Robot autonomy in harsh outdoor environments can be achieved through methods and robots
Through lidar acquisition, European clustering, K-Means denoising cloud data, combined with probability positioning and deep learning analysis, the problem of robot perception accuracy and autonomous motion accuracy in the wild environment is solved, and more efficient Independent movement in the wild.
Patent Information
- Application Number
- CN202310250617.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-03-15
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2043-03-15
AI Technical Summary
Existing robots have poor environmental perception map construction and positioning accuracy in harsh environments in the wild, and consume a lot of computing resources, resulting in unstable obstacle recognition and inaccurate autonomous movement.
LiDAR is used to collect 3D point cloud data in real time, combine European clustering and K-Means algorithm to remove noise, and optimize point cloud data processing and path planning through probability positioning and deep learning analysis.
It improves data stability and positioning efficiency, reduces the uncertainty of passability analysis, and solves the problems of large amount of point cloud data, real-time positioning and poor passability analysis accuracy in field environments.
Smart Images

Figure CN116088536B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of robot autonomous motion, and in particular to a robot autonomous motion method and a robot in harsh outdoor environments. Background Art
[0002] With the development of social and economic level and the yearning for smart cities, the country has increased its investment in unmanned systems and intelligent agents, and typical intelligent agents such as robots are widely used in life. The purpose of the application of traditional robots is to reduce some tedious tasks of people, such as industrial robots in factories and food delivery robots in restaurants. The autonomous movement of robots mainly consists of three parts, namely environmental perception, autonomous planning and behavior control. The environmental perception part uses the sensors carried by the robot body to collect information about objects in the surrounding environment and then analyze its own posture; autonomous planning is to design the route to reach the destination in the perceived environment according to the destination required by the task; behavior control is a design of how the robot body reaches the destination. In the wild environment, the unknown environment makes some existing maps or GPS positioning inaccurate, which makes it difficult to meet the requirements in the implementation of some specific tasks.
[0003] The existing autonomous motion solutions of robots generally use lidar or cameras as data acquisition devices, and fix sensors on the front or upper part of the robot. For the collected data, real-time processing is generally used to integrate the data to construct a map that the robot perceives in real time, and then determine whether there are some obvious obstacles in the area around the robot. The radius of the obstacle is set to more than 1.2 times the actual radius to ensure that the robot can avoid and pass through obstacles during autonomous movement.
[0004] Solution 1: Use the binocular camera on the robot to collect data in the environment, build an environmental map through real-time computer processing, and achieve real-time positioning of the robot through feature point matching. Convert the image data (left and right image data or depth data) in the binocular camera into three-dimensional point cloud data, use the point cloud data to select the cost range of obstacles, and then select the passable area of the robot.
[0005] Solution 2: Use the laser radar on the robot to collect data in the environment, remove some erroneous points in the data through real-time computer processing, and then build a map. Use the actual point cloud data and the ICP algorithm to calculate the current position and get the real-time position of the robot. Use the constructed 3D point cloud map to select the actual passable area of the robot.
[0006] There are two main disadvantages of the existing methods: 1) Poor accuracy in environmental perception map construction and positioning. Traditional robots generally use all the data collected by sensors to construct maps and real-time positioning for environmental perception. However, in harsh outdoor environments, the data collected by the camera is relatively noisy, resulting in poor image feature extraction, which further leads to poor subsequent mapping and positioning effects; while the data collected by the lidar has more noise points and contributes less to actual mapping and positioning. In summary, poor environmental perception and positioning will also make it difficult and unstable to identify obstacles. 2) In harsh outdoor scenes, the irregular environment and rapid changes in the robot's posture result in a large amount of data collected by the sensor, which contains a large number of corner points and messy invalid image features. When the robot processes in real time, it will increase the consumption of the entire computing resources, resulting in a decrease in subsequent computing power, resulting in a deterioration in overall real-time performance, and unable to accurately plan the robot's real-time passability.
