Motion precision control system for four limbs of humanoid robot

Abstract

The invention discloses a humanoid robot four-limb movement precision control system, and relates to the technical field of robot control. Comprising a data acquisition module, the data acquisition module is in butt joint with a data processing module, and the data acquisition module is used for acquiring multi-dimensional original data of the humanoid robot and sending the multi-dimensional original data to the data processing module; the data processing module is in butt joint with the vibration suppression module, and the data processing module is used for conducting finite element analysis and neural network processing on the multi-dimensional original data and the motion planning parameters, obtaining vibration pre-judgment information and sending the vibration pre-judgment information to the vibration suppression module; and the vibration suppression module is used for generating a vibration suppression mode according to the vibration pre-judgment information and adjusting motion planning parameters of the humanoid robot according to the vibration suppression mode. According to the method, the motion planning parameters of the humanoid robot can be inhibited in advance according to the vibration pre-judgment information, and high-precision motion control can be completed without depending on subsequent precision compensation.

Description

Technical Field

[0001] This invention relates to the field of robot control technology, specifically to a precision control system for the limb movements of a humanoid robot. Background Technology

[0002] As humanoid robots develop towards lightweight and high flexibility, their limb links and joints are mostly made of low-stiffness materials such as carbon fiber and aluminum alloy. Although these materials can reduce the robot's weight and improve its movement flexibility, they are easily affected by external excitations such as load force and ground reaction force under typical working conditions such as high-speed grasping, rotation operation, and walking, resulting in elastic vibration. This leads to deviations in joint rotation angles and end effector position, affecting movement accuracy.

[0003] The control of the limb movement accuracy of existing humanoid robots usually adopts the method of post-accident accuracy compensation, which has obvious response lag. Summary of the Invention

[0004] In view of the deficiencies in the existing technology, the technical problem to be solved by this application is: how to predict the vibration parameters during the movement of a humanoid robot, and improve the limb movement accuracy of the humanoid robot based on the predicted vibration parameters.

[0005] To achieve the above objectives, this application provides a humanoid robot limb motion precision control system, comprising:

[0006] A data acquisition module is connected to a data processing module. The data acquisition module is used to collect multi-dimensional raw data of the humanoid robot and send the multi-dimensional raw data to the data processing module.

[0007] The data processing module is connected to the vibration suppression module. The data processing module is used to perform finite element analysis and neural network processing on multi-dimensional raw data and motion planning parameters to obtain vibration prediction information and send it to the vibration suppression module.

[0008] The vibration suppression module is used to generate a vibration suppression mode based on vibration prediction information and adjust the motion planning parameters of the humanoid robot according to the vibration suppression mode.

[0009] In one embodiment, the multi-dimensional raw data includes vibration acceleration signals, elastic strain signals, and actual joint rotation angle signals.

[0010] In one embodiment, the data processing module includes a data preprocessing unit, an execution scenario determination unit, a finite element analysis unit, and a neural network processing unit;

[0011] The data preprocessing unit is connected to the execution scenario determination unit, the execution scenario determination unit is connected to the finite element analysis unit, and the finite element analysis unit is connected to the neural network processing unit.

[0012] In one embodiment, the data preprocessing unit is used to filter, amplify, and reduce noise in the multi-dimensional raw data to obtain preprocessed multi-dimensional raw data.

[0013] In one embodiment, the execution scenario determination unit is used to determine the execution scenario based on the preprocessed multi-dimensional raw data and motion planning parameters. The determination logic for the execution scenario includes:

[0014] A high-speed upper limb grasping scenario is determined when the upper limb movement speed is ≥0.3m / s and the load is 0.1-1kg.

[0015] A high-speed walking scenario is defined as a lower limb movement speed ≥1.0m / s.

[0016] An upper limb rotation operation scenario is determined when the upper limb joint rotation angle is ≥180° and the angular velocity is ≥2rad / s.

