CN-122015822-A - BIM-driven robot multi-mode fusion positioning method

Abstract

A multi-mode fusion positioning method of a robot based on BIM driving comprises 1) extracting core structural elements from a BIM model in the construction and initialization stages of the BIM prior map to generate a high-precision prior point cloud map, and performing rough matching with an initial laser radar point cloud when the robot is started to finish pose alignment under a BIM global coordinate system. 2) In the stage of real-time data parallel processing and feature extraction, the system processes high-frequency data from the IMU in parallel to perform pre-integration so as to correct laser radar point cloud distortion, and extracts environmental geometric features and BIM-matching-oriented structural features from the laser radar point cloud. 3) In the factor graph construction and optimization positioning stage, a multi-constraint factor graph model is constructed, the motion estimation provided by the IMU, the relative pose provided by the scanning matching of the geometric features and the local sub-map and the strong global constraint obtained by registering the structural features and the global BIM prior map are fused, so that the positioning with high precision and high robustness is realized, and the accumulated drift is effectively restrained.

Inventors

• MENG YUXIANG
• CHENG LE
• LIU JIN
• ZHANG XIAOGUO
• YANG YUAN
• FAN YONGCHAO
• WANG YIHAO
• WANG ZHENGDONG
• ZHAO XIAOBING
• JIANG ZHIJUN
• SHEN YUE
• MENG XIANGYIN
• ZU YU

Assignees

• 中交建筑集团有限公司
• 中交建筑集团第一工程有限公司
• 东南大学

Dates

Publication Date

20260512

Application Date

20251231

Claims (6)

1. 1. A robot multi-mode fusion positioning method based on BIM driving is characterized by comprising the following specific steps: step1, building a BIM prior map and aligning with an initial pose: extracting core structural elements from a building information model BIM to generate a global priori point cloud map, and registering with initial laser radar scanning data when a robot is started to finish initial pose alignment under a BIM global coordinate system; Step 2, IMU and LiDAR data parallel processing and feature extraction: real-time data of an Inertial Measurement Unit (IMU) and a laser radar LiDAR are processed in parallel, motion distortion correction is carried out on LiDAR point clouds by utilizing IMU data pre-integration, and geometric features for odometers and structural features for global matching are extracted synchronously from the point clouds; And 3, constructing and positioning and optimizing a multi-constraint factor graph model, fusing an IMU pre-integral, a laser radar odometer and global constraints obtained by matching structural features with a global BIM priori map, and outputting the robot pose with high precision and global consistency through nonlinear optimization solution.
2. 2. The method for multi-modal fusion positioning of a robot based on BIM driving of claim 1, wherein the step 1 is specifically as follows: and 1.1, inputting a BIM model of an industrial basic type IFC format, selectively extracting relevant core structural elements of a wall body, a column and a floor slab, and ignoring non-structural or volatile elements to form a simplified structural framework model.
3. 3. Step 1.2, carrying out surface treatment on the extracted structural frame model, and adopting a poisson disk sampling algorithm to generate a uniform and high-density three-dimensional point cloud on the surface of the model as a global BIM priori map; And 1.3, when the robot is started, acquiring and splicing laser radar point clouds of initial frames to form an initial local map, matching the initial local map with a global BIM priori map by adopting a RANSAC global registration method, and calculating an initial transformation matrix of the robot under a BIM global coordinate system to finish initialization.
4. 4. The method for multi-modal fusion positioning of a robot based on BIM driving of claim 1, wherein the step 2 is specifically as follows: Step 2.1, pre-integrating all IMU readings between two continuous laser radar key frames to obtain relative pose increment, correcting motion distortion in a laser radar scanning period by interpolation operation by utilizing the increment, and generating undistorted accurate point cloud; Step 2.2, calculating the local curvature of the point along each scanning line for the undistorted point cloud, and classifying the point cloud into edge points and plane point related geometric features for laser radar odometer matching according to curvature threshold values; And 2.3, identifying and extracting a structured characteristic point set belonging to the plane structure such as the wall surface from the point cloud by adopting a self-adaptive sliding window and straight line fitting method, wherein the process can effectively filter interference points generated by temporary obstacles, and the extracted point set is used for subsequent global matching with a BIM prior map.
5. 5. The method for multi-modal fusion localization of a robot based on BIM driving as set forth in claim 3, wherein said step 2.3 is specifically as follows: Firstly, moving a sliding window with a self-adaptive size along a point cloud scanning line, secondly, carrying out least square straight line fitting on a point set in each window, calculating a fitting residual error, then, starting from a window with the residual error meeting a preset threshold value, rearwards aggregating all continuous points with the distance smaller than the threshold value from the fitting straight line, marking the continuous points as structural points, and finally, collecting all marked structural points to form a structured characteristic point cloud for BIM matching.
6. 6. The method for multi-modal fusion positioning of a robot based on BIM driving of claim 1, wherein the step 3 is specifically as follows: Step 3.1, constructing a factor graph comprising at least three types of constraint factors: IMU pre-integration factor, which is to connect pose nodes of two continuous key frames to provide high frequency motion constraint; the laser radar odometer factor is that the geometric characteristics of the current frame are matched with the local sub-map to provide high-precision relative pose constraint; BIM global constraint factors, namely directly associating pose nodes of the current frame with a global BIM coordinate system to provide global absolute pose constraint; step 3.2, setting a trigger mechanism, and triggering a BIM global constraint factor when the accumulated displacement or the uncertainty of the odometer of the robot exceeds a threshold value, registering the structural feature point cloud extracted from the current frame with a global BIM prior map after triggering to obtain an observed quantity of the global absolute pose as the factor; Step 3.3, constructing all factors into a large nonlinear least square problem, and adopting iSAM and other back-end optimizers to solve in real time to obtain optimal estimation of all robot pose nodes, so as to output a globally consistent and high-precision positioning result; 。

