CN-121994207-A - Construction method, device and storage medium for mobile robot map

CN121994207ACN 121994207 ACN121994207 ACN 121994207ACN-121994207-A

Abstract

The application discloses a construction method, a construction device and a storage medium for a mobile robot map, and belongs to the technical field of positioning and map construction. The construction method comprises the steps of obtaining a current measurement point cloud and a current prediction pose of a mobile robot at a current time, carrying out normal distribution transformation matching on the current measurement point cloud and a point cloud corresponding to the current prediction pose on a pre-constructed global voxel map to obtain a point cloud global matching residual error, carrying out iterative closest point matching on the current measurement point cloud and the point cloud corresponding to the current prediction pose on a pre-constructed local dense map to obtain a current observation pose and a point cloud local matching residual error, determining a current actual pose according to the current observation pose, the current prediction pose and the point cloud local matching residual error, obtaining a current actual point cloud matched with the current actual pose, and constructing the mobile robot map according to the current actual pose and the current actual point cloud. The method and the device are used for improving the construction precision of the mobile robot map.

Inventors

WU FAN
YANG JUN
Xie Changzuo
TONG XING
SHE LINGJUAN

Assignees

中科云谷科技有限公司

Dates

Publication Date: 20260508
Application Date: 20251231

Claims (10)

1. A construction method for a mobile robot map, the construction method comprising: Acquiring a current measurement point cloud obtained by measuring the mobile robot through a radar device at a current time and a current predicted pose generated by an inertial measurement unit; Performing normal distribution transformation matching on the current measurement point cloud and the point cloud corresponding to the current prediction pose on a pre-constructed global voxel map to obtain a point cloud global matching residual error of the mobile robot at the current moment; Performing iterative closest point matching on the current measurement point cloud and the point cloud corresponding to the current prediction pose on a pre-constructed local dense map under the condition that the global point cloud matching residual is larger than or equal to a first preset residual threshold value so as to obtain the current observation pose and the point cloud local matching residual of the mobile robot at the current moment; Determining a current actual pose of the mobile robot at a current time according to the current observation pose, the current prediction pose and the point cloud local matching residual error, and obtaining a current actual point cloud of the mobile robot matched with the current actual pose at the current time based on the current actual pose; and constructing a mobile robot map of the mobile robot according to the current actual pose and the current actual point cloud.
2. The method of claim 1, wherein the generating of the current predicted pose comprises: Acquiring the last actual pose of the mobile robot at the last moment and the mobile measurement parameters of the mobile robot at the current moment, which are measured by the inertial measurement unit; Based on a preset pose prediction algorithm, predicting the pose of the mobile robot at the current moment according to the last actual pose and the mobile measurement parameters so as to obtain the current predicted pose.
3. The method of claim 1, wherein the determining the current actual pose of the mobile robot at the current time based on the current viewing pose, the current predicted pose, and the point cloud local matching residual comprises: under the condition that the point cloud local matching residual error is smaller than a second preset residual error threshold, determining the current actual pose of the mobile robot at the current moment according to the current observation pose, the current prediction pose and the Kalman gain at the current moment; And under the condition that the point cloud local matching residual error is larger than or equal to the second preset residual error threshold, determining the current actual pose of the mobile robot at the current moment according to the current observation pose, the current prediction pose and the corrected Kalman gain at the current moment.
4. A method according to claim 3, wherein said determining a current actual pose of the mobile robot at a current time based on the current viewing pose, the current predicted pose, and a kalman gain at the current time comprises: Wherein, the The mobile robot is at present The current actual pose at the moment, The mobile robot is at present The current predicted pose at the moment in time, To be at present The Kalman gain corresponding to the Kalman filtering algorithm under the moment, Currently for the mobile robot The current observation pose at the moment, Currently for the mobile robot An observation model corresponding to the current predicted pose at the moment.
5. A method according to claim 3, wherein said determining a current actual pose of the mobile robot at a current time based on the current viewing pose, the current predicted pose, and a modified kalman gain at the current time comprises: Wherein, the The mobile robot is at present The current actual pose at the moment, The mobile robot is at present The current predicted pose at the moment in time, To be at present The corrected kalman gain at the time instant, Currently for the mobile robot The current observation pose at the moment, Currently for the mobile robot An observation model corresponding to the current predicted pose at the moment.
6. A method according to claim 3, wherein the determination of the corrected kalman gain at the current time instant comprises: Determining a penalty factor according to the point cloud local matching residual error; and reducing the Kalman gain of the Kalman filtering algorithm at the current time according to the penalty factor to obtain the corrected Kalman gain at the current time.
7. The method according to claim 1, wherein the method further comprises: And under the condition that the global matching residual error of the point cloud is smaller than the first preset residual error threshold, determining the current actual pose of the mobile robot at the current moment based on a Kalman filtering algorithm according to the initial observation pose and the current predicted pose obtained by matching the normal distribution change.
8. A construction apparatus for a mobile robot map, comprising: A memory configured to store instructions, and A processor configured to invoke the instructions from the memory and when executing the instructions is capable of implementing the construction method for a mobile robot map according to any of claims 1 to 7.
9. A mobile robot, comprising: the construction device for a mobile robot map according to claim 8.
10. A machine-readable storage medium having stored thereon instructions for causing a machine to perform the method of constructing a map for a mobile robot according to any one of claims 1 to 7.