[0007] In order to avoid the problems of high positioning accuracy and large amount of calculation in map construction in the above environment, the data collected by the robot's sensors is subjected to Euclidean clustering to remove corner points, depth information error points, and meaningless points (such as points generated by dynamic objects). This is a new method for environmental perception based on data removal. This method uses point cloud data after reducing dynamic points as dynamic input, but when repeatedly passing the same position, it will cause data accumulation and redundancy, resulting in low accuracy of environmental perception, large drift when constructing the map, and poor passability.
[0008] The main disadvantage of the above methods is that the mapping and positioning accuracy of the robot when perceiving the environment and the analysis of passability are still limited by the amount of data collected by the sensor, and the accuracy of data removal is low. The poor quality of removal leads to the deletion of some static features in the environment, which reduces the positioning accuracy and leads to worse passability. At the same time, the information of the vehicle itself is not considered in the algorithm's passability, which leads to unexpected situations such as rollover in actual autonomous movement. Summary of the invention
[0009] The purpose of the present invention is to provide a robot autonomous passability method and robot for harsh outdoor environments, so as to solve the problem that the robot cannot adaptively learn due to large amount of point cloud data, poor positioning real-time and passability analysis accuracy in the outdoor environment.
[0010] To achieve the above object, the present invention provides the following solutions:
[0011] A method for autonomous robot passage in harsh outdoor environments, comprising:
[0012] When the robot moves in the field operation area, it uses LiDAR to collect 3D point cloud data of the field environment in real time;
[0013] The 3D point cloud data is segmented and noise removed by using the Euclidean clustering algorithm and the K-Means clustering algorithm to obtain the currently processed point cloud data;
[0014] When there is no prior point cloud map in the robot or the matching error between the prior point cloud map and the currently processed point cloud data is greater than an error threshold, constructing a three-dimensional point cloud map using the currently processed point cloud data, and obtaining the position and posture of the robot body according to the three-dimensional point cloud map;
[0015] According to the currently processed point cloud data, the current position of the robot body is located in a probabilistic manner;
[0016] Based on the robot's position and the currently processed point cloud data, a deep learning method is used to analyze the environment's passability.
[0017] If the passability analysis fails, the point cloud data of the outdoor environment is collected again by using the laser radar at the current position, and the process returns to the step of "using the Euclidean clustering algorithm and the K-Means clustering algorithm to segment and remove noise from the 3D point cloud data to obtain the currently processed point cloud data";
[0018] If the passability analysis is successful, a path is planned between the current position and the target position according to the result of the passability analysis, and the robot is controlled to reach the target position according to the planned path;
[0019] When there is a prior point cloud map in the robot and the matching error between the prior point cloud map and the currently processed point cloud data is less than or equal to the error threshold, the position and posture of the robot body is obtained according to the prior point cloud map, and the process returns to the step of "based on the position and posture of the robot body and the currently processed point cloud data, using a deep learning method to perform a passability analysis on the environment".
[0020] Optionally, the 3D point cloud data is segmented and noise removed by using a Euclidean clustering algorithm and a K-Means clustering algorithm to obtain currently processed point cloud data, which also includes:
[0021] The robot hand-eye calibration method is used to calibrate the position transformation matrix between the laser radar's own coordinate system and the robot's base coordinate system and the field environment;
[0022] According to the position coordinates of the point cloud data, using the formula Calculate the world coordinates of the robot base coordinate system 3 p i ; In the formula, is the pose transformation matrix between the laser radar’s own coordinate system and the robot’s base coordinate system, 4T is the pose transformation matrix between the laser radar’s own coordinate system and the field environment, and p i is the location coordinate of the point cloud data.
