CN-122004030-A - Robot control method, robot, and readable storage medium

CN122004030ACN 122004030 ACN122004030 ACN 122004030ACN-122004030-A

Abstract

The application relates to a robot control method, a robot and a readable storage medium, wherein the method comprises the steps of obtaining a time sequence looking-around image of the current environment of the robot and state information of the current robot; the time sequence looking-around image comprises a plurality of frames of looking-around images of the current environment of the robot and time information of the plurality of frames of looking-around images, the time sequence looking-around images and the state information are input into a pre-trained action control model to obtain action parameters of the robot, a control instruction is generated according to the action parameters, and the movement of the robot is controlled according to the control instruction. The robot control method provided by the application can realize accurate control of the robot.

Inventors

LU ZHILIN
WAN LEI
ZHANG YUAN
ZHOU RUNNAN
LI QIDI
HAN BINGBING
XU YIXIN

Assignees

安克创新科技股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260326

Claims (11)

1. A method of controlling a robot, the method comprising: Acquiring a time sequence looking-around image of the current environment of the robot and the state information of the robot, wherein the time sequence looking-around image comprises a multi-frame looking-around image of the current environment of the robot and time information of the multi-frame looking-around image; Inputting the time sequence looking-around image and the state information into a pre-trained motion control model to obtain motion parameters of the robot; And generating a control instruction according to the action parameter, and controlling the robot to move according to the control instruction.
2. The method of claim 1, wherein the motion control model comprises a feature extraction network, a feature stitching network, a time series modeling network, and a motion output network connected in sequence, the inputting the time series looking-around image and the state information into a pre-trained motion control model, obtaining motion parameters of the robot, comprising: extracting features of the multi-frame looking-around image through the feature extraction network to obtain image features and time sequence features, wherein the image features comprise semantic features representing boundary lines, barriers, covered areas and uncovered areas; Performing feature stitching on the image features, the time sequence features and the state information through the feature stitching network to obtain environment feature representation; information fusion is carried out on the environmental characteristic representation through the time sequence modeling network, and a high-dimensional state representation is obtained; mapping the high-dimensional state representation to the action parameters through the action output network.
3. The method according to claim 1, wherein the method further comprises: Acquiring a moving path and a coverage area of the robot based on the control instruction; Based on a reinforcement learning algorithm, updating network parameters of the motion control model according to the time sequence looking-around image, the state information, the motion parameters of the robot, the moving path and the coverage area.
4. The method of claim 3, wherein updating the network parameters of the motion control model based on the time series look-around image, the state information, the motion parameters of the robot, and the movement path and the coverage area based on a reinforcement learning algorithm comprises: Acquiring a historical moving path of the robot and a corresponding historical coverage area; Constructing a new training sample according to the time sequence looking-around image, the state information, the action parameters of the robot, the moving path, the coverage area, the historical moving path and the historical coverage area; And updating the network parameters of the action control model according to the new training sample.
5. The method of claim 4, wherein the new training samples include a current state, an action, a prize value, and a next state, and wherein constructing new training samples based on the time series look-around image, the state information, the action parameters of the robot, the movement path and the coverage area, and the historical movement path and the historical coverage area comprises: determining the current state of the robot according to the time sequence looking-around image and the state information; determining the action of the robot according to the action parameters of the robot; Calculating a reward value based on a preset reward function according to the moving path and the coverage area and the historical moving path and the historical coverage area; and determining the next state of the robot according to the state of the robot after the robot moves based on the control instruction.
6. The method of claim 5, wherein the reward function comprises a positive reward including a newly added coverage area and a negative reward including at least one of a total path length penalty, a collision event penalty, a boundary event penalty, and a path overlap length penalty.
7. The method of claim 6, wherein said calculating a prize value based on a predetermined prize function based on said movement path and coverage area and said historical movement path and historical coverage area comprises: determining the newly added coverage area according to the coverage area and the historical coverage area; Determining the total length of the path and the path overlap length according to the moving path and the historical moving path; Determining collision events and boundary events according to the moving path; substituting the newly added coverage area, the total path length, the path overlapping length, the collision event and the boundary event into the reward function to obtain the reward value.
8. The method of claim 1, wherein the acquiring a time-series look-around image of the environment in which the robot is currently located comprises: acquiring a multi-frame looking-around image of the current environment of the robot; and stacking the multi-frame looking-around images in the time dimension to obtain the time sequence looking-around image.
9. The method of any one of claims 1-8, wherein the training process of the motion control model comprises: Obtaining a training sample; The method comprises the steps of training an initial network model according to a training sample to obtain an action control model, wherein the initial network model comprises a feature extraction network, a feature splicing network, a time sequence modeling network, an action output network and a value estimation network, the feature extraction network, the feature splicing network, the time sequence modeling network and the action output network are sequentially connected, and the time sequence modeling network is also connected with the value estimation network.
10. The robot is characterized by comprising a robot body, an image acquisition assembly, a state acquisition assembly and a control assembly, wherein the image acquisition assembly, the state acquisition assembly and the control assembly are arranged on the robot body, and the control assembly is connected with the image acquisition assembly and the state acquisition assembly; The system comprises an image acquisition component, a control component and a control component, wherein the image acquisition component is used for acquiring a time sequence looking-around image of the current environment of the robot body and transmitting the time sequence looking-around image to the control component; The state acquisition component is used for acquiring the current state information of the robot body and transmitting the state information to the control component; The control component is used for inputting the time sequence looking-around image and the state information into a pre-trained motion control model to obtain motion parameters of the robot body, generating a control instruction according to the motion parameters and controlling the robot body to move according to the control instruction.
11. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 9.

