Aircraft control method based on L1 adaptive control
By integrating two-dimensional lidar and depth cameras on UAVs, and using improved Bayesian fusion algorithms and OctoMap for data fusion and three-dimensional modeling, the problem of UAV obstacle avoidance systems being difficult to identify long-distance obstacles is solved. Improved the autonomous navigation and obstacle avoidance capabilities of UAVs.
Patent Information
- Application Number
- CN202411497139.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-25
- Publication Date
- 2025-01-17
AI Technical Summary
The existing UAV obstacle avoidance system is difficult to identify long-distance obstacles in a timely manner. The limited computing power leads to slow response speed, which is prone to false alarms or missed reports, increasing the risk of collision.
It adopts an aircraft control method based on L1 adaptive control, integrates two-dimensional lidar and depth cameras, and achieves comprehensive capture and precise synchronization of environmental information through spatial synchronization and time synchronization sensors. Use the improved Bayesian fusion algorithm and OctoMap for multi-sensor data fusion and three-dimensional modeling to improve data accuracy and computing efficiency.
It enhances the adaptability and autonomous navigation capabilities of UAVs in complex environments, improves the response speed and accuracy of the obstacle avoidance system, reduces false alarms and omissions, and reduces the risk of collisions.
Smart Images
Figure CN119322523A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of aviation technology, and in particular to an aircraft control method based on L1 adaptive control. Background Art
[0002] Drone obstacle avoidance technology is one of the important directions of current drone technology development. Its background mainly stems from the need for safe flight of drones in complex environments. With the increasing application of drones in aerial photography, environmental monitoring, agriculture, rescue and other fields, the flight environment faced by drones is becoming increasingly complex. In order to ensure the safe flight of drones in these environments, obstacle avoidance technology has become an indispensable key technology.
[0003] At present, the obstacle avoidance systems of many drones can only sense obstacles at a short distance, and may not be able to identify obstacles at a long distance in time. This increases the risk of collision between drones and obstacles during flight. The obstacle avoidance system needs to process and analyze a large amount of data in real time, which requires high computing power. However, the computing power of drones may be limited, resulting in a slower response speed of the obstacle avoidance system, or even failure to respond in time. The obstacle avoidance system may have false alarms or missed alarms due to sensor errors, imperfect algorithms, or environmental interference. False alarms may cause the drone to perform unnecessary obstacle avoidance actions, while missed alarms may cause the drone to fail to identify and avoid obstacles in time. To this end, we propose an aircraft control method based on L1 adaptive control to solve the above problems. Summary of the invention
[0004] The object of the present invention is to provide an aircraft control method based on L1 adaptive control to solve the problems raised in the above background technology.
[0005] The technical solution of the present invention is: an aircraft control method based on L1 adaptive control, including an aircraft, on which a two-dimensional laser radar and a depth camera are arranged, the two-dimensional laser radar and the depth camera signals are connected to a space synchronization sensor, the space synchronization sensor signal is connected to a time synchronization sensor, the time synchronization sensor is electrically connected to a data preprocessing sensor, the data preprocessing signal is connected to an improved Baye new fusion algorithm sensor, the improved Baye new fusion algorithm sensor is communicatively connected to a fused point cloud, and the fused point cloud is communicatively connected to OctoMap.