[0023] Optionally, constructing a three-dimensional point cloud map using the currently processed point cloud data, and obtaining the position and posture of the robot body according to the three-dimensional point cloud map, specifically includes:
[0024] Extract features from two adjacent frames of point cloud data;
[0025] Match the extracted features to construct radar odometer and 3D point cloud map;
[0026] The Kalman filter is used to fuse the radar odometer and the three-dimensional point cloud map to obtain the position and posture of the robot.
[0027] Optionally, according to the currently processed point cloud data, the current position of the robot body is located in a probabilistic manner, specifically including:
[0028] Performing three-dimensional grid division on the currently processed point cloud data;
[0029] Calculate the multidimensional normal distribution parameters for each grid;
[0030] Convert the currently processed point cloud data into point cloud data in the robot base coordinate system;
[0031] Determine the probability distribution model based on the point cloud data and multi-dimensional normal distribution parameters in the robot base coordinate system;
[0032] The current position of the robot body is located according to the score of the probability distribution model.
[0033] Optionally, based on the posture of the robot body and the currently processed point cloud data, a deep learning method is used to perform a passability analysis of the environment, specifically including:
[0034] The robot body’s posture and the currently processed point cloud data are input into a convolutional neural network, and the results of the passability analysis are output.
[0035] Optionally, the point cloud data of the outdoor environment is collected again using a laser radar at the current location, specifically including:
[0036] At the current position, the robot rotates or moves itself and uses the lidar to collect point cloud data of the field environment again.
[0037] Optionally, a path is planned between the current location and the target location based on the result of the passability analysis, specifically including:
[0038] According to the results of the passability analysis, the A* algorithm or the D* algorithm is used to plan the path between the current position and the target position.
[0039] An autonomous robot for outdoor harsh environments, including: an AGV, a switch, an industrial computer and a laser radar;
[0040] The switch, industrial computer and laser radar are all installed on the AGV;
[0041] The signal output end of the laser radar is connected to the industrial computer through a switch, and the industrial computer is connected to the control end of the AGV car;
[0042] The laser radar is used to perform 3D point cloud scanning of the outdoor environment driven by the AGV, collect 3D point cloud data of the outdoor environment in real time, and transmit the 3D point cloud data to the industrial computer through the switch;
[0043] The industrial computer is used to perform a passability analysis based on the 3D point cloud data, and to perform real-time path planning based on the result of the passability analysis, and then to control the AGV vehicle to reach the target position according to the planned path.
[0044] Optionally, a switch, an industrial computer and a laser radar are fixedly installed on the AGV vehicle in sequence from bottom to top.
[0045] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0046] The present invention discloses a robot autonomous passability method and a robot for outdoor harsh environments. The method improves data stability by segmenting and removing noise based on Euclidean clustering and K-Means clustering for 3D point cloud data of the outdoor environment collected by a laser radar. The method is suitable for complex and large-scale outdoor harsh scenes. The real-time positioning of the robot is solved in a probability-based manner, which reduces the amount of calculation in the entire positioning process and improves the efficiency of positioning. The robot's position and posture are also included in the passability analysis, which reduces the uncertainty in the passability analysis and solves the problem that the robot cannot learn adaptively due to the large amount of point cloud data, poor positioning real-time and poor passability analysis accuracy in the outdoor environment. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0048] Figure 1A flowchart of a method for autonomously passing a robot in a harsh outdoor environment provided by an embodiment of the present invention;
[0049] Figure 2 A schematic diagram of a method for autonomously passing a robot in a harsh outdoor environment provided by an embodiment of the present invention;
[0050] Figure 3 A schematic diagram of 3D point cloud data collected in real time by a laser radar provided in an embodiment of the present invention.
[0051] Figure 4 A schematic diagram of the structure of an autonomous traversable robot for outdoor harsh environments provided by an embodiment of the present invention.
[0052] Explanation of symbols: 1-AGV, 2-switch, 3-industrial computer, 4-laser radar. DETAILED DESCRIPTION
[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0054] The purpose of the present invention is to provide a robot autonomous passability method and robot for harsh outdoor environments, so as to solve the problem that the robot cannot learn and adapt due to large amount of point cloud data, poor positioning real-time and passability analysis accuracy in the outdoor environment.
