Limited space autonomous exploration unmanned aerial vehicle system and method based on multi-vehicle cooperation

By adopting multi-machine collaboration and the combination of vision sensors and lidar sensors in the UAV system, the autonomous exploration and mapping of drones in confined spaces is achieved in stages, and the limitation of drones in the existing technology is solved. The robustness and accuracy issues of space autonomous exploration are suitable for autonomous navigation and task execution in complex environments.

CN119088071APending Publication Date: 2024-12-06SUZHOU TUOYIZHE INTELLIGENT TECHNOLOGY CO LTD

Patent Information

Application Number
CN202411307894.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-19
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing multi-machine collaborative SLAM algorithms are difficult to achieve real-time mutual observation and large-bandwidth environment-aware data interaction in confined spaces, resulting in the failure of autonomous drones' exploration and unable to meet the needs of emergency disaster relief and military reconnaissance.

Method used

Adopt multi-machine collaboration and confined space autonomously to explore the drone system, through the drone nodes and central command stations that communicate with each other, use vision sensors and lidar sensors for autonomous navigation and mapping, which is divided into initial stages and autonomous exploration and rough construction. The model phase and offline precise modeling phase ensure that the system operates robustly in the event of communication loss between drones.

Benefits of technology

It realizes autonomous navigation and mission execution of drones in complex, narrow or dangerous environments, can establish rough real-time maps on the central command station, and improve map accuracy through offline precise modeling stages, suitable for emergency search and rescue and military operations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119088071A_ABST
    Figure CN119088071A_ABST
Patent Text Reader

Abstract

The invention discloses a limited space autonomous exploration unmanned aerial vehicle system and method based on multi-vehicle cooperation, and the system comprises a plurality of unmanned aerial vehicle nodes and a central command station which communicate with each other. A plurality of unmanned aerial vehicles autonomously carry out positioning and mapping in a limited space through airborne visual sensors and laser radar sensors, autonomously plan an exploration trajectory, traverse the limited space, and establish a rough real-time map on a central command station; after the unmanned aerial vehicle returns, the sensor original data stored by the multiple unmanned aerial vehicles are merged and imported into the central command station, and an accurate map model is calculated and established. According to the method, a collaborative exploration process is divided into a real-time autonomous exploration stage which only depends on local map information of each unmanned aerial vehicle node and an offline accurate modeling stage which needs global information, the previous process does not depend on real-time large-bandwidth data communication between the unmanned aerial vehicles, and the next process is carried out after the unmanned aerial vehicles return. Therefore, the method has the feasibility of practical engineering application.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to an autonomous exploration unmanned aerial vehicle system and method, and in particular to an autonomous exploration unmanned aerial vehicle system and method for confined space based on multi-machine collaboration, and belongs to the technical field of unmanned aerial vehicles. Background Art

[0002] The UAV autonomous exploration system is a complex system that integrates multiple advanced technologies, designed to enable UAVs to autonomously perform tasks, explore unknown environments and make decisions without human control. It has good application prospects in many fields, including but not limited to military reconnaissance, environmental monitoring, traffic monitoring, agriculture, wildlife protection, disaster response and search and rescue operations.

[0003] The invention patent with application number 2020100510052 discloses a detection and positioning system and method for drones, including: a central processing station and multiple reconnaissance stations; the central processing station includes: a first satellite module, a first processor, a first satellite antenna, and a first data link module; wherein the first satellite module and the first data link module are respectively connected to the first processor; each reconnaissance station includes: a second satellite module, a second processor, a second satellite antenna, an air-to-ground link signal detection module, a detection antenna, a timestamp calibration module, and a second data link module; wherein the air-to-ground link signal detection module is respectively connected to the detection antenna and the timestamp calibration module, and the second satellite module, the air-to-ground link signal detection module, the timestamp calibration module, and the second data link module are respectively connected to the second processor. This patented technology can improve the detection and positioning accuracy of drones. However, most drone systems in the above patents rely on GNSS systems to obtain positioning, and can only work in outdoor open spaces, and cannot work in confined spaces such as indoors, tunnels, pipelines, tanks, and underground.

[0004] With the advancement of technology, some manufacturers have designed drone products that are equipped with a laser radar and run SLAM algorithms, so that they can obtain positioning and fly in confined spaces. Existing multi-machine collaborative SLAM algorithms often rely on mutual observation between multiple machines and mutual communication of surrounding environment perception data, but in actual application scenarios, these two conditions are often not met, resulting in the existing multi-machine collaborative SLAM algorithms remaining only in the academic stage and difficult to be applied in engineering. For example, specifically for autonomous exploration of confined spaces, each drone node often flies to different floors and different tunnel forks, and is invisible to each other, which simply cannot meet the needs of continuous real-time mutual observation and continuous large-bandwidth environmental perception data among multiple machines. Most of the existing multi-machine collaborative SLAM algorithms rely on real-time mutual observation and mutual communication data for posture optimization. Once the nodes enter a mutually invisible state, the SLAM algorithm is easy to diverge, which leads to the failure of autonomous exploration of drones.