[0006] Preferably, step S1: the two-dimensional laser radar and the depth camera are installed at different positions of the drone, and the two-dimensional laser radar and the depth camera are used to simultaneously observe the same position P in the space, and the coordinates of point P in the depth camera coordinate system are set to ( , ), the coordinates in the two-dimensional laser radar coordinate system are ( , , ), the conversion relationship between the depth camera coordinates and the two-dimensional lidar coordinates can be obtained as follows: ; Step S2: Solve the R and T matrices by the least squares method, and then determine the rotation and translation relationship between the 2D laser radar coordinate system and the depth camera coordinate system to complete the joint calibration. Under the ROS platform, use the static_transform_publisher tool in the tf2_ros software package to perform static coordinate transformation, convert the 2D laser radar and depth camera to the same coordinate system, and achieve spatial synchronization; Step S3: directly collect point cloud information in the environment through the depth camera, and use statistical filtering and voxel filtering to process the point cloud collected by the depth camera to filter out isolated noise points caused by the jitter of the drone, thereby reducing the number of point clouds without destroying the appearance structure of the point cloud; Step S4: The laser radar data sequence collected by the two-dimensional laser radar needs to be converted into a point cloud data format, and then statistical filtering is performed. The laser radar data sequence ( , )T and its point cloud data format ( , , ) are as follows: ; Preferably, the improved Bayesian fusion algorithm steps are as follows: Step S1: Use the observation vector Z to infer the unknown state vector X. Assume that in a state space, the observation value is known , and the probability of the state vector X at time k is , its posterior distribution is as follows: ; Step S2: When the point cloud data is updated by using the Bayesian estimation method, O indicates that the point cloud is observed to exist, O indicates that the point cloud is observed to not exist, E indicates that the point cloud actually exists, and E indicates that the point cloud does not exist. The posterior probability is as follows: ; ; The point cloud information of the 2D laser radar is fused with the point cloud information of the depth camera, and the detection range problem of the 2D laser radar is compensated. The probability formula for updating the fused point cloud is as follows: ; Step S3: Based on the indoor environment and sensor characteristics, the update of the fused point cloud is mainly based on the point cloud information of the two-dimensional laser radar, and the point cloud information of the depth camera is used to compensate for the area that the two-dimensional laser radar cannot detect. Suppose the coordinates of a certain point cloud of the laser radar ( , , ), the coordinates of a certain point cloud of the depth camera ( , , ), then the distance between the two point clouds for ; The likelihood function of the improved Bayesian fusion algorithm is transformed from the sigmoid function, as shown below: ; Step S4: OctoMap uses the octree data structure to divide the environment into voxels of different sizes and binarizes each voxel to represent its occupancy state; Step S5: OctoMap uses the octree data structure to divide the environment into voxels of different sizes, and binarizes each voxel to represent its occupancy status. The stored octree map represents the occupied or idle state in the environment. Under the ROS platform, the fused point cloud is converted into an octree map by calling the octomap_server function package, so that the drone can autonomously avoid obstacles.
[0007] Preferably, in the formula, R represents the transformation rotation matrix from the depth camera coordinate system to the two-dimensional laser radar coordinate system, and T represents the transformation translation matrix from the depth camera coordinate system to the two-dimensional laser radar coordinate system.
[0008] Preferably, in the formula Indicates the distance from the coordinate center of the two-dimensional laser radar to a target point in the measured environment. It represents the angle between the coordinate center of the 2D laser radar and a target point in the measured environment. h represents the sum of the flight altitude of the drone and the installation height of the 2D laser radar. represents the likelihood function of the observation model, represents the prior distribution function of the transformation system model, represents the normalized probability density function, P represents the probability value of the fused point cloud update, and They represent the prior probability of the existence of point clouds in indoor environments, with a size of 0.5, Represents the likelihood probability of the depth camera detecting the existence of the point cloud.
[0009] Preferably, it is characterized in that: Indicates the maximum distance value, and its value is 0.1m.
[0010] The present invention provides a grass mosaic aviation through improvement, which has the following improvements and advantages compared with the prior art: First, in the present invention, the aircraft integrates a two-dimensional laser radar and a depth camera, and with the help of spatial synchronization and temporal synchronization sensors, realizes comprehensive capture and precise synchronization of environmental information, which not only enhances the adaptability of the aircraft in complex environments, but also improves its autonomous navigation and obstacle avoidance capabilities. By introducing data preprocessing sensors, the aircraft effectively improves data quality and reduces the adverse effects caused by sensor noise or errors. At the same time, the application of the improved Baye new fusion algorithm sensor further improves the efficiency and accuracy of data fusion, enabling the aircraft to more accurately identify targets, analyze the environment and make decisions. OctoMap is used to perform three-dimensional modeling of the fused point cloud, which not only improves the accuracy and readability of the map, but also provides strong data support for tasks such as aircraft path planning and target tracking. This efficient three-dimensional modeling method helps the aircraft better understand and adapt to complex and changing environments, thereby improving its overall performance.