[0017] In one embodiment, the finite element analysis unit is used to construct stiffness-vibration coupled mathematical models under different execution scenario combinations based on material property parameters and geometric feature parameters, and to obtain initial vibration prediction information based on preprocessed multi-dimensional raw data and motion planning parameters; the initial vibration prediction information includes stress and strain distribution prediction data and vibration response characteristic prediction data.

[0018] In one embodiment, the neural network processing unit is used to construct a BP neural network model, taking initial vibration prediction information, preprocessed multi-dimensional raw data and motion planning parameters as input, and outputting final vibration prediction information; the vibration prediction information includes vibration amplitude prediction value, decay time prediction value and angle hysteresis deviation prediction value.

[0019] In one embodiment, the process by which the vibration suppression module generates a vibration suppression method based on vibration prediction information includes:

[0020] When a high-speed upper limb grasping scenario occurs, based on the initial torque, the torque correction coefficient and velocity change rate are dynamically adjusted in combination with the real-time vibration amplitude during the movement, so that the real-time vibration amplitude is less than or equal to the maximum allowable vibration amplitude.

[0021] When high-speed walking scenarios occur, the driving torque of the lower limb joints is determined based on the basic driving torque and the compensation torque for vibration suppression in high-speed walking scenarios;

[0022] When upper limb rotation operation scenarios occur, the real-time correction torque is determined based on the torsional damping torque and the basic torque of joint rotation, so that the real-time torsional vibration amplitude is less than or equal to the target threshold.

[0023] In one embodiment, when a high-speed grasping scenario of the upper limb occurs and the predicted vibration amplitude is greater than a preset threshold, the damping torque of the high-speed grasping scenario of the upper limb is superimposed on the correction torque.

[0024] In one embodiment, when an upper limb rotation operation scenario occurs, and the actual rotation angle deviation of the joint in the upper limb rotation operation scenario exceeds a preset threshold, a fine-tuning torque is superimposed on the real-time correction torque. The calculation formula for this fine-tuning torque is:

[0025] ;

[0026] In the formula, To fine-tune the torque, This is the angular deviation compensation coefficient for upper limb rotation operation scenarios, and , This represents the deviation of the actual rotation angle of the joint.

[0027] Compared with the prior art, the advantages of this application are:

[0028] While acquiring vibration prediction information, it can suppress the motion planning parameters of the humanoid robot in advance based on the vibration prediction information, thereby avoiding vibration causes in advance and achieving high-precision motion control without relying on subsequent accuracy compensation.

[0029] By determining the execution scenario, a targeted adaptation basis is provided for subsequent vibration suppression methods. Furthermore, the initial vibration prediction information at the structural level is obtained through a stiffness-vibration coupling mathematical model, and the final accurate vibration prediction information is then optimized and output through a BP neural network model. This dual processing mechanism can significantly improve the accuracy and timeliness of vibration prediction.

[0030] Different vibration suppression methods are configured for different execution scenarios, and motion planning parameters can be adjusted in real time during the suppression process to suppress vibration generation from the source. This solves the response lag problem of traditional post-compensation and provides reliable support for improving the motion accuracy of humanoid robot limbs. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1This is a schematic diagram of the architecture of the humanoid robot's limb motion precision control system in this application;

[0033] Figure 2 This is a schematic diagram illustrating the usage process of the humanoid robot limb motion precision control system in this application. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0035] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0036] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0037] Reference Figure 1 This application provides a humanoid robot limb motion precision control system, including: a data acquisition module, a data processing module, and a vibration suppression module. The data acquisition module is connected to the data processing module, and the data processing module is connected to the vibration suppression module. The data acquisition module is used to collect multi-dimensional raw data of the humanoid robot and send the multi-dimensional raw data to the data processing module. After receiving the multi-dimensional raw data, the data processing module performs finite element analysis and neural network processing on the multi-dimensional raw data and motion planning parameters (control parameters generated by the humanoid robot before action according to the target requirements, so that the humanoid robot adjusts the joint driving torque and motion parameters according to the control parameters to complete the target requirements) to obtain vibration prediction information and send it to the vibration suppression module. After receiving the vibration prediction information, the vibration suppression module generates a vibration suppression mode according to the vibration prediction information and adjusts the motion planning parameters of the humanoid robot according to the vibration suppression mode.