[0055] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0056] In view of the shortcomings of the prior art, the present invention proposes a method for autonomous robot passability in harsh outdoor environments. The method realizes real-time perception of the robot's autonomous movement and analysis of its passability through a 3D environment perception self-positioning system of a lidar, thereby solving the problem that the robot cannot work in harsh outdoor environments.
[0057] like Figure 1 As shown, an embodiment of the present invention provides a method for autonomous robot movement in harsh outdoor environments, including:
[0058] Step S1, when the robot moves in the field operation area, it uses the laser radar to collect 3D point cloud data of the field environment in real time.
[0059] Collect the initial point cloud data in the robot environment. First, the AGV drives the laser radar to scan the field environment in 3D point cloud, and collects the following data: Figure 3 The point cloud data shown, where the position coordinates of the point cloud data are set to p i .
[0060] Before proceeding to step S2, use the formula Calculate the world coordinates of the robot base coordinate system 3 p i After obtaining the world coordinates in the robot coordinate system, the world coordinates are processed in the robot's own coordinate system, that is, the world coordinates in the robot coordinate system are used as the original input of the subsequent algorithm.
[0061] In the formula, is the pose transformation matrix between the laser radar's own coordinate system and the robot's base coordinate system, and 4T is the pose transformation matrix between the laser radar's own coordinate system and the field environment.
[0062] Calibrate the relative pose transformation matrix between the LIDAR’s own coordinate system and the robot’s base coordinate system using any mature robot hand-eye calibration method (e.g. Zhao Z. Hand-eye calibration using convex optimization [C] / / 2011 IEEE International Conference on Robotics and Automation. IEEE, 2011: 2947-2952. or Park Y, Yun S, Won CS, et al. Calibration between color camera and 3D LIDAR instruments with a polygonal planarboard [J]. Sensors, 2014, 14 (3): 5333-5353.).
[0063] The pose transformation matrix 4T between the laser radar's own coordinate system and the field environment is calibrated using any mature robot hand-eye calibration method (same as above).
[0064] Step S2, using the Euclidean clustering algorithm and the K-Means clustering algorithm to segment and remove noise from the 3D point cloud data to obtain the currently processed point cloud data.
[0065] The point cloud data in the robot coordinate system is segmented and noise is removed. Due to the physical characteristics of the laser radar, the initial point cloud is a bunch of disordered points with a lot of noise. In order to achieve high-quality mapping of the robot in the field and reduce the calculation amount of positioning and passability analysis, the present invention uses two methods, Euclidean clustering and K-Means clustering, to segment and remove point cloud data and noise. The initial threshold of Euclidean clustering is set to t, and the minimum number of point clouds for a single cluster is set to N; the results of Euclidean clustering and K-Means clustering are combined, that is, clusters with a point cloud number less than N can be stably removed and the overlap and redundancy of point cloud data in repeated positions can be eliminated. Therefore, on this basis, the segmentation of noise and objects in the preliminary point cloud data is achieved.
[0066] Step S3, when there is no prior point cloud map in the robot or the matching error between the prior point cloud map and the currently processed point cloud data is greater than the error threshold, a three-dimensional point cloud map is constructed using the currently processed point cloud data, and the position and posture of the robot body is obtained according to the three-dimensional point cloud map.
[0067] Whether the robot has prior point cloud map information and the integrity of the prior map is used to determine whether the real-time mapping mode needs to be turned on. The prior map data is provided by the operator before field operations. If there is no prior point cloud map or the matching error between the prior point cloud map information and the currently collected point cloud data is large, the robot abandons the prior map and automatically turns on the mapping mode. In this step, the operator can subjectively choose whether to turn on the real-time mapping mode.