[0011] Second: The present invention improves the autonomous navigation and obstacle avoidance capabilities of the aircraft by optimizing data quality and realizing deep fusion of multi-sensor data. The efficient three-dimensional modeling method also provides strong data support for subsequent applications of the aircraft. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] The present invention will be further explained below in conjunction with the accompanying drawings and embodiments: Figure 1 It is a flow chart of the autonomous obstacle avoidance method of a UAV based on multi-sensor fusion of the present invention; Figure 2 is a flow chart of the UAV trajectory replanning based on the A ∗ algorithm of the present invention; Figure 3 is a relationship diagram between the likelihood estimate value of the present invention and the distance value between two point clouds; Figure 4 It is a drone trajectory map. DETAILED DESCRIPTION
[0013] The present invention is described in detail below, and the technical solutions in the embodiments of the present invention are clearly and completely described. 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.
[0014] The present invention provides an aircraft control method based on L1 adaptive control through improvement. The technical solution of the present invention is: like Figure 1 - Figure 4As shown, an aircraft control method based on L1 adaptive control includes an aircraft, on which a two-dimensional laser radar and a depth camera are arranged, the two-dimensional laser radar and the depth camera signals are connected with a space synchronization sensor, the space synchronization sensor signal is connected with a time synchronization sensor, the time synchronization sensor is electrically connected with a data preprocessing sensor, the data preprocessing signal is connected with an improved Baye new fusion algorithm sensor, the improved Baye new fusion algorithm sensor communication is connected with a fused point cloud, the fused point cloud communication is connected with OctoMap, and the point cloud collected by using two-dimensional laser radar or depth camera alone will have outliers caused by noise. After filtering, the maximum offset of the point cloud obtained by simple superposition for the point cloud collected by the two-dimensional laser radar and the point cloud collected by the depth camera was reduced by 75.0% and 81.7%, respectively, while the maximum offset of the fused point cloud obtained by this method was reduced by 25.5% compared with the point cloud obtained by simple superposition. The Hausdorff distance of the fused point cloud obtained by this method was reduced by 19.4%, 41.5% and 9.81% compared with the point cloud collected by the two-dimensional laser radar, the point cloud collected by the depth camera and the point cloud obtained by simple superposition, respectively, and the root mean square error was reduced by 39.2%, 47.6% and 13.3%, respectively. In summary, the difference between the fused point cloud obtained by this method and the true value of the obstacle point cloud is small, the similarity is high, and the accuracy of the surrounding environment composition can be guaranteed.
[0015] Further, step S1: the two-dimensional laser radar and the depth camera are installed at different positions of the drone, and the two-dimensional laser radar and the depth camera are used to simultaneously observe the same position P in the space, and the coordinates of point P in the depth camera coordinate system are set to ( , ), the coordinates in the two-dimensional laser radar coordinate system are ( , , ), the conversion relationship between the depth camera coordinates and the two-dimensional lidar coordinates can be obtained as follows: ; Step S2: Solve the R and T matrices by the least squares method, and then determine the rotation and translation relationship between the 2D laser radar coordinate system and the depth camera coordinate system to complete the joint calibration. Under the ROS platform, use the static_transform_publisher tool in the tf2_ros software package to perform static coordinate transformation, convert the 2D laser radar and depth camera to the same coordinate system, and achieve spatial synchronization; Step S3: directly collect point cloud information in the environment through the depth camera, and use statistical filtering and voxel filtering to process the point cloud collected by the depth camera to filter out isolated noise points caused by the jitter of the drone, thereby reducing the number of point clouds without destroying the appearance structure of the point cloud; Step S4: The laser radar data sequence collected by the two-dimensional laser radar needs to be converted into a point cloud data format, and then statistical filtering is performed. The laser radar data sequence ( , )T and its point cloud data format ( , , ) are as follows: ; Furthermore, the steps of the improved Bayesian fusion algorithm are as follows: Step S1: Use the observation vector Z to infer the unknown state vector X. Assume that in a state space, the observation value is known , and the probability of the state vector X at time k is , its posterior distribution is as follows: ; Step S2: When the point cloud data is updated by using the Bayesian estimation method, O indicates that the point cloud is observed to exist, O indicates that the point cloud is observed to not exist, E indicates that the point cloud actually exists, and E indicates that the point cloud does not exist. The posterior probability is as follows: ; ; The point cloud information of the 2D laser radar is