[0038] This allows for the acquisition of vibration prediction information, while simultaneously suppressing the motion planning parameters of the humanoid robot in advance based on the vibration prediction information, thus enabling early avoidance of vibration-induced factors and achieving high-precision motion control without relying on subsequent accuracy compensation.

[0039] In one embodiment, the process of the data acquisition module acquiring multi-dimensional raw data includes:

[0040] Sensor arrays (including MEMS accelerometers, fiber optic strain sensors, and angle encoders) are deployed at the midpoints of the links in the limbs of the humanoid robot, at the joint connections, and at the end effector.

[0041] Vibration acceleration signals during the movement of the humanoid robot's limbs are collected using MEMS accelerometers (sampling frequency not less than 1kHz), elastic strain signals of the linkage structure are collected using fiber optic strain sensors (resolution at least 0.1με), and actual joint rotation angle signals are collected using angle encoders.

[0042] By deploying multiple types of sensors in key parts of the limbs, multi-dimensional raw data such as vibration acceleration, elastic strain, and joint rotation angle can be collected synchronously, providing comprehensive and reliable data source support for subsequent vibration prediction and avoiding prediction errors caused by a single data dimension.

[0043] In one embodiment, the data processing module includes a data preprocessing unit, an execution scenario determination unit, a finite element analysis unit, and a neural network processing unit; the data preprocessing unit is connected to the execution scenario determination unit, the execution scenario determination unit is connected to the finite element analysis unit, and the finite element analysis unit is connected to the neural network processing unit.

[0044] Based on this, the data processing module's process for obtaining vibration prediction information from multi-dimensional raw data includes:

[0045] After receiving the multi-dimensional raw data, the data preprocessing unit performs filtering, amplification and noise reduction on the multi-dimensional raw data to obtain the preprocessed multi-dimensional raw data, and sends the preprocessed multi-dimensional raw data to the execution scenario determination unit.

[0046] After receiving the preprocessed multi-dimensional raw data, the execution scenario determination unit determines the execution scenario by combining it with motion planning parameters. The determination logic for the execution scenario includes:

[0047] When the upper limb movement speed is ≥0.3m / s and the load (the load carried by the end effector) is 0.1-1kg, a high-speed upper limb grasping scenario is determined to occur.

[0048] A high-speed walking scenario is defined as a lower limb movement speed ≥1.0m / s.

[0049] When the upper limb joint rotation angle is ≥180° and the angular velocity is ≥2rad / s, an upper limb rotation operation scenario is determined to have occurred.

[0050] The finite element analysis unit serves as the stiffness-vibration coupled mathematical model. The preprocessed multi-dimensional raw data and motion planning parameters are input into the stiffness-vibration coupled mathematical model, and the stiffness-vibration coupled mathematical model outputs the initial vibration prediction information: stress and strain distribution prediction data and vibration response characteristic prediction data.

[0051] The neural network processing unit, as a BP neural network model, takes the initial vibration prediction information, the preprocessed multi-dimensional raw data, and motion planning parameters as inputs, and outputs the final vibration prediction information: vibration amplitude prediction value, decay time prediction value, and angle hysteresis deviation prediction value.

[0052] In this embodiment, by determining the execution scenario, a targeted adaptation basis is provided for subsequent vibration suppression methods. Furthermore, the initial vibration prediction information at the structural level is obtained through a stiffness-vibration coupling mathematical model, and then the final accurate vibration prediction information is optimized and output through a BP neural network model. This dual processing mechanism can significantly improve the accuracy and timeliness of vibration prediction.

[0053] In one embodiment, the process of constructing stiffness-vibration coupling mathematical models for different combinations of execution scenarios based on material property parameters and geometric feature parameters includes:

[0054] S1. After standardizing the material property parameters (such as elastic modulus, density, Poisson's ratio, etc.) and geometric feature parameters (such as link length, cross-sectional dimensions, etc.) of the humanoid robot's limbs, determine the initial geometric constraints (such as the initial value of the joint rotation angle) under the current motion posture of the structure by combining the actual joint rotation angle signal.

