Search

CN-121989237-A - Robot control method and device based on linear velocity, electronic equipment and storage medium

CN121989237ACN 121989237 ACN121989237 ACN 121989237ACN-121989237-A

Abstract

The invention discloses a robot control method and device based on linear velocity, electronic equipment and a storage medium. The method comprises the steps of carrying out gesture prediction and action simulation on first data through a first model to obtain second data, processing the second data through the second model to obtain third data, carrying out linear velocity estimation on the third data to obtain fourth data, and controlling the robot according to the fourth data. According to the method, the second data are processed through the second model, real noise data in the second data can be effectively removed, and third data are obtained. The linear velocity estimation is performed on the third data, so that the equal stable linear velocity output can be obtained, and the control precision of the robot is improved.

Inventors

  • YANG ZHUANG

Assignees

  • 杭州软通天擎机器人科技有限公司

Dates

Publication Date
20260508
Application Date
20260130

Claims (10)

  1. 1. A robot control method based on a linear velocity, comprising: the method comprises the steps of carrying out gesture prediction and action simulation on first data through a first model to obtain second data, wherein the first data are used for representing motion change of a robot, the first model is provided with a reward layer, the first gesture data of the reward layer are determined and used for correcting the first model, and the first gesture data are gesture data obtained by predicting the first model according to the first data; The second data is processed through a second model to obtain third data, the second model is provided with a noise adding layer and a reverse denoising layer, the noise adding layer is used for adding noise to the first data according to a preset step length, and the reverse denoising layer is used for recovering the first data added with the noise; Performing linear velocity estimation on the third data to obtain fourth data; And controlling the robot according to the fourth data.
  2. 2. The method of claim 1, wherein predicting the pose of the first data by the first model to obtain the second data comprises: determining first data, and weighting the first data according to preset weights to obtain second posture data; fitting the second gesture data through an activation function to obtain first gesture data; And simulating the running state of the robot according to the first gesture data to obtain second data.
  3. 3. The method of claim 1, wherein the at least one manner of determining the prize layer comprises: Calculating norm square and index for a first linear velocity and a second linear velocity to obtain a first reward function, wherein the first linear velocity is the linear velocity of a base in first posture data, the second linear velocity is the linear velocity of the base in third posture data, and the third posture data is posture change data of the robot obtained after simulation according to the first posture data; calculating norm square and index for a first angular velocity and a second angular velocity to obtain a second rewarding function, wherein the first angular velocity is yaw angular velocity in the first gesture data; Calculating a norm square and an index for a third line speed to obtain a third rewarding function, wherein the third line speed is the foot line speed in the third gesture data; Calculating norm square and index for a third angular velocity and a first moment to obtain a fourth rewarding function, wherein the third angular velocity is the joint angular velocity in the third posture data; generating a reward layer according to the first reward function, the second reward function, the third reward function and the fourth reward function.
  4. 4. The method of claim 1, wherein processing the second data with the second model to obtain third data comprises: Adding preset noise to the second data according to a preset time step to obtain first noise data, wherein the preset noise is used for simulating the actual distortion characteristics of the robot; Determining a first step length, and matching corresponding noise data from the first noise data according to the first step length to obtain second noise data, wherein the first step length is the addition frequency selected from any one of preset addition frequencies; And predicting the second noise data, and updating the first noise data according to a prediction result to obtain third data.
  5. 5. The method of claim 4, wherein predicting the second noise data, updating the first noise data based on the prediction, and obtaining third data, comprises: performing noise prediction according to the second noise data to obtain first noise; Generating third noise data according to the second noise data and the first noise, wherein the third noise data is noise data corresponding to a second step length generated according to the first noise, and the second step length is the previous adding frequency of the first step length; And updating the first noise data according to the third noise data to obtain third data.
  6. 6. The method of claim 1, wherein the performing linear velocity estimation on the third data to obtain fourth data includes: Calculating the third data to obtain a first matrix; converting the third data according to the first matrix to obtain fifth data; and integrating the fifth data to obtain fourth data.
  7. 7. A robot control device based on linear velocity, comprising: The system comprises a first data determining module, a first model, a second data determining module, a first data processing module and a second data processing module, wherein the first data is used for carrying out gesture prediction and action simulation on first data through the first model to obtain second data, the first data is used for representing motion change of a robot, the first model is configured with a reward layer, the first gesture data of the reward layer is determined and used for correcting the first model, and the first gesture data is gesture data obtained by predicting the first model according to the first data; The system comprises a first data determining module, a second data adding module, a third data determining module and a first data processing module, wherein the first data determining module is used for processing second data through a second model to obtain first data, the second model is configured with a noise adding layer and a reverse denoising layer, the noise adding layer is used for adding noise to first data according to a preset step length, and the reverse denoising layer is used for recovering the first data added with the noise; The fourth data determining module is used for estimating the linear velocity of the third data to obtain fourth data; and the control module is used for controlling the robot according to the fourth data.
  8. 8. The apparatus of claim 7, wherein the third data determination module comprises: the first noise data determining unit is used for adding preset noise to the second data according to a preset time step to obtain first noise data, wherein the preset noise is used for simulating the actual distortion characteristics of the robot; The second noise data determining unit is used for determining a first step length, and matching corresponding noise data from the first noise data according to the first step length to obtain second noise data, wherein the first step length is the addition frequency selected from any one of the preset addition frequencies; and the third data determining unit is used for predicting the second noise data and updating the first noise data according to a prediction result to obtain third data.
  9. 9. An electronic device, the electronic device comprising: At least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the linear velocity based robot control method of any one of claims 1-6.
  10. 10. A computer readable storage medium, characterized in that the computer readable storage medium stores computer instructions for causing a processor to implement the linear velocity based robot control method of any one of claims 1-6 when executed.