fused with the point cloud information of the depth camera, and the detection range problem of the 2D laser radar is compensated. The probability formula for updating the fused point cloud is as follows: ; Step S3: Based on the indoor environment and sensor characteristics, the update of the fused point cloud is mainly based on the point cloud information of the two-dimensional laser radar, and the point cloud information of the depth camera is used to compensate for the area that the two-dimensional laser radar cannot detect. Suppose the coordinates of a certain point cloud of the laser radar ( , , ), the coordinates of a certain point cloud of the depth camera ( , , ), then the distance between the two point clouds for ; The likelihood function of the improved Bayesian fusion algorithm is transformed from the sigmoid function, as shown below: ; Step S4: OctoMap uses the octree data structure to divide the environment into voxels of different sizes and binarizes each voxel to represent its occupancy state; Step S5: OctoMap uses the octree data structure to divide the environment into voxels of different sizes, and performs binary processing on each voxel to represent its occupancy status. The stored octree map represents the occupied or idle state in the environment. Under the ROS platform, the fused point cloud is converted into an octree map by calling the octomap_server function package, so that the UAV can autonomously avoid obstacles and meet the real-time requirements of the UAV in an unknown environment. First, it is necessary to use the A∗ algorithm to generate a global optimal reference track at the beginning of the UAV's flight mission. During the flight of the UAV, its sensors continuously collect information about the surrounding environment in order to identify and avoid obstacles in a timely manner. If the UAV detects a new obstacle in the surrounding environment and the planned track may collide with the obstacle, it is necessary to re-plan the track. At this time, taking the current position as the starting point, the A∗ algorithm is used to re-search and generate a new track to avoid the new obstacle until the UAV reaches the end point. This strategy can ensure that the UAV responds in time when encountering obstacles in an unknown environment to ensure flight safety. The track replanning process based on the A∗ algorithm is as follows: Figure 2 shown.
[0016] Furthermore, R in the formula represents the transformation rotation matrix from the depth camera coordinate system to the two-dimensional lidar coordinate system, and T represents the transformation translation matrix from the depth camera coordinate system to the two-dimensional lidar coordinate system.
[0017] Furthermore, in the formula Indicates the distance from the coordinate center of the two-dimensional laser radar to a target point in the measured environment. It represents the angle between the coordinate center of the 2D laser radar and a target point in the measured environment. h represents the sum of the flight altitude of the drone and the installation height of the 2D laser radar. represents the likelihood function of the observation model, represents the prior distribution function of the transformation system model, represents the normalized probability density function, P represents the probability value of the fused point cloud update, and They represent the prior probability of the existence of point clouds in indoor environments, with a size of 0.5, Represents the likelihood probability of the depth camera detecting the existence of the point cloud.
[0018] Further, Represents the maximum distance value, and the size is 0.1m. When the distance between the two point clouds is greater than the maximum distance, the update of the fused point cloud is mainly based on the point cloud information of the depth camera. The likelihood function uses the distance value between the two-dimensional lidar point cloud and the depth camera point cloud to determine the size of the likelihood estimate, such as Figure 3As shown in , this function ensures that when the distance is very small, the corresponding likelihood estimate is close to 0, that is, at this time, the fused point cloud is mainly based on the point cloud information of the 2D laser radar. As the distance between the 2D laser radar point cloud and the depth camera point cloud increases, the likelihood estimate gradually increases to a maximum value of 1, that is, at this time, the 2D laser radar point cloud information is supplemented by the depth camera point cloud information to obtain the fused point cloud. The improved Bayesian fusion algorithm can process complementary or redundant information from multiple sensors, remove duplicate or erroneous point clouds, and improve the accuracy of the fused point cloud.
[0019] The above description enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. An aircraft control method based on L1 adaptive control, comprising an aircraft, characterized in that: A two-dimensional laser radar and a depth camera are arranged on the aircraft, the two-dimensional laser radar and the depth camera signals are connected with a space synchronization sensor, the space synchronization sensor signal is connected with a time synchronization sensor, the time synchronization sensor is electrically connected with a data preprocessing sensor, the data preprocessing signal is connected with an improved Baye new fusion algorithm sensor, the improved Baye new fusion algorithm sensor is communicatively connected with a fused point cloud, and the fused point cloud is communicatively connected with OctoMap.