[0055] S2. Based on the standardized geometric feature parameters, the limbs of the humanoid robot are structured and discretized, and the limb links and joint structures are discretized into discrete elements that can be used for mechanical analysis; for example, the limb link structure (such as the upper arm and thigh links) is discretized into beam elements along the length direction; the joint connection is discretized into rigid and flexible connection elements (rigid represents the joint rotation core, and flexible represents the low stiffness characteristics of the joint).

[0056] Nodes are set at both ends (beam elements) or the center (rigid and flexible connection elements) of each discrete unit, and the nodes are numbered according to the sequence of structural motion posture;

[0057] S3. After calculating the element mass matrix, element damping matrix and element elastic stiffness matrix of each discrete element, according to the association between the node number and the discrete element, assemble the element mass matrix, element damping matrix and element elastic stiffness matrix of all discrete elements in the order of the overall degrees of freedom to form the overall structural mass matrix, overall damping matrix and overall elastic stiffness matrix.

[0058] S4. The vibration acceleration signal in the multi-dimensional raw data is converted into an inertial load, and the elastic strain signal in the multi-dimensional raw data is converted into an equivalent load. After applying the corresponding load according to the motion planning parameters under different execution scenario combinations, the external force vector of the discrete unit node is formed.

[0059] Based on the limb movement characteristics of humanoid robots, fixed boundaries (such as the connection between the torso and upper limbs, and the connection between the torso and lower limbs, constraining translational degrees of freedom) and motion boundaries (such as joint rotation, releasing rotational degrees of freedom, and constraining translational degrees of freedom in non-rotational directions) are set, and the actual joint rotation angle signal is substituted as the dynamic constraint condition of the motion boundary.

[0060] S5. Substituting the element mass matrix, element damping matrix, element elastic stiffness matrix, discrete element nodal force vectors, and dynamic constraint conditions of the moving boundary into the element dynamic equilibrium equation, the expression of the element dynamic equilibrium equation is:

[0061]

[0062] In the formula, The unit mass matrix, This refers to the nodal acceleration vector, i.e., the vibration acceleration signal. For the element damping matrix, The unit node velocity vector is obtained by integrating the vibration acceleration signal to reflect the vibration velocity. The element's elastic stiffness matrix, The nodal displacement vector is obtained through the elastic strain signal (i.e., strain = displacement gradient). For discrete element nodes, external force vectors. For time variables, This is used to identify discrete units.

[0063] Based on this, the methods for obtaining initial vibration prediction information include:

[0064] The Newmark-β method is used to solve the dynamic equilibrium equations of the element. The calculation process includes:

[0065] The time domain is discretized into several time steps according to the motion period (time step size ≤ 1 / 2 of the sensor sampling period).

[0066] Based on the initial time ( nodal displacements (=0) ,speed and acceleration Calculate the nodal displacement vector at each time step. velocity vector acceleration vector ;

[0067] Based on the displacement, velocity, and acceleration data of each node at each moment, combined with the element stiffness matrix and mass matrix, the stress (such as bending stress of the link element and shear stress of the joint element), strain (such as elastic strain and torsional strain), and vibration response characteristics (such as vibration amplitude and decay rate) of each discrete element are calculated.

[0068] After filtering, integrating, and standardizing the stress, strain, and vibration response characteristics of each discrete unit, initial vibration prediction data corresponding to different execution scenario combinations are obtained. Specifically:

[0069] Stress and strain distribution prediction data: Integrate the stress magnitude and strain distribution state of each discrete unit, mark the stress peak and strain peak of key parts such as the midpoint of the connecting rod and the joint connection, form stress and strain distribution cloud map data, and verify it in combination with the elastic strain signal in the multi-dimensional original data.