[0005] It can be seen that the existing UAV system has low detection efficiency and cannot meet the needs of emergency rescue, military reconnaissance, etc., which require traversing a confined space and establishing a map model in a short time. In view of the above situation, there is an urgent need for a UAV system that can collaborate with multiple aircraft to conduct autonomous flight exploration in confined spaces. Summary of the invention

[0006] In order to address the deficiencies of the prior art, the purpose of the present invention is to provide a drone system and method for autonomous exploration of confined spaces based on multi-machine collaboration. The system and method can autonomously navigate and perform tasks in complex, narrow or dangerous environments, and are feasible for practical engineering applications.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] The present invention first discloses a drone system for autonomous exploration of confined space based on multi-machine collaboration, which includes: a plurality of drone nodes and a central command station that communicate with each other, on which a multi-node real-time online collaborative SLAM algorithm and an offline high-precision SLAM modeling algorithm are run; the drone node includes: a flight controller, an actuator, a plurality of visual sensors and laser radar sensors, a high-performance processor, a memory, and a multi-point communication device, the visual sensor is used to collect image information of the surrounding environment, the laser radar sensor is used to collect point cloud data of the surrounding environment, the high-performance processor runs the local SLAM mapping and traversal exploration algorithm of each drone node, and is connected to the visual sensor, the laser radar sensor, the memory, and the multi-point communication device, and the high-performance processor controls the flight controller, and then sends instructions to the actuator. The visual sensor and the laser radar sensor are connected to the high-performance processor via Ethernet, MIPI or USB high-speed data interface.

[0009] Preferably, the aforementioned memory is used to store raw data information of various sensors carried by the drone, and the multi-point communication device is used to establish data communication between each drone and the central command station.

[0010] The present invention also discloses an exploration method for the aforementioned multi-machine collaborative autonomous exploration UAV system in a confined space, which includes three stages:

[0011] (1) The initial stage is used to detect the initial posture of each UAV and initialize the initial value of each SLAM node;

[0012] (2) Autonomous exploration and coarse modeling stage, each UAV node conducts autonomous exploration and builds a real-time coarse model on the central command station. This step is robust to communication loss between nodes.

[0013] (3) In the offline precise modeling stage, each UAV node transmits or copies the data to the central command station after returning to build a precise model offline.

[0014] Preferably, in the aforementioned initial stage, the drones are concentrated near the central node to wait for take-off, and one or more laser cooperation targets may also be deployed in the environment to enhance the visibility of the surrounding environment of each drone in the initial stage.

[0015] More preferably, the specific workflow of the aforementioned initial stage is:

[0016] S1.1. Keep each drone node stationary for a few seconds to obtain the zero bias and gravity observation of the IMU sensor data on the drone. The synchronized drone node will accumulate the visual and lidar observation data for a few seconds, and integrate to obtain the initial observation P0 of the surrounding environment point cloud data. i , where the subscript 0 represents the initial moment, and the superscript i represents the drone node number. The drones are arranged in the order of power-on. The first drone powered on is recorded as 0, and the subsequent drones powered on are numbered from 1 to n. The body coordinate system of the drone numbered 0 when it is powered on is the global coordinate system;

[0017] S1.2, each drone node downsamples its point cloud observation of the surrounding environment Transmitted to the central command station; at this time, each drone node has not yet set off and is still near the central node, so the transmission is feasible.

[0018] S1.3. The central command station collects the point cloud observation data sampled by each drone. Point cloud observation data sampled by the first drone Match each other to get the relative position of each drone before departure and the transformation matrix in Refers to the position vector of the i-th UAV in global coordinates at the initial moment, Refers to the attitude angle quaternion of the i-th UAV in the global coordinate system at the initial moment, Refers to the transfer matrix from the i-th UAV body coordinates to the global coordinate system at the initial moment;

[0019] S1.4. The central command station sends the relative position of each drone before departure back to each drone node to initialize the initial position of each SLAM node state variable.

