CN-122023592-A - Scene map construction method, equipment and medium based on mobile robot

CN122023592ACN 122023592 ACN122023592 ACN 122023592ACN-122023592-A

Abstract

The invention discloses a scene map construction method, equipment and medium based on a mobile robot, and relates to the technical field of mobile robot navigation, comprising the steps of acquiring a reference scene map and a reference key frame pose by a high-precision offline optimization method based on an initial scene map and a pose map which are repeatedly operated in the same environment, extracting key frame topological features and key frame geometric features by combining the initial scene map and the pose map with the reference scene map and the reference key frame pose, and training a global deformation priori model according to the space position difference of the key frames; in the long-term running process of the mobile robot, online key frame topological features and online key frame geometric features of an online pose graph and an online scene map are extracted, a global deformation priori model is input, deformation suggestion information is generated, graph optimization is performed according to the deformation suggestion information, and the online scene map is updated. The invention improves the consistency of the scene map.

Inventors

WU JIAN
MA YE
WANG FEI
YUAN MENGRU
SHI MIAO

Assignees

杭州它人机器人技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260403

Claims (10)

1. A scene map construction method based on a mobile robot is characterized by comprising the following steps of, Establishing a unified vehicle body coordinate reference relation on the mobile robot, and collecting original observation data and pose initial values; preprocessing the original observation data, selecting key frames, distributing pose initial values for each key frame, constructing a pose graph, and generating an initial scene map according to the pose superposition point cloud of the key frames; Based on an initial scene map and a pose map which are repeatedly operated in the same environment, acquiring a reference scene map and a reference key frame pose by a high-precision offline optimization method, combining the initial scene map and the pose map with the reference scene map and the reference key frame pose, extracting key frame topological features and key frame geometric features, and training a global deformation priori model according to the space position difference of the key frames; In the long-term running process of the mobile robot, online key frame topological features and online key frame geometric features of an online pose graph and an online scene map are extracted, a global deformation priori model is input, deformation suggestion information is generated, graph optimization is performed according to the deformation suggestion information, and the online scene map is updated.
2. The method for constructing a scene map based on a mobile robot according to claim 1, wherein the step of establishing a unified body coordinate reference relationship on the mobile robot, collecting original observation data and pose initial values comprises the steps of, Establishing a unified vehicle body coordinate reference relation by carrying out joint calibration on all sensors on the mobile robot, and configuring unified time stamps for all the sensors through time synchronization; Based on the unified vehicle body coordinate reference relation and the timestamps, pose initial values are calculated for each timestamp and combined with data acquired in the timestamps corresponding to all the sensors, so that original observation data are generated.
3. The method for constructing a scene map based on a mobile robot according to claim 2, wherein the steps of preprocessing the original observation data and selecting key frames, assigning a pose initial value to each key frame, constructing a pose map are as follows, Performing distortion correction, filtering, downsampling and time alignment on the original observed data to obtain preprocessed observed data; Selecting key frames based on the preprocessed observation data and the pose initial values, reserving the pose initial value corresponding to each key frame, and executing closed loop detection between non-adjacent key frames to acquire closed loop constraint; And constructing a pose graph by calculating the relative motion constraint of adjacent key frames and combining the relative motion constraint with the closed loop constraint.
4. The method for building a scene map based on a mobile robot according to claim 3, wherein said generating an initial scene map based on the key frame pose superimposing point cloud comprises the steps of, Performing map optimization on the pose map to obtain the pose of the key frame; extracting laser radar point clouds corresponding to key frame time stamps in the pose graph based on the preprocessed observation data; and converting the laser radar point cloud into a unified vehicle body coordinate reference relationship, converting the laser radar point cloud into a global coordinate reference relationship according to the key frame pose, and superposing the laser radar point cloud and the global coordinate reference relationship to generate an initial scene map.
5. The method for constructing a scene map based on a mobile robot according to claim 4, wherein the steps of acquiring a reference scene map and a reference key frame pose by a high-precision offline optimization method based on the initial scene map and the pose map acquired by repeatedly running in the same environment are as follows, Reading an initial scene map and a pose map which are obtained by repeatedly running in the same environment, and constructing a combined pose map according to key frames, relative motion constraints and closed loop constraints in the pose map operated at each time; Performing joint optimization on the relative motion constraint and the closed-loop constraint in the joint pose graph by using the keyframe pose in the joint pose graph as an optimization variable through a high-precision offline optimization method to obtain a reference keyframe pose corresponding to the keyframe in the joint pose graph; And carrying out coordinate transformation and fusion on point clouds in the initial scene map based on the reference key frame pose and the initial scene map, and generating a reference scene map under a global coordinate reference relationship.
6. The method for constructing a scene map based on a mobile robot according to claim 5, wherein the steps of combining an initial scene map and a pose map with a reference scene map and a reference key frame pose, extracting key frame topological features and key frame geometric features, and training a global deformation priori model according to the key frame spatial position difference are as follows, Based on the pose graph and the pose of the reference key frame, counting the number of adjacent key frames, constraint type distribution and local communication structure of each key frame, and generating key frame topological characteristics; Performing plane fitting, edge detection and point cloud density statistics on the local point clouds through the local point clouds corresponding to the key frame indexes in the initial scene map and the reference scene map to generate key frame geometric features; Calculating a key frame space position difference according to the key frame pose and the reference key frame pose in the pose graph, and combining key frame topological features, key frame geometric features and key frame space position differences into a training sample set; And based on the training sample set, performing multi-round forward calculation and parameter updating on the global deformation prior model to complete the global deformation prior model training.
7. The method for constructing a scene map based on a mobile robot according to claim 6, wherein in the long-term running process of the mobile robot, online key frame topological features and online key frame geometric features of the online pose map and the online scene map are extracted and input into a global deformation priori model to generate deformation suggestion information, the steps are as follows, In the long-term running process of the mobile robot, selecting an online keyframe set from the online pose graph according to the online pose graph and the online scene map which are continuously updated; based on the online pose graph and the online key frame set, counting the number of adjacent online key frames, constraint types and local connected structures of each online key frame in the online key frame set, and generating an online key frame topological feature set; Based on an online scene map and an online key frame set, intercepting a local point cloud area corresponding to each online key frame in the online key frame set from the online scene map, and obtaining an online key frame geometric feature set through calculation of the local point cloud area; and splicing the online key frame topological feature set and the online key frame geometric feature set to generate an online key frame input feature set and input a global deformation priori model to obtain deformation suggestion information.
8. The method for building a scene map based on a mobile robot according to claim 7, wherein said performing map optimization based on the deformation advice information updates the on-line scene map as follows, Correlating the deformation suggestion information with the online key frame pose in the online pose graph to generate a corrected initial pose set; Based on the corrected initial pose set, relative motion constraint and closed loop constraint in the online pose graph, constructing a graph optimization problem by taking the corrected initial pose set as an optimization initial value, and performing nonlinear optimization on the graph optimization problem to obtain an online key frame pose set; and according to the pose set of the online key frame, performing coordinate transformation and point cloud fusion operation on a local point cloud area corresponding to the online key frame in the online scene map, and updating the online scene map.
9. A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the method is characterized in that the processor realizes the steps of the scene map construction method based on the mobile robot according to any one of claims 1-8 when executing the computer program.
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 mobile robot-based scene map construction method according to any one of claims 1 to 8.