[0070] Vibration response characteristic prediction data: Extract the preliminary prediction value of vibration amplitude and vibration attenuation trend data of each discrete element (such as the preliminary prediction value of attenuation rate and the estimated attenuation time range).

[0071] In this embodiment, differentiated vibration suppression methods are configured for different execution scenarios. The motion planning parameters can be adjusted in real time during the suppression process to suppress vibration generation from the source. This solves the response lag problem of traditional post-compensation and provides reliable support for improving the motion accuracy of humanoid robot limbs.

[0072] In one embodiment, the process for outputting the final vibration prediction information includes:

[0073] The aforementioned BP neural network model adopts a three-layer architecture, consisting of an input layer, two hidden layers, and an output layer. The input layer has 10 nodes, which correspond to the initial vibration prediction data (peak stress, peak strain, preliminary prediction of vibration amplitude, and preliminary prediction of decay rate), preprocessed multi-dimensional raw data (vibration acceleration, elastic strain, and actual joint rotation angle), motion planning parameters (motion speed and load size), and execution scenario data.

[0074] The data input into the input layer is processed by two hidden layers and then output by the output layer. The output layer includes three nodes, which correspond to the predicted vibration amplitude, the predicted decay time, and the predicted angle hysteresis deviation, respectively.

[0075] The above execution scenario data is the execution scenario code, which uses binary encoding. From the least significant bit to the most significant bit, the scenarios are: high-speed upper limb grasping scenario, upper limb rotation operation scenario, and high-speed walking scenario. Bit=0 indicates that the scenario is not superimposed, and bit=1 indicates that the scenario is superimposed, as shown in Table 1.

[0076] Execution scenario binary Encoded value none 000 0 High-speed upper limb grasping scenario 001 1 Upper limb rotation operation scenario 010 2 High-speed walking scene 100 4 High-speed grasping scenarios with upper limbs and rotational manipulation scenarios with upper limbs 011 3 High-speed grasping and high-speed walking scenarios 101 5 Upper limb rotation operation scenario, high-speed walking scenario 110 6 High-speed grasping scenarios with upper limbs, high-speed rotational manipulation scenarios, and high-speed walking scenarios. 111 7

[0077] Table 1

[0078] The iterative training process of this BP neural network model includes:

[0079] Dataset determination: Select vibration response data of humanoid robots in three typical scenarios (high-speed grasping scenario, upper limb rotation operation scenario, and high-speed walking scenario) under multiple working conditions, and use the real-time perception results of the corresponding working conditions as labels. After standardization, these data are used as training samples. Each scenario contains 500 sets of valid samples (covering different speeds, loads, and postures, such as 0.1-1kg load and 0.3-0.8m / s upper limb movement speed in the grasping scenario).

[0080] Divide the dataset into training and testing sets according to a preset ratio (e.g., 8:2);

[0081] The training set is input into the input layer (with the activation function being the Linear function), processed by the activation function of the hidden layer (the Sigmoid function), and then processed by the activation function of the output layer (the Purelin linear function) to output vibration prediction information (predicted vibration amplitude, predicted decay time, and predicted angle hysteresis deviation).

[0082] The minimum mean square error between vibration prediction information and labels is used as the target loss function for iterative training, with an initial learning rate of 0.01 and optimization using the adaptive momentum gradient descent algorithm (Adam).

[0083] When the number of iterations reaches a preset number (e.g., 10,000 times) or the deviation between the vibration prediction information and the label is less than or equal to a preset threshold (e.g., 0.001), the iterated BP neural network model is obtained.

[0084] The accuracy of the BP neural network model after iteration is verified by inputting the test set. When the deviation between the vibration prediction information and the label is less than a preset threshold (such as 5%), it is determined that the accuracy requirement is met.