[0068] Use the processed point cloud data to build the environment map in real time. Extract the features of the previous and next frames of the processed point cloud data, match the extracted features, build the radar odometer and create the map. Odometer data refers to the information collected by the robot's own sensors during movement, which is processed by the algorithm to estimate the body's posture. However, it will be inaccurate due to the drift of the trajectory. Using the observation data of the map will cause the posture fed back by the map information to be inaccurate due to the imperfection and inaccuracy of the map information. According to the perfection of the map information, choose whether to use the posture fed back by the map data as the main body, use the Kalman filter to fuse the map data and odometer data, and output the accurate posture information of the robot body.
[0069] Step S4: based on the currently processed point cloud data, locate the current position of the robot body in a probabilistic manner.
[0070] Real-time positioning is achieved using point cloud data collected in real time. In order to solve the problem of large ICP calculation in traditional point cloud map positioning, the present invention estimates the position of the robot itself in a probabilistic way. The 3D point cloud data collected in real time is divided into three-dimensional grids of specified sizes, and the multi-dimensional normal distribution parameters are calculated to calculate the probability distribution model of the grid. The model score is used to evaluate the accuracy of the current estimate to achieve the positioning of the current position.
[0071] The calculation process of the multidimensional normal distribution parameters is as follows: the multidimensional normal distribution parameters of each grid have two q (the mean of the point cloud in each three-dimensional grid) and the covariance matrix ∑.
[0072] The calculation process of the probability distribution model is as follows: first, use 4T to convert the collected point cloud data into point cloud data in the robot coordinate system. Then use the multidimensional normal distribution parameters to calculate the probability distribution model.
[0073] The process of using the model's score to obtain the robot's real-time position is as follows: The score calculation formula is the addition of probabilities, that is, The score is derived and LM optimization is used to maximize the score. The Jacobian matrix and Hessian matrix of the point transformation can be calculated, that is, the position of the robot can be obtained.
[0074] Step S5, based on the posture of the robot body and the currently processed point cloud data, a deep learning method is used to perform a passability analysis on the environment.
[0075] In order to solve the problem that the traditional passability analysis method directly uses the mapping results for analysis, which leads to model instability and does not consider the information correlation between the environment and the robot body, the present invention uses a deep learning method based on the robot body posture and real-time environmental point cloud data to perform passability analysis on the environment. The use of deep learning methods can enable the robot to make its own analysis and judgment of complex environments in real-time passability calculations, greatly enhancing the robot's applicability in field environments.
[0076] That is, the posture of the robot body and the currently processed point cloud data are input into a convolutional neural network, and the result of the passability analysis is output.
[0077] Step S6: If the passability analysis fails, the point cloud data of the outdoor environment is collected again using the laser radar at the current position, and the process returns to step S2.
[0078] If the passability analysis fails, the robot collects more point cloud data at the current position (using its own rotation or small displacement, etc.) and makes another judgment until a stable passability analysis result is obtained. The judgment basis for "stable passability analysis result" is: after three calculations, the same passability path (i.e., movement direction) is obtained.
[0079] Step S7: If the passability analysis is successful, a path is planned between the current position and the target position according to the result of the passability analysis, and the robot is controlled to reach the target position according to the planned path.
[0080] The robot path and body control are performed based on the results of the passability analysis. Since the passability analysis method proposed in the present invention directly takes into account the raw data collected by the sensor and the information of the robot body, the robot can move autonomously and efficiently in harsh outdoor environments. The analysis results are used to use the A* algorithm or the D* algorithm to plan the path between the current position and the target position, and further control the robot to reach the destination.
[0081] Step S8, when there is a priori point cloud map in the robot and the matching error between the prior point cloud map and the currently processed point cloud data is less than or equal to the error threshold, the position and posture of the robot body is obtained according to the prior point cloud map, and return to step S5.
[0082] Using the posture of the robot body fed back by the prior point cloud map, continue to execute step S5: based on the posture of the robot body and the currently processed point cloud data, use a deep learning method to perform a passability analysis on the environment.