Description

Robot control method and device based on linear velocity, electronic equipment and storage medium Technical Field The present invention relates to the field of robot control technologies, and in particular, to a method and apparatus for controlling a robot based on linear velocity, an electronic device, and a storage medium. Background The linear speed is one of the most critical observables for describing the dynamic state of the organism, and in the existing bipedal robot control system, the linear speed is often only estimated as an external module, is independent of the control strategy training, and lacks a unified optimization and collaborative learning mechanism. Because of the lack of a sensor capable of directly measuring the linear velocity, various control frameworks generally depend on independent velocity estimation modules (such as IMU integration, filtering fusion and the like), and the estimation result cannot be ensured to be consistent with the dynamic modeling in the strategy, so that the problems of unstable performance, insufficient generalization and the like of the whole system are caused. The existing linear velocity estimation method mainly depends on IMU integral, complementary filtering or EKF and other traditional filters, and the method is extremely susceptible to noise, offset drift and impact interference on a real bipedal robot, and has obviously limited output stability and accuracy. Since the velocity estimation module typically exists as a stand-alone component, its noise characteristics can directly impact the training quality of the downstream strategy network, resulting in the strategy receiving noise-dominated, biased velocity observations during the training phase. The strategy obtained in this way is often highly sensitive to speed errors during actual deployment, and phenomena such as oscillation, gait instability, insufficient speed drift compensation or slow response to external disturbance are easy to occur. Disclosure of Invention The invention provides a robot control method, a robot control device, electronic equipment and a storage medium based on linear speed, which are used for solving the problem that the action precision caused by the fact that a bipedal robot strategy is highly sensitive to speed errors does not meet the requirements. According to an aspect of the present invention, there is provided a robot control method based on a linear velocity, including: the method comprises the steps of carrying out gesture prediction and action simulation on first data through a first model to obtain second data, wherein the first data are used for representing motion change of a robot, the first model is provided with a reward layer, the first gesture data of the reward layer are determined and used for correcting the first model, and the first gesture data are gesture data obtained by predicting the first model according to the first data; The second data is processed through a second model to obtain third data, the second model is provided with a noise adding layer and a reverse denoising layer, the noise adding layer is used for adding noise to the first data according to a preset step length, and the reverse denoising layer is used for recovering the first data added with the noise; Performing linear velocity estimation on the third data to obtain fourth data; And controlling the robot according to the fourth data. According to another aspect of the present invention, there is provided a robot control device based on a linear velocity, including: The system comprises a first data determining module, a first model, a second data determining module, a first data processing module and a second data processing module, wherein the first data is used for carrying out gesture prediction and action simulation on first data through the first model to obtain second data, the first data is used for representing motion change of a robot, the first model is configured with a reward layer, the first gesture data of the reward layer is determined and used for correcting the first model, and the first gesture data is gesture data obtained by predicting the first model according to the first data; The system comprises a first data determining module, a second data adding module, a third data determining module and a first data processing module, wherein the first data determining module is used for processing second data through a second model to obtain first data, the second model is configured with a noise adding layer and a reverse denoising layer, the noise adding layer is used for adding noise to first data according to a preset step length, and the reverse denoising layer is used for recovering the first data added with the noise; The fourth data determining module is used for estimating the linear velocity of the third data to obtain fourth data; and the control module is used for controlling the robot according to the fourth data. According to another aspect of the