CN-121999051-A - Robot repositioning method, apparatus, computer device and storage medium

CN121999051ACN 121999051 ACN121999051 ACN 121999051ACN-121999051-A

Abstract

The application relates to a robot repositioning method, a robot repositioning device, computer equipment and a storage medium. The method comprises the steps of obtaining at least one candidate initial pose of a robot at a restarting moment and current point cloud data of the robot at the restarting moment, primarily screening at least one reference initial pose according to the candidate initial poses and the current point cloud data, obtaining predicted poses of the robot after a preset planning action is executed on the basis of the reference initial poses and measured point cloud data of the robot after the preset planning action is executed according to the reference initial poses under the condition that the reference initial poses meet preset verification conditions, and finely screening target initial poses of the robot at the restarting moment from the at least one reference initial pose according to the measured point cloud data and the predicted poses corresponding to the reference initial poses. The method can accurately screen the initial pose of the target, and improves the repositioning accuracy to a certain extent.

Inventors

WU JIE
YANG ZONGLIN
WU YUPENG
KANG HANYU
MA YINGYING
YING XIANG

Assignees

重庆凤凰技术有限公司

Dates

Publication Date: 20260508
Application Date: 20260317

Claims (10)

1. A method of robotic repositioning, the method comprising: Acquiring at least one candidate initial pose of a robot at a restarting moment and current point cloud data of the robot at the restarting moment; According to each candidate initial pose and the current point cloud data, at least one reference initial pose is initially screened out from the at least one candidate initial pose; Under the condition that the reference initial pose meets a preset verification condition, acquiring a predicted pose of the robot after performing a preset planning action based on the reference initial pose and measuring point cloud data of the robot after performing the preset planning action according to each reference initial pose; And finely screening a target initial pose of the robot at the restarting time from the at least one reference initial pose according to the measured point cloud data and the predicted poses corresponding to the reference initial poses.
2. The method according to claim 1, wherein the fine screening the target initial pose of the robot at the restart time from the at least one reference initial pose according to the measured point cloud data and the predicted poses respectively corresponding to the reference initial poses comprises: for each reference initial pose, acquiring a first error of primarily screening out the reference initial pose; determining a second error of the reference initial pose according to the measurement point cloud data and the predicted pose corresponding to the reference initial pose; Determining a third error of the reference initial pose according to the planned measurement pose and the predicted pose corresponding to the reference initial pose, wherein the planned measurement pose corresponding to the reference initial pose is the measurement pose of the robot under the reference initial pose after the preset planning action is executed; and fine screening out the target initial pose of the robot at the restarting moment from the at least one reference initial pose according to the second error and/or the third error and the first error which are respectively corresponding to the reference initial poses.
3. The method of claim 2, wherein the determining the second error of the reference initial pose based on the measured point cloud data and the predicted pose corresponding to the reference initial pose comprises: Selecting a first local point cloud map from the global point cloud map of the area where the robot is located according to the predicted pose corresponding to the reference initial pose; And registering the measured point cloud data corresponding to the reference initial pose with the first local point cloud map to obtain the second error.
4. The method of claim 2, wherein determining a third error for the reference initial pose based on the planned measured pose and the predicted pose corresponding to the reference initial pose comprises: And establishing a logarithmic mapping between the predicted pose corresponding to the reference initial pose and the corresponding planning measurement pose, and obtaining a third error of the reference initial pose.
5. The method of claim 1, wherein each of the reference initial poses satisfies a preset verification condition, comprising any one of: the difference degree between the two reference initial poses with the minimum first error is smaller than the preset difference degree; The first error is minimum and a preset number of reference initial poses are distributed and diverged in a global point cloud map of the area where the robot is located.
6. The method of claim 1, wherein initially screening at least one reference initial pose from the at least one candidate initial pose based on each of the candidate initial pose and the current point cloud data, comprising: Registering the current point cloud data with a second local point cloud map corresponding to each candidate initial pose for each candidate initial pose to obtain pose transformation data for correcting the candidate initial pose; Determining a first error corresponding to the candidate initial pose according to the position information of each registration point and pose transformation data corresponding to the candidate initial pose; And primarily screening at least one reference initial pose from the at least one candidate initial pose according to a first error corresponding to each candidate initial pose.
7. The method of any of claims 1-6, wherein initially screening at least one reference initial pose from the at least one candidate initial pose based on each of the candidate initial pose and the current point cloud data, comprising: acquiring gravity alignment rotation data of the robot at the restarting moment; Correcting the candidate initial pose according to the gravity alignment rotation data aiming at each candidate initial pose to obtain a corrected initial pose corresponding to the candidate initial pose; and according to the current point cloud data and the corrected initial pose corresponding to each candidate initial pose, initially screening at least one reference initial pose from the at least one candidate initial pose.
8. A robotic repositioning apparatus, the apparatus comprising: the first acquisition module is used for acquiring at least one candidate initial pose of the robot at the restarting moment and current point cloud data of the robot at the restarting moment; the primary screening module is used for primarily screening at least one reference initial pose from the at least one candidate initial pose according to each candidate initial pose and the current point cloud data; The second acquisition module is used for acquiring, for each reference initial pose, predicted poses of the robot after performing a preset planning action based on the reference initial poses and measurement point cloud data of the robot after performing the preset planning action under the condition that each reference initial pose meets a preset verification condition; And the fine screening module is used for fine screening the target initial pose of the robot at the restarting moment from the at least one reference initial pose according to the measured point cloud data and the predicted pose corresponding to each reference initial pose.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 7 when the computer program is executed.
10. 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 7.