[0085] In one embodiment, the process of generating a vibration suppression method includes:

[0086] S1. In high-speed upper limb grasping scenarios, based on the initial torque, and combined with the real-time vibration amplitude during movement, the torque correction coefficient and velocity change rate are dynamically adjusted every 1ms to ensure that the real-time vibration amplitude is less than or equal to the maximum allowable vibration amplitude (determined by material stiffness and end effector positioning requirements). Specific vibration suppression methods include:

[0087] Motion parameter adjustment: An S-shaped acceleration / deceleration curve is used for segmented planning, dividing the acceleration / deceleration process into an initial acceleration phase, a constant speed phase, and a deceleration / stop phase. During the initial acceleration phase, the acceleration linearly increases from 0 to a preset maximum acceleration value (dynamically adjusted according to the load; for example, acceleration ≤ 0.5 m / s² when the load is 0.1-0.5 kg). 2 When the load is 0.5-1kg, the acceleration is ≤0.3m / s². 2 During the deceleration phase, the acceleration linearly decreases from the maximum acceleration value to 0 to avoid abrupt acceleration changes; the upper limb movement speed is constrained to ≤0.8m / s, and the rate of change of speed is ≤0.2m / s². 3 To reduce the impact of inertial forces;

[0088] When adjusting the joint drive torque, the torque correction coefficient is calculated using the following formula:

[0089] ;

[0090] In the formula, This is the torque correction factor. This is the predicted value for vibration amplitude. The maximum permissible vibration amplitude for high-speed upper limb grasping scenarios is preferably 0.5 mm;

[0091] The initial torque of joint actuation is calculated based on motion planning parameters, using the following formula:

[0092] ;

[0093] In the formula, For the initial torque, To capture load force, The length of the link. The moment of inertia of the upper limb joints. To capture the angular acceleration of the planning direction;

[0094] The real-time corrected torque is obtained based on the torque correction factor and the initial torque. The calculation formula is:

[0095] .

[0096] Furthermore, in Preset threshold (e.g.) At that time, the damping torque of the high-speed grasping scenario of the upper limb is superimposed on the real-time correction torque. The formula for calculation is:

[0097] ;

[0098] In the formula, This refers to the damping coefficient of the upper limb joints. This represents the real-time angular velocity of the upper limb joints.

[0099] By smoothly correcting the output torque of the motor, acceleration and deceleration vibrations are offset, reducing the impact of inertial forces on low-rigidity structures and suppressing vibration generation at its source.

[0100] S2. In high-speed walking scenarios, the driving torque of the lower limb joints is determined based on the basic driving torque and the compensation torque for vibration suppression in high-speed walking scenarios; specific vibration suppression methods include:

[0101] The basic driving torque is determined by the following formula:

[0102] ;

[0103] In the formula, Based on the driving torque, To bear the load of the lower limbs, This is the ground reaction force (collected and converted by MEMS sensors). The length of the lower limb link. This refers to the rotation angle of the lower limb joints;

[0104] The compensation torque for vibration suppression in high-speed walking scenarios is calculated using the following formula:

[0105] ;

[0106] In the formula, To compensate for the torque, The stiffness compensation coefficient is preferably 5 N·m / mm. This represents the real-time vibration amplitude during high-speed walking. The strain compensation coefficient for high-speed walking scenarios is preferably 0.1 N·m / με. Real-time elastic strain for high-speed walking scenarios; and the direction of the compensation torque is opposite to the direction of vibration (determined by the phase of the vibration acceleration signal).

[0107] The driving torque of the lower limb joints is determined based on the vibration damping compensation torque, and the calculation formula is as follows:

[0108] ;

[0109] In the formula, The value represents the driving torque of the lower limb joints, and is "-" for positive vibration and "+" for negative vibration.

[0110] By collecting ground reaction force and vibration signals in real time, calculating the compensation torque every 2ms and adjusting the joint output, the vibration decay time is ensured to be ≤50ms.