Description

BIM-driven robot multi-mode fusion positioning method Technical Field The invention belongs to the technical field of autonomous navigation and positioning of robots, and particularly relates to a multi-mode fusion positioning method of a robot based on BIM driving. Background In recent years, along with the continuous improvement of automation and intelligent level, the autonomous mobile robot has been increasingly used in the fields of building construction, post maintenance, inspection monitoring and the like. For example, the robot for carrying out autonomous spray curing on a newly poured concrete structure can remarkably improve the integrity and durability of the building structure. The reliable autonomous operation of the robot is realized, and the primary premise is to perform accurate and robust self-positioning in a complex and unstructured operation environment. However, the construction site is significantly different from the indoor environments such as regular offices and houses, and the like, and the existing positioning technology is severely challenged by lack of infrastructure and signal interference, namely the construction site (especially at the stage of just finishing a main structure) is generally lack of stable power supply, doors and windows and other facilities, so that the positioning scheme such as Ultra Wideband (UWB), wi-Fi or Bluetooth relying on pre-deployment beacons is high in deployment cost and difficulty, and wireless signals are extremely easy to be interfered by absorption, reflection and diffraction of reinforced concrete structures, so that signals are unstable and positioning accuracy is reduced. The perceived environment is bad, and the construction site environment is extremely complex and changeable. First, open architecture, severe variations and even missing natural illumination create significant problems for visual SLAM (V-SLAM) methods that rely on illumination and textural features. Secondly, a large amount of dust and smoke generated in the construction process can seriously influence the propagation of laser radar (LiDAR) beams, so that a large amount of noise and even failure are generated in point cloud data. The construction site is filled with barriers which dynamically change, such as movable scaffolds, building material piles, constructors and the like. These dynamic elements can severely interfere with conventional SLAM algorithms based on static environment assumptions. Meanwhile, the scenes such as large-area walls, floors and the like lack enough geometric characteristics, so that the sensor data are easy to degrade, and drift and even complete failure of a positioning algorithm are caused. In order to address the above challenges, a variety of synchronous positioning and map building (SLAM) solutions have been proposed by the academy. Early filtering-based methods, such as extended kalman filters (EKF-SLAM), have tended to perform poorly in highly dynamic, non-linear building environments due to their linear and gaussian noise assumptions, and state vector dimensions expand dramatically as map scale increases, resulting in high computational complexity and prone to inconsistency problems. Particle filtering (PF-SLAM) can handle non-Gaussian distribution, but requires massive particles in a high-dimensional state space, has huge calculation cost, and is difficult to meet real-time requirements. Currently, optimization-based methods, particularly Graph-based SLAM (Graph-based SLAM), have become mainstream. According to the method, the pose and the landmark points of the robot are taken as nodes of the graph, the sensor observation is taken as edges (constraint) of the connecting nodes, and the optimal estimation of all the nodes is solved through the nonlinear optimization of the rear end. The Factor Graph (Factor Graph) is used as a general representation form of Graph optimization, and heterogeneous information from different sensors such as a laser radar, a camera and an Inertial Measurement Unit (IMU) can be flexibly and efficiently fused. By constructing tightly coupled multi-modal fusion models, such as LIO-SAM, FAST-LIO and the like, motion estimation can be provided by means of other sensors (such as IMU) when a single sensor (such as LiDAR) is degraded temporarily, so that the robustness of the system is improved to a certain extent. However, these methods still rely on high quality sensor inputs and reliable loop-back detection to correct long-term accumulated errors in nature. In construction environments with similar structures (such as long corridor and repeated floors), sparse features or dynamic scene changes, effective loop detection is difficult to realize, so that the graph optimization method also faces serious long-term drift problems, and the severe requirements of tasks such as building maintenance and mapping on global position accuracy cannot be met. In order to solve the above problems, the prior art is as follows: Appli