Description

Robot control method, robot, and readable storage medium Technical Field The present application relates to the field of robotics, and in particular, to a robot control method, a robot, and a readable storage medium. Background Along with the increasing importance of the national ecological civilization construction, from road greening to district parks, from outdoor leisure to entertainment competition, the lawn is ubiquitous, and has great development potential for lawn maintenance and beautification due to wide market and clear requirements. The intelligent mowing robot can realize outdoor autonomous unmanned mowing, improves efficiency, saves time and labor cost and is widely applied. Existing lawn mowing robots rely on a "sense-map-plan-control" serial path in lawn-covering operations. However, there is information loss and error accumulation between perception, mapping and planning, which can lead to inaccurate control of the robot. Disclosure of Invention In view of the above, it is necessary to provide a robot control method, a robot, and a readable storage medium that can improve the control accuracy of the robot. In a first aspect, the present application provides a robot control method, the method comprising: Acquiring a time sequence looking-around image of the current environment of the robot and state information of the current robot, wherein the time sequence looking-around image comprises a multi-frame looking-around image of the current environment of the robot and time information of the multi-frame looking-around image; Inputting the time sequence looking-around image and the state information into a pre-trained motion control model to obtain motion parameters of the robot; and generating a control instruction according to the action parameter, and controlling the robot to move according to the control instruction. In one embodiment, the motion control model includes a feature extraction network, a feature stitching network, a time sequence modeling network, and a motion output network that are sequentially connected, and the time sequence looking-around image and the state information are input into the pre-trained motion control model to obtain motion parameters of the robot, including: Extracting features of the multi-frame looking-around image through a feature extraction network to obtain image features and time sequence features, wherein the image features comprise semantic features representing boundary lines, barriers, covered areas and uncovered areas; Performing feature stitching on the image features, the time sequence features and the state information through a feature stitching network to obtain environment feature representation; information fusion is carried out on the environmental characteristic representation through a time sequence modeling network, and a high-dimensional state representation is obtained; the high-dimensional state representation is mapped to an action parameter through an action output network. In one embodiment, the method further comprises: acquiring a moving path and a coverage area of the robot based on a control instruction; Based on the reinforcement learning algorithm, the network parameters of the motion control model are updated according to the time sequence looking-around image, the state information, the motion parameters of the robot, the moving path and the coverage area. In one embodiment, updating network parameters of the motion control model according to the time sequence looking-around image, the state information, the motion parameters of the robot, and the moving path and the coverage area based on the reinforcement learning algorithm comprises: Acquiring a historical moving path of the robot and a corresponding historical coverage area; constructing a new training sample according to the time sequence looking around image, the state information, the action parameters, the moving path and the coverage area of the robot, and the historical moving path and the historical coverage area; and updating the network parameters of the motion control model according to the new training sample. In one embodiment, the new training samples comprise a current state, an action, a reward value and a next state, and constructing the new training samples according to the time sequence looking-around image, the action parameters, the moving path and the coverage area of the state information robot, and the historical moving path and the historical coverage area comprises the following steps: Determining the current state of the robot according to the time sequence looking around image and the state information; Determining the action of the robot according to the action parameters of the robot; Calculating a reward value based on a preset reward function according to the moving path and the coverage area and the historical moving path and the historical coverage area; and determining the next state of the robot according to the state of the robot af