[0020] More preferably, the specific workflow of the aforementioned autonomous exploration stage is:

[0021] S2.0, each UAV and the central command station rasterize the space to be explored into three-dimensional voxel grids, and initialize all grids as unreached;

[0022] S2.1, each drone node runs its own SLAM algorithm and updates its status through IMU sensor data. Status transmission to Where k refers to the current moment, and k-1 refers to the previous moment;

[0023] S2.2. Each UAV node matches the point cloud observation data to its own local map M locali Observe and update to optimize its own posture information and the transformation matrix And use this posture information to control the posture of the drone. Refers to the transfer matrix from the i-th UAV body coordinates to the global coordinate system at the current moment k;

[0024] S2.3. Each drone node maintains and splices its own local map:

[0025] S2.4. Each drone node sends its own position information Transformation Matrix And the downsampled observation point cloud data Transmit to the central command station;

[0026] S2.5, the central command station receives the transformation matrix in S2.2 The new point cloud data observed by each drone Transform to the global coordinate system and stitch them into a rough real-time global map:

[0027] S2.6. The central command station will provide a rough real-time global map Map to the three-dimensional voxel grid map, and mark the mapped grids as arrived, while the rest of the grids remain in their original state;

[0028] S2.7, the central command station broadcasts the updated information of the grid that has arrived to each drone node;

[0029] S2.8. Each drone node searches for grids near itself according to the distance until it finds the nearest unreached grid, then guides the drone to fly in that direction and repeats the work from step S2.1; if all surrounding grids have been reached, the drone is controlled to return.

[0030] Further preferably, each of the three-dimensional voxel grids can store an exploration state, and the exploration state is divided into: arrived and not arrived.

[0031] Further preferably, the specific workflow of the aforementioned offline precise modeling stage is:

[0032] S3.1. The central command station controls each UAV node to return to the initial position, and the original sensor data of each UAV is transmitted to the central command station;

[0033] S3.2, according to the initial posture observed in S1.3, assign initial values ​​to each sub-SLAM algorithm;

[0034] S3.3. Splice the initial observation point clouds of each node as the initial accurate global map:

[0035] S3.4, in turn according to the sampling time sequence of each node, perform the update step for each sub-SLAM algorithm according to the IMU data of each sub-SLAM node, Status transmission to

[0036] S3.5, perform the observation step for each sub-SLAM algorithm in turn according to the sampling time sequence of each node, match the observed point cloud to the refined global map, and optimize the pose accuracy in the state variable. The difference between this step and S2.2 is that the point cloud observation will be in the entire refined global map M globalP Therefore, it is possible to better match the point cloud data perceived by multiple drones;

[0037] S3.6. Attach a detailed global map at each step:

[0038] S3.7, perform loop detection and graph optimization to further improve pose and modeling accuracy;

[0039] S3.8, loop to S3.3 and restart the process until all sensor data are processed;

[0040] S3.9, output the final global map M globalP , as a refined global model.

[0041] Preferably, in step S3.1, the original sensor data of each UAV is transmitted or copied to the central command station via a data link or a removable storage medium.

[0042] The present invention is beneficial in that:

[0043] The innovative multi-machine cooperative confined space autonomous exploration drone system of the present invention can, under the control of the central command station, enable multiple drones to autonomously locate and build maps in the confined space through the onboard visual sensors and lidar sensors, autonomously plan exploration trajectories, traverse the confined space, and build a rough real-time map on the central command station. After the drones return, the raw sensor data stored by the multiple drones are merged and imported into the central command station, and an accurate map model is calculated and established, which can realize autonomous navigation and execution of tasks in complex, narrow or dangerous environments, and is used for emergency search and rescue and military operations command.

[0044] Compared with other similar systems in the prior art, the present invention divides the collaborative exploration process into a real-time autonomous exploration phase that only relies on the local map information of each drone node and an offline precise modeling phase that requires global information. The former process does not need to rely on real-time, high-bandwidth data communication between drones, and the latter process is performed after the drone returns, so the present invention has the feasibility of practical engineering applications. Other similar systems require real-time data interaction of observation information between drone nodes to perform position calculation and modeling during the exploration phase, but in the confined space exploration scenario targeted by the present invention, this transmission is not feasible due to the obstruction of terrain, walls, and obstacles, so other similar systems are difficult to be applied in the scenario of the present invention without targeted optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a structural block diagram of a drone in a multi-machine cooperative confined space autonomous exploration drone system of the present invention;

[0046] Figure 2 This is the first phase workflow diagram of the multi-machine collaborative confined space autonomous exploration UAV system of the present invention;

[0047] Figure 3 This is the second phase workflow diagram of the multi-machine collaborative confined space autonomous exploration UAV system of the present invention;

[0048] Figure 4 This is the third phase workflow diagram of the multi-machine collaborative confined space autonomous exploration UAV system of the present invention. DETAILED DESCRIPTION

[0049] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] Example 1

[0051] This embodiment discloses an autonomous exploration drone system for confined spaces based on multi-drone collaboration, which includes: a number of drone nodes and a central command station, wherein the central command station communicates with the drone nodes to coordinate and command multiple drones to perform exploration work.