[0083] Reference Figure 2 The overall process of a robot autonomously passing method for outdoor harsh environments provided by an embodiment of the present invention is as follows:
[0084] Step 1: Calibrate the pose transformation matrix between the laser radar's own coordinate system and the robot and the field environment;
[0085] Step 2: Move the AVG car to the field operation area;
[0086] Step 3: Use LiDAR to collect initial point cloud data in the environment;
[0087] Step 4: Segment the initial point cloud data and remove noise to obtain processed point cloud data;
[0088] Step 5: Determine whether to enable real-time map building mode. If yes, proceed to the next step. If no, skip to step 7.
[0089] Step 6: Use the processed point cloud data to build an environment map in real time, and obtain the trajectory of the robot body, that is, the position and posture through the map;
[0090] Step 7: Use the point cloud data collected in real time to realize the real-time positioning of the robot;
[0091] Step 8: Use the robot posture and point cloud data collected in real time to perform passability analysis. If the passability analysis is successful, proceed to the next step. If it fails, the robot collects more point cloud data at the current position and makes another judgment until a stable passability analysis result is obtained.
[0092] Step 9: Control the robot path and body based on the results of the passability analysis.
[0093] The key points of the present invention are as follows:
[0094] 1. The present invention first calibrates itself in the robot body, laser radar and environment, and removes noise based on Euclidean clustering and K-Means clustering and segment objects in the environment for the point cloud data collected by the laser radar on the AGV. Compared with traditional point cloud data processing methods, the accuracy of point cloud segmentation and the stability of noise removal in the present invention are significantly enhanced, and it can be applied to complex, large-scale and harsh outdoor scenes.
[0095] 2. The present invention first divides the three-dimensional space into grids when the robot is positioned in real time, and further obtains the probability matching score of the current position and the environmental point cloud data collected by the sensor by calculating the parameters of the multi-dimensional normal distribution of the grid, and then obtains the position of the current position in the 3D point cloud map. Compared with the traditional positioning method of using ICP point cloud matching, the method proposed by the present invention can reduce the amount of calculation in the entire positioning process and improve the efficiency of positioning. Under the premise of the same data scale, the ICP matching calculation method takes twice as long as the method proposed by the present invention.
[0096] 3. The present invention identifies feature points in the environment based on three-dimensional point clouds, uses map data and the two frames of data before and after the radar feature points to calculate the odometer and fuses the built map data to estimate its own posture. Compared with the traditional method of collecting environmental feature points based on two-dimensional images, the present invention is more accurate in extracting feature depth information in the environment, thereby improving the accuracy of mapping and posture estimation.
[0097] 4. The present invention uses a deep learning method to analyze the robot's passability by combining the robot's posture information and environmental point cloud data. Compared with the traditional passability analysis method, the present invention reduces the uncertainty in the passability analysis, that is, the traditional passability analysis relies on the static map constructed by the system, which loses a lot of original information in the environment. At the same time, the present invention takes the robot's posture into overall consideration so that the robot will not be unable to pass through some special areas (such as ditches, etc.) due to some of its own structures when interacting with the environment.
[0098] The existing robot autonomous motion scheme uses laser radar or camera as data acquisition device, and the perception part of the entire robot autonomous system is limited by the amount of data and the validity of the data, resulting in a decrease in the robot's perception of the environment, that is, the accuracy of map construction and positioning is poor. In addition, the poor environmental perception ability makes the position of obstacles in the map relatively uncertain, resulting in poor passability analysis effect. It is a new point cloud preprocessing method to remove corner points, depth information error points, noise points and dynamic object points based on Euclidean clustering from the data collected by the sensor, and then perform real-time robot mapping and positioning, and then perform road passability analysis based on the mapping results. However, this method will cause point cloud redundancy at the same position, and the efficiency of passability analysis is related to the accuracy of map construction. In view of the above shortcomings, the method proposed in the present invention removes dynamic points and noise points based on Euclidean clustering and K-Means for the environmental point cloud data collected by the laser radar, adopts a probability-based model to solve the real-time positioning of the robot, and incorporates the robot's posture into the passability analysis, solving the problem that the robot has a large amount of point cloud data in the wild environment, cannot achieve real-time positioning, and has poor passability analysis accuracy and cannot learn and adapt.