Description

Robot repositioning method, apparatus, computer device and storage medium Technical Field The present application relates to the field of robot navigation and positioning technologies, and in particular, to a method and apparatus for repositioning a robot, a computer device, and a storage medium. Background In the field of mobile robot navigation and positioning, in the process that the robot needs to be repositioned, firstly, loading an offline map, then collecting the current laser point cloud, matching the laser point cloud with a global descriptor, estimating the initial pose of the robot under the offline map coordinate system, and entering a navigation operation stage. However, in an indoor scene, repeated structures such as a corridor, a shelf, a door opening and the like often exist, a single-frame point cloud or a short-time point cloud possibly corresponds to a plurality of similar positions in a global map, and only an optimal matching pose is directly selected according to a global descriptor, namely, a scheme for considering successful repositioning is considered, so that a false repositioning phenomenon is easy to occur, and an inaccurate repositioning condition is caused. Disclosure of Invention Based on the above, the present application provides a method, an apparatus, a computer device and a storage medium for repositioning a robot, which can improve repositioning accuracy. In a first aspect, the present application provides a robot repositioning method, comprising: acquiring at least one candidate initial pose of the robot at the restarting moment and current point cloud data of the robot at the restarting moment; According to each candidate initial pose and the current point cloud data, at least one reference initial pose is initially screened from at least one candidate initial pose; Under the condition that each reference initial pose meets a preset verification condition, aiming at each reference initial pose, acquiring a predicted pose of the robot after the robot executes a preset planning action based on the reference initial pose and measuring point cloud data of the robot after the robot executes the preset planning action; and fine screening the target initial pose of the robot at the restarting moment from at least one reference initial pose according to the measured point cloud data and the predicted pose corresponding to each reference initial pose. According to the robot repositioning method, at least one reference initial pose is screened out from at least one candidate initial pose according to the candidate initial poses and the current point cloud data, and under the condition that the reference initial poses meet the preset verification conditions, the fine screening process can be continuously carried out on the at least one reference initial pose, namely, the target initial pose is continuously screened out from the at least one reference initial pose so as to further disambiguate, and the target initial pose can be screened out more accurately through the two screening processes, so that the repositioning accuracy can be improved to a certain extent. In one embodiment, the method comprises the steps of screening out target initial pose of the robot at the restarting time from at least one reference initial pose according to the measured point cloud data and the predicted pose corresponding to each reference initial pose, wherein the method comprises the steps of obtaining first errors of the initially screened reference initial pose for each reference initial pose; the method comprises the steps of determining a second error of a reference initial pose according to measured point cloud data and a predicted pose corresponding to the reference initial pose, determining a third error of the reference initial pose according to a planned measured pose corresponding to the reference initial pose and the predicted pose, wherein the planned measured pose corresponding to the reference initial pose is a measured pose of a robot under the reference initial pose after a preset planning action is executed, and fine screening a target initial pose of the robot at a restarting moment from at least one reference initial pose according to the second error and/or the third error and the first error respectively corresponding to each reference initial pose. In this embodiment, based on the second error and/or the third error and the first error corresponding to each reference initial pose, the integrated error corresponding to each reference initial pose can be obtained, so that the integrated error is used as a screening basis, the target initial pose can be screened out more accurately, and the repositioning accuracy can be improved to a certain extent. In one embodiment, determining the second error of the reference initial pose according to the measured point cloud data and the predicted pose corresponding to the reference initial pose comprises selecting a first local point cloud