[0052] The block diagram of each drone node is as follows: Figure 1 As shown, they all include: a flight controller, an actuator (consisting of a motor, a driver and blades), a number of visual sensors and lidar sensors, a high-performance processor, a memory and a multi-point communication device. Among them, the visual sensor is used to collect image information of the surrounding environment, and the lidar sensor is used to collect point cloud data of the surrounding environment. Both are connected to the high-performance processor through high-speed data interfaces such as Ethernet, MIPI or USB. The high-performance processor runs the local SLAM mapping and traversal exploration algorithms of each drone node, and controls the flight controller, and then sends instructions to the actuator to control the movement of the drone. The high-performance processor is also connected to the memory and the multi-point communication device, among which the memory is used to store the raw data information of various sensors carried by the drone, and the multi-point communication device is used to establish data communication between each drone and the central command station.

[0053] The central command station runs a multi-node real-time online collaborative SLAM algorithm and an offline high-precision SLAM modeling algorithm. The SLAM algorithm running on each drone node is local and is responsible for processing the data on the drone. The SLAM algorithm running on the central command station is global and will integrate the data of multiple drones to coordinate positioning and mapping.

[0054] In the autonomous exploration process of the present invention, the entire exploration process is divided into two stages: local rough modeling and offline fine modeling. The local rough modeling stage can tolerate the loss of data communication between drones and will not cause the failure of the entire exploration and modeling process. It can be divided into three stages:

[0055] (1) The initial stage is used to detect the initial posture of each UAV and initialize the initial value of each SLAM node;

[0056] (2) Autonomous exploration and coarse modeling stage, each UAV node conducts autonomous exploration and builds a real-time coarse model on the central command station. This step is robust to communication loss between nodes.

[0057] (3) In the offline precise modeling stage, each UAV node transmits or copies the data to the central command station after returning to build a precise model offline.

[0058] Example 2

[0059] This embodiment discloses an exploration method for a multi-machine cooperative confined space autonomous exploration UAV system based on the embodiment 1. The method is divided into three stages: initial stage, autonomous exploration stage, and offline precise modeling. The specific workflow of each stage is as follows:

[0060] 1. Initial stage: Figure 2As shown in the figure, the initial state information of each drone node is measured in this stage, so that each SLAM algorithm in the whole system is correctly initialized. In this stage, multiple drones are usually gathered near the central node to prepare for takeoff. At this time, wireless communication is more reliable, and one or more laser cooperation targets can be deployed in the environment to enhance the observability of the surrounding environment of each drone in the initial stage. The specific workflow of this stage is as follows:

[0061] S1.1. Keep each drone node in a static state for a few seconds to obtain the zero bias and gravity observation of the drone's onboard IMU sensor data. The synchronized drone node will accumulate a few seconds of visual and lidar observation data, and integrate to obtain the initial observation of the surrounding environment point cloud data. The subscript 0 represents the initial moment, and the superscript i represents the drone node number. The drones are arranged in the order in which they are powered on. The first drone powered on is recorded as 0, and the subsequent drones powered on are numbered from 1 to n. The body coordinate system of the drone numbered 0 when it is powered on is the global coordinate system.

[0062] S1.2, each drone node downsamples its point cloud observation of the surrounding environment Transmit to the central command station. At this time, each drone node has not yet set off and is still near the central node, so the transmission is feasible.

[0063] S1.3. The central command station collects the point cloud observation data sampled by each drone. Point cloud observation data sampled by the first drone Match them together to get the relative position of each drone before departure and the transformation matrix In the formula Refers to the position vector of the i-th UAV in global coordinates at the initial moment, Refers to the attitude angle quaternion of the i-th UAV in the global coordinate system at the initial moment. Refers to the transfer matrix from the body coordinates of the i-th UAV to the global coordinate system at the initial moment.

[0064] S1.4. The central command station sends the relative position of each drone before departure back to each drone node to initialize the initial position of each SLAM node state variable.

[0065] Second, the independent exploration stage: Figure 3 As shown in the figure, at this stage, each UAV will fly autonomously under the guidance of the local SLAM algorithm, and perform real-time online rough modeling at the central command station. The following algorithms will run in parallel on each UAV.