[0099] The present invention also provides an autonomous robot for outdoor harsh environments, such as Figure 4 As shown, it includes: an AGV car 1, a switch 2, an industrial computer 3 and a laser radar 4.
[0100] The switch 2, the industrial computer 3 and the laser radar 4 are all arranged on the AGV 1. The signal output end of the laser radar 4 is connected to the industrial computer 3 through the switch 2, and the industrial computer 3 is connected to the control end of the AGV 1. The laser radar 4 is used to perform 3D point cloud scanning of the outdoor environment driven by the AGV 1, collect 3D point cloud data of the outdoor environment in real time, and transmit the 3D point cloud data to the industrial computer 3 through the switch 2. The industrial computer 3 is used to perform passability analysis based on the 3D point cloud data, and perform real-time path planning based on the results of the passability analysis, and then control the AGV 1 to reach the target position according to the planned path.
[0101] Exemplarily, a switch 2, an industrial computer 3 and a laser radar 4 are fixedly installed on the AGV 1 in sequence from bottom to top.
[0102] An autonomous robot for harsh outdoor environments can perform the aforementioned autonomous robot method for harsh outdoor environments. The laser radar performs step S1, and the industrial computer 3 performs steps S2 to S8. In step S7, the industrial computer 3 controls the AGV 1 to reach the target position according to the planned path.
[0103] The advantages of using switch 2 for processing are: 1) avoiding the real-time transmission of a large amount of data to the industrial computer 3, which would cause the entire system to crash and fail to work; 2) increasing the scalability of the robot system, allowing it to connect to other devices to complete development and make the entire robot function more complete.
[0104] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.
[0105] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of the present invention. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.
Claims
1. A method for autonomous robot movement in harsh outdoor environments, characterized in that: include: When the robot moves in the field operation area, it uses LiDAR to collect 3D point cloud data of the field environment in real time; The 3D point cloud data is segmented and noise removed by using the Euclidean clustering algorithm and the K-Means clustering algorithm to obtain the currently processed point cloud data; When there is no prior point cloud map in the robot or the matching error between the prior point cloud map and the currently processed point cloud data is greater than an error threshold, a three-dimensional point cloud map is constructed using the currently processed point cloud data, and the position and posture of the robot body is obtained according to the three-dimensional point cloud map; wherein, constructing a three-dimensional point cloud map using the currently processed point cloud data, and obtaining the position and posture of the robot body according to the three-dimensional point cloud map specifically includes: extracting features from two adjacent frames of point cloud data; matching the extracted features to construct a radar odometer and a three-dimensional point cloud map; and fusing the radar odometer and the three-dimensional point cloud map using a Kalman filter to obtain the position and posture of the robot body; According to the currently processed point cloud data, the current position of the robot body is located in a probabilistic manner; Based on the robot's position and the currently processed point cloud data, a deep learning method is used to analyze the environment's passability. If the passability analysis fails, the point cloud data of the outdoor environment is collected again by using the laser radar at the current position, and the process returns to step "using the Euclidean clustering algorithm and the K-Means clustering algorithm to segment and remove noise from the 3D point cloud data to obtain the currently processed point cloud data"; If the passability analysis is successful, a path is planned between the current position and the target position according to the result of the passability analysis, and the robot is controlled to reach the target position according to the planned path; When there is a priori point cloud map in the robot and the matching error between the prior point cloud map and the currently processed point cloud data is less than or equal to the error threshold, the position and posture of the robot body is obtained according to the prior point cloud map, and the process returns to the step of "based on the position and posture of the robot body and the currently processed point cloud data, using a deep learning method to perform a passability analysis on the environment".