Description

Scene map construction method, equipment and medium based on mobile robot Technical Field The invention relates to the technical field of mobile robot navigation, in particular to a scene map construction method, equipment and medium based on a mobile robot. Background With the popularization of service robots and warehouse logistics robots, the construction of an environment scene map based on a mobile robot becomes a basic technology for realizing autonomous positioning and path planning, and the conventional technology generally collects point clouds, images and motion information in the running process according to multi-source sensors such as a laser radar, a camera, an inertial measurement device, a wheel type odometer and the like which are arranged on the robot, predicts the pose of the robot through odometer calculation and image optimization, and superimposes the point clouds at all moments under a unified coordinate reference relationship to generate the environment scene map. In addition, the online map optimization is dependent on relative motion constraint and local closed loop constraint in the current operation, the integral deformation accumulated for a long time is mainly indirectly absorbed through local adjustment, the deformation constraint of a global layer is not established by combining the deformation rule obtained through statistics of historical operation data, and the consistency maintenance of the topological structure and the measurement precision of the scene map under the long-time operation condition still has room for improvement. Disclosure of Invention The present invention has been made in view of the above-described problems occurring in the prior art. Therefore, the invention provides a scene map construction method based on a mobile robot, which solves the problems that a unified calibration reference scene map is difficult to construct when the same environment is operated for a plurality of times, and global constraint and active correction on map deformation by utilizing historical operation data are lacked in a long-term operation process. In order to solve the technical problems, the invention provides the following technical scheme: In a first aspect, the present invention provides a mobile robot-based scene map construction method, comprising, Establishing a unified vehicle body coordinate reference relation on the mobile robot, and collecting original observation data and pose initial values; preprocessing the original observation data, selecting key frames, distributing pose initial values for each key frame, constructing a pose graph, and generating an initial scene map according to the pose superposition point cloud of the key frames; Based on an initial scene map and a pose map which are repeatedly operated in the same environment, acquiring a reference scene map and a reference key frame pose by a high-precision offline optimization method, combining the initial scene map and the pose map with the reference scene map and the reference key frame pose, extracting key frame topological features and key frame geometric features, and training a global deformation priori model according to the space position difference of the key frames; In the long-term running process of the mobile robot, online key frame topological features and online key frame geometric features of an online pose graph and an online scene map are extracted, a global deformation priori model is input, deformation suggestion information is generated, graph optimization is performed according to the deformation suggestion information, and the online scene map is updated. The method for constructing the scene map based on the mobile robot is used as an optimal scheme, wherein a unified vehicle body coordinate reference relation is established on the mobile robot, original observation data and pose initial values are collected, the steps are as follows, Establishing a unified vehicle body coordinate reference relation by carrying out joint calibration on all sensors on the mobile robot, and configuring unified time stamps for all the sensors through time synchronization; Based on the unified vehicle body coordinate reference relation and the timestamps, pose initial values are calculated for each timestamp and combined with data acquired in the timestamps corresponding to all the sensors, so that original observation data are generated. The method for constructing the scene map based on the mobile robot is characterized by comprising the following steps of preprocessing the original observed data, selecting key frames, distributing pose initial values for each key frame, constructing a pose map, Performing distortion correction, filtering, downsampling and time alignment on the original observed data to obtain preprocessed observed data; Selecting key frames based on the preprocessed observation data and the pose initial values, reserving the pose initial value corresponding to each key frame,