estimation using the torque data deviation and the angular velocity data deviation in combination with a basic energy consumption estimation model; When the robot motion state is recognized, the joint torque and angular velocity data are obtained, the torque data deviation and the angular velocity data deviation are calculated, and the energy consumption estimation of each joint per unit time is calculated in combination with the basic energy consumption estimation model.

2. The humanoid robot motion control method based on motion data analysis according to claim 1, characterized in that: When obtaining multimodal motion data of the robot during motion, the multimodal motion data is obtained by installing an inertial measurement unit, a force sensor and an angle encoder on each joint of the humanoid robot, and the multimodal motion data includes joint angle, angular velocity, acceleration, joint torque and end effector force; Perform time alignment and interpolation on multimodal motion data to generate a standardized time data series.

3. The humanoid robot motion control method based on motion data analysis according to claim 1, characterized in that: The energy consumption estimation is obtained by using the torque data deviation and the angular velocity data deviation in combination with a basic energy consumption estimation model, including: Retrieve all torque data deviations and angular velocity data deviations obtained by monitoring the current joint point; Performing standardization processing on the torque data deviation and the angular velocity data deviation to obtain the torque data deviation and the angular velocity data deviation after the standardization processing; Integrate all torque data deviations after standardization to generate a torque data deviation data set; Obtaining a torque deviation ratio average value according to each standardized torque data deviation in the torque data deviation data set; Integrate all the angular velocity data deviations after the standardization process to generate an angular velocity data deviation data set; Obtaining an angular velocity deviation ratio average value according to each standardized angular velocity data deviation in the angular velocity data deviation data set; Obtaining an energy loss coefficient according to the average value of the torque deviation ratio and the average value of the angular velocity deviation ratio; The energy consumption loss coefficient is combined with a basic energy consumption estimation model to obtain an energy consumption estimation.

4. The humanoid robot motion control method based on motion data analysis according to claim 1, characterized in that: When constructing the robot motion model, the key parameters change rules under each motion state are analyzed based on the extracted key motion features; Constructing a model of the angle, angular velocity, acceleration and torque changes of each joint under different motion states, wherein the change model is used to describe the dynamic change trend of the joint over time; According to the complete motion process of the robot from starting to stopping, the transfer relationship between different motion states is analyzed based on the time data series, and a robot motion state transfer model is established through a state machine modeling method. The motion state transfer model is used to define the transfer conditions and transition modes of different motion states; The mutual influence relationship of joint motion under different motion states is analyzed, which includes the synergy between adjacent joints, inertial influence and interference factors of the external environment on the robot motion, and a state dependence model is established based on the multivariate regression method.

5. The humanoid robot motion control method based on motion data analysis according to claim 4, characterized in that: When constructing the robot motion model, the Lagrangian method is used to establish the robot motion equation, which is used to describe the dynamic characteristics of each joint of the robot in different states. By analyzing the relationship between joint torque, speed, acceleration and energy consumption, the robot motion model of the robot in different motion states is established; Combined with the actual operation data of the robot, the robot motion model parameters are optimized in a data-driven way, and the robot motion model is trained through machine learning; Combined with the robot's historical motion data, the constructed robot motion model is simulated and tested and error analyzed. The error between the model prediction value and the actual measurement value is calculated, and the model is optimized through parameter adjustment.

6. The humanoid robot motion control method based on motion data analysis according to claim 1, characterized in that: When generating and optimizing the robot motion trajectory, based on the robot motion model, combined with the current motion state and the target state, the key parameters of trajectory generation are determined, including joint angles, angular velocities, accelerations and path point information, and a trajectory parameterization model is established; Generate discrete trajectory points between the robot's current state and the target state; Based on the robot motion state transfer model, the motion time allocation of different trajectory segments is calculated, and the motion time allocation includes the time proportion of acceleration, uniform speed and deceleration stages; For the initially generated motion trajectory, constraints are set in combination with the robot's structural characteristics and task requirements. The constraints include joint angle range constraints, maximum speed constraints, maximum acceleration constraints, and joint torque constraints. The motion trajectory is optimized and adjusted using constraint optimization. Based on the energy consumption estimation in the robot motion model, the energy consumption of different motion trajectories is calculated, the motion trajectory is optimized through the energy optimal control method, and the optimized motion trajectory is smoothed.

7. The humanoid robot motion control method based on motion data analysis according to claim 1, characterized in that: When generating corresponding control instructions according to the motion trajectory, based on the optimized motion trajectory, the target state of each joint of the robot is extracted, and the target state includes angle, angular velocity and acceleration parameters. Combined with the trajectory time allocation information, the expected motion state of each joint at different time steps is determined; Based on the robot motion model, calculate the joint torque and control torque required for each joint of the robot to perform the target action, and generate the target joint torque distribution; The calculated joint target state, joint torque and control torque are converted into standardized control signals, and the control signals are sent to the robot joint drive unit through a real-time control bus.

8. The humanoid robot motion control method based on motion data analysis according to claim 1, characterized in that: When calculating the error deviation and dynamically adjusting the control instruction based on the error deviation, the motion state data of each joint during the robot's motion is collected in real time, and the motion state data includes the joint angle, angular velocity, acceleration, joint torque and the force condition of the end effector; Based on the optimized motion trajectory, the robot's current joint angle data is compared with the target angle in the preset trajectory to calculate the angle error of each joint; Combined with the robot motion model, the actual trajectory of the robot end effector is calculated, and compared with the preset trajectory to calculate the trajectory deviation, which includes the spatial position deviation and the direction deviation; Based on the joint angle error and trajectory deviation, the error compensation factor is generated in combination with the robot motion model. The control signal of each joint is adjusted based on the error compensation factor to correct the target joint angle, angular velocity and control torque in real time.

Images