2. The aircraft control method based on L1 adaptive control according to claim 1, characterized in that: Step S1: The two-dimensional laser radar and the depth camera are installed at different positions on the drone. The two-dimensional laser radar and the depth camera are used to simultaneously observe the same position P in the space. The coordinates of point P in the depth camera coordinate system are set to ( , ), the coordinates in the two-dimensional laser radar coordinate system are ( , , ), the conversion relationship between the depth camera coordinates and the two-dimensional lidar coordinates can be obtained as follows: ； Step S2: Solve the R and T matrices by the least squares method, and then determine the rotation and translation relationship between the 2D laser radar coordinate system and the depth camera coordinate system to complete the joint calibration. Under the ROS platform, use the static_transform_publisher tool in the tf2_ros software package to perform static coordinate transformation, convert the 2D laser radar and depth camera to the same coordinate system, and achieve spatial synchronization; Step S3: directly collect point cloud information in the environment through the depth camera, and use statistical filtering and voxel filtering to process the point cloud collected by the depth camera to filter out isolated noise points caused by the jitter of the drone, thereby reducing the number of point clouds without destroying the appearance structure of the point cloud; Step S4: The laser radar data sequence collected by the two-dimensional laser radar needs to be converted into a point cloud data format, and then statistical filtering is performed. The laser radar data sequence ( , )T and its point cloud data format ( , , ) are as follows: 。 3. The aircraft control method based on L1 adaptive control according to claim 1, characterized in that: The steps of the improved Bayesian fusion algorithm are as follows: Step S1: Use the observation vector Z to infer the unknown state vector X. Assume that in a state space, the observation value is known , and the probability of the state vector X at time k is , its posterior distribution is as follows: ; Step S2: When the point cloud data is updated by using the Bayesian estimation method, O indicates that the point cloud is observed to exist, O indicates that the point cloud is observed to not exist, E indicates that the point cloud actually exists, and E indicates that the point cloud does not exist. The posterior probability is as follows: ; ; The point cloud information of the 2D laser radar is fused with the point cloud information of the depth camera, and the detection range problem of the 2D laser radar is compensated. The probability formula for updating the fused point cloud is as follows: ； Step S3: Based on the indoor environment and sensor characteristics, the update of the fused point cloud is mainly based on the point cloud information of the two-dimensional laser radar, and the point cloud information of the depth camera is used to compensate for the area that the two-dimensional laser radar cannot detect. Suppose the coordinates of a certain point cloud of the laser radar ( , , ), the coordinates of a certain point cloud of the depth camera ( , , ), then the distance between the two point clouds for ； The likelihood function of the improved Bayesian fusion algorithm is transformed from the sigmoid function, as shown below: ； Step S4: OctoMap uses the octree data structure to divide the environment into voxels of different sizes and binarizes each voxel to represent its occupancy state; Step S5: OctoMap uses the octree data structure to divide the environment into voxels of different sizes, and binarizes each voxel to represent its occupancy status. The stored octree map represents the occupied or idle state in the environment. Under the ROS platform, the fused point cloud is converted into an octree map by calling the octomap_server function package, so that the drone can autonomously avoid obstacles.
4. The aircraft control method based on L1 adaptive control according to claim 2, characterized in that: In the formula, R represents the transformation rotation matrix from the depth camera coordinate system to the two-dimensional lidar coordinate system, and T represents the transformation translation matrix from the depth camera coordinate system to the two-dimensional lidar coordinate system.
5. The aircraft control method based on L1 adaptive control according to claim 3, characterized in that: In the formula Indicates the distance from the coordinate center of the two-dimensional laser radar to a target point in the measured environment. It represents the angle between the coordinate center of the 2D laser radar and a target point in the measured environment. h represents the sum of the flight altitude of the drone and the installation height of the 2D laser radar. represents the likelihood function of the observation model, represents the prior distribution function of the transformation system model, represents the normalized probability density function, P represents the probability value of the fused point cloud update, and They represent the prior probability of the existence of point clouds in indoor environments, with a size of 0.5, Represents the likelihood probability of the depth camera detecting the existence of the point cloud.
6. The aircraft control method based on L1 adaptive control according to claim 3, characterized in that: Said Indicates the maximum distance value, and its size is 0.1m.