2. The method for autonomous robot passage in harsh outdoor environments according to claim 1, characterized in that: The 3D point cloud data is segmented and noise removed using the Euclidean clustering algorithm and the K-Means clustering algorithm to obtain the currently processed point cloud data, which also includes: The robot hand-eye calibration method is used to calibrate the position transformation matrix between the laser radar's own coordinate system and the robot's base coordinate system and the field environment; According to the position coordinates of the point cloud data, using the formula Calculate the world coordinates of the robot base coordinate system 3 p i ; In the formula, is the pose transformation matrix between the laser radar’s own coordinate system and the robot’s base coordinate system, 4T is the pose transformation matrix between the laser radar’s own coordinate system and the field environment, and p i is the location coordinate of the point cloud data.
3. The method for autonomous robot passage in harsh outdoor environments according to claim 1, characterized in that: According to the currently processed point cloud data, the current position of the robot body is located in a probabilistic manner, specifically including: Performing three-dimensional grid division on the currently processed point cloud data; Calculate the multidimensional normal distribution parameters for each grid; Convert the currently processed point cloud data into point cloud data in the robot base coordinate system; Determine the probability distribution model based on the point cloud data and multi-dimensional normal distribution parameters in the robot base coordinate system; The current position of the robot body is located according to the score of the probability distribution model.
4. The method for autonomous robot passage in harsh outdoor environments according to claim 1, characterized in that: Based on the robot's position and the currently processed point cloud data, a deep learning method is used to perform a passability analysis of the environment, including: The robot body’s posture and the currently processed point cloud data are input into a convolutional neural network, and the results of the passability analysis are output.
5. The method for autonomous robot passage in harsh outdoor environments according to claim 1, characterized in that: Use LiDAR to collect point cloud data of the outdoor environment again at the current location, including: At the current position, the robot rotates or moves itself and uses the lidar to collect point cloud data of the field environment again.
6. The method for autonomous robot passage in harsh outdoor environments according to claim 1, characterized in that: Based on the results of the passability analysis, a path is planned between the current location and the target location, including: According to the results of the passability analysis, the A* algorithm or the D* algorithm is used to plan the path between the current position and the target position.
7. An autonomous robot that can pass through harsh outdoor environments, characterized in that: The autonomous robot for harsh outdoor environments is used to execute the autonomous robot method for harsh outdoor environments described in any one of claims 1 to 6, and the autonomous robot for harsh outdoor environments comprises: an AGV, a switch, an industrial computer, and a laser radar; The switch, industrial computer and laser radar are all installed on the AGV; The signal output end of the laser radar is connected to the industrial computer through a switch, and the industrial computer is connected to the control end of the AGV car; The laser radar is used to perform 3D point cloud scanning of the outdoor environment driven by the AGV, collect 3D point cloud data of the outdoor environment in real time, and transmit the 3D point cloud data to the industrial computer through the switch; The industrial computer is used to perform a passability analysis based on the 3D point cloud data, and to perform real-time path planning based on the result of the passability analysis, and then to control the AGV vehicle to reach the target position according to the planned path.
8. The autonomous traversable robot for outdoor harsh environments according to claim 7, characterized in that: A switch, an industrial computer and a laser radar are fixedly installed on the AGV vehicle in sequence from bottom to top.
Citation Information
Patent Citations
Complex environment-oriented passable area detection method and system
CN112395956A
Large-scale park mapping positioning method based on multi-sensor touch
CN112967392A
Method for extracting Buddha niche from grotto point cloud
CN114429537A
Unknown environment autonomous exploration method suitable for air-ground dual-purpose robot
CN114578824A
Road marking segmentation method and system based on vehicle-mounted laser point cloud data
CN114863376A