Description

Construction method, device and storage medium for mobile robot map Technical Field The application relates to the technical field of positioning and mapping, in particular to a method and a device for constructing a mobile robot map and a storage medium. Background The construction of the smart mobile device map enables autonomous movement of the smart mobile device (e.g., mobile robot). At present, point cloud data of an intelligent mobile device at a certain moment is matched with a local map constructed at a previous moment, so that pose data of the intelligent mobile device at the certain moment is updated, and the local map is updated according to the pose data at the certain moment, so that state update of the intelligent mobile device is realized. However, in the prior art, if a large error occurs in matching between the point cloud data and the local map at a certain moment, corresponding pose data with the large error is obtained, and the pose data generated by the matching method continuously influences the construction of the subsequent map, so that the error increases with the lapse of time, and the global map is distorted integrally, so that the construction precision of the map of the intelligent mobile device is reduced. Disclosure of Invention The embodiment of the application aims to provide a method and a device for constructing a map of a mobile robot, the mobile robot and a storage medium, which are used for solving the problem of low construction precision of an intelligent mobile device map in the prior art. In order to achieve the above object, a first aspect of the present application provides a construction method for a mobile robot map, the construction method comprising: Acquiring a current measurement point cloud obtained by measuring the mobile robot through a radar device at a current time and a current predicted pose generated by an inertial measurement unit; performing normal distribution transformation matching on the current measurement point cloud and the point cloud corresponding to the current prediction pose on the pre-constructed global voxel map to obtain a point cloud global matching residual error of the mobile robot at the current moment; Performing iterative closest point matching on the current measurement point cloud and the point cloud corresponding to the current predicted pose on the pre-constructed local dense map under the condition that the global point cloud matching residual is larger than or equal to a first preset residual threshold value so as to obtain the current observation pose of the mobile robot at the current moment and the local point cloud matching residual; determining the current actual pose of the mobile robot at the current time according to the current observation pose, the current predicted pose and the point cloud local matching residual error, and obtaining the current actual point cloud of the mobile robot matched with the current actual pose at the current time based on the current actual pose; And constructing a mobile robot map of the mobile robot according to the current actual pose and the current actual point cloud. The method comprises the steps of obtaining the last actual pose of the mobile robot at the last moment and the movement measurement parameters of the mobile robot at the current moment, which are measured by an inertial measurement unit, and predicting the pose of the mobile robot at the current moment according to the last actual pose and the movement measurement parameters based on a preset pose prediction algorithm so as to obtain the current predicted pose. The method comprises the steps of determining the current actual pose of the mobile robot at the current time according to the current observation pose, the current predicted pose and the Kalman gain at the current time under the condition that the point cloud local matching residual is smaller than a second preset residual threshold, and determining the current actual pose of the mobile robot at the current time according to the current observation pose, the current predicted pose and the Kalman gain at the current time under the condition that the point cloud local matching residual is larger than or equal to the second preset residual threshold. In the embodiment of the application, determining the current actual pose of the mobile robot at the current time according to the current observation pose, the current predicted pose and the Kalman gain at the current time comprises the following steps: Wherein, the Mobile robot is at presentThe current actual pose at the moment,Mobile robot is at presentThe current predicted pose at the moment in time,To be at presentThe Kalman gain corresponding to the Kalman filtering algorithm under the moment,At present for mobile robotThe current observation pose at the moment,At present for mobile robotAn observation model corresponding to the current predicted pose at the moment. In the embodiment of the application, ac