[0111] S3. When upper limb rotation occurs, a real-time correction torque is determined every 1 ms based on the torsional damping torque and the basic joint rotation torque, ensuring that the real-time torsional vibration amplitude is ≤ the target threshold (e.g., 0.2 mm). Specific vibration suppression methods include:

[0112] The rotation process is divided into an acceleration phase, a constant speed phase, and a deceleration and stop phase. During the constant speed phase, the angular velocity is ≤3 rad / s (to avoid torsional resonance). The angular velocity is dynamically adjusted based on the torsional strain signal (e.g., when the torsional strain signal is >0.5με, the angular velocity is reduced by 20%; when the torsional strain signal is >1με, acceleration is paused and the current angular velocity is maintained until the torsional strain signal drops back down). The rotation deceleration phase adopts a stepped deceleration, such as reducing the angular velocity by 0.2 rad / s every 10 ms, to avoid torsional vibration caused by sudden changes in angular velocity.

[0113] When adjusting the joint driving torque, a torsional damping coefficient is introduced, and its calculation formula is as follows:

[0114] ;

[0115] In the formula, The torsional damping coefficient is... The torsional strain signal is for an upper limb rotation operation scenario;

[0116] The torsional damping moment is determined based on the torsional damping coefficient, and the calculation formula is as follows:

[0117] ;

[0118] In the formula, For torsional damping moment, Real-time angular velocity of joints in upper limb rotation operation scenarios;

[0119] The formula for calculating the basic torque of joint rotation is as follows:

[0120] ;

[0121] In the formula, The basic torque for joint rotation. To meet the torsional torque requirements in upper limb rotation operation scenarios, The radius of the upper limb rotational joint;

[0122] Real-time corrected torque is obtained based on the joint rotational basic torque and torsional damping torque. The calculation formula is:

[0123] ;

[0124] Furthermore, when the actual joint rotation angle deviation in an upper limb rotation operation scenario exceeds a preset threshold (e.g., 0.5°), a fine-tuning torque is superimposed on the real-time correction torque. The formula for calculating this fine-tuning torque is:

[0125] ;

[0126] In the formula, To fine-tune the torque, This is the angular deviation compensation coefficient for upper limb rotation operation scenarios, and The unit is N·m / °. This represents the deviation of the actual rotation angle of the joint.

[0127] Based on this, and according to the judgment result of the execution scenario, the vibration suppression method corresponding to the execution scenario is invoked to adjust the motion planning parameters. In this way, by controlling the motion planning parameters in advance, the control accuracy of the humanoid robot's limb movements is effectively improved.

[0128] Reference Figure 2 , Figure 2 This is a schematic diagram illustrating the usage flow of the humanoid robot's limb motion precision control system in this application. Figure 2 As shown, the usage process includes:

[0129] S1. Collect multi-dimensional raw data of the humanoid robot: vibration acceleration signal, elastic strain signal and actual joint rotation angle signal;

[0130] S2. Filter, amplify and reduce noise on the multi-dimensional raw data to obtain the preprocessed multi-dimensional raw data;

[0131] S3. Determine the execution scenario based on the preprocessed multi-dimensional raw data and motion planning parameters (see the above-mentioned execution scenario determination logic).

[0132] S4. Input the preprocessed multi-dimensional raw data and motion planning parameters into the stiffness-vibration coupling mathematical model corresponding to the execution scenario determined in S2 (see the above process for constructing stiffness-vibration coupling mathematical models under different execution scenario combinations) to obtain initial vibration prediction information (see the above method for obtaining initial vibration prediction information).

[0133] S5. Input the initial vibration prediction information, the preprocessed multi-dimensional raw data and motion planning parameters into the BP neural network model to obtain the final vibration prediction information (see the above process for outputting the final vibration prediction information).

[0134] S6. Generate a vibration suppression method based on the vibration prediction information (see the above process for generating a vibration suppression method), and adjust the motion planning parameters of the humanoid robot according to the vibration suppression method.

[0135] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0136] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0137] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0138] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0139] The above are merely specific embodiments of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the scope of the claims.

Claims

1. A precision control system for the limbs of a humanoid robot, characterized in that, include: A data acquisition module is connected to a data processing module. The data acquisition module is used to collect multi-dimensional raw data of the humanoid robot and send the multi-dimensional raw data to the data processing module. The data processing module is connected to the vibration suppression module. The data processing module is used to perform finite element analysis and neural network processing on multi-dimensional raw data and motion planning parameters to obtain vibration prediction information and send it to the vibration suppression module. The vibration suppression module is used to generate a vibration suppression mode based on vibration prediction information and adjust the motion planning parameters of the humanoid robot according to the vibration suppression mode.

2. The humanoid robot limb motion precision control system according to claim 1, characterized in that, The multi-dimensional raw data includes vibration acceleration signals, elastic strain signals, and actual joint rotation angle signals.

3. The humanoid robot limb motion precision control system according to claim 2, characterized in that, The data processing module includes a data preprocessing unit, an execution scenario determination unit, a finite element analysis unit, and a neural network processing unit; The data preprocessing unit is connected to the execution scenario determination unit, the execution scenario determination unit is connected to the finite element analysis unit, and the finite element analysis unit is connected to the neural network processing unit.

4. The humanoid robot limb motion precision control system according to claim 3, characterized in that, The data preprocessing unit is used to filter, amplify, and reduce noise in the multi-dimensional raw data to obtain preprocessed multi-dimensional raw data.

5. A humanoid robot limb motion precision control system according to claim 4, characterized in that, The execution scenario determination unit is used to determine the execution scenario based on the preprocessed multi-dimensional raw data and motion planning parameters. The determination logic for the execution scenario includes: A high-speed upper limb grasping scenario is determined when the upper limb movement speed is ≥0.3m / s and the load is 0.1-1kg. A high-speed walking scenario is defined as a lower limb movement speed ≥1.0m / s. An upper limb rotation operation scenario is determined when the upper limb joint rotation angle is ≥180° and the angular velocity is ≥2rad / s.

6. A humanoid robot limb motion precision control system according to claim 5, characterized in that, The finite element analysis unit is used to construct stiffness-vibration coupled mathematical models under different execution scenario combinations based on material property parameters and geometric feature parameters, and to obtain initial vibration prediction information based on preprocessed multi-dimensional raw data and motion planning parameters; the initial vibration prediction information includes stress and strain distribution prediction data and vibration response characteristic prediction data.

7. A humanoid robot limb motion precision control system according to claim 6, characterized in that, The neural network processing unit is used to construct a BP neural network model, taking initial vibration prediction information, preprocessed multi-dimensional raw data and motion planning parameters as input, and outputting the final vibration prediction information; the vibration prediction information includes vibration amplitude prediction value, decay time prediction value and angle hysteresis deviation prediction value.

8. A humanoid robot limb motion precision control system according to claim 7, characterized in that, The process by which the vibration suppression module generates a vibration suppression method based on vibration prediction information includes: When a high-speed upper limb grasping scenario occurs, based on the initial torque, the torque correction coefficient and velocity change rate are dynamically adjusted in combination with the real-time vibration amplitude during the movement, so that the real-time vibration amplitude is less than or equal to the maximum allowable vibration amplitude. When high-speed walking scenarios occur, the driving torque of the lower limb joints is determined based on the basic driving torque and the compensation torque for vibration suppression in high-speed walking scenarios; When upper limb rotation operation scenarios occur, the real-time correction torque is determined based on the torsional damping torque and the basic torque of joint rotation, so that the real-time torsional vibration amplitude is less than or equal to the target threshold.

9. A humanoid robot limb motion precision control system according to claim 8, characterized in that, When a high-speed grasping scenario of the upper limb occurs and the predicted vibration amplitude is greater than the preset threshold, a damping torque of the high-speed grasping scenario of the upper limb is superimposed on the correction torque.

10. A humanoid robot limb motion precision control system according to claim 8, characterized in that, When an upper limb rotation operation scenario occurs, and the actual joint rotation angle deviation in the upper limb rotation operation scenario exceeds a preset threshold, a fine-tuning torque is superimposed on the real-time correction torque. The calculation formula for this fine-tuning torque is: ； In the formula, To fine-tune the torque, This is the angular deviation compensation coefficient for upper limb rotation operation scenarios, and , This represents the deviation of the actual rotation angle of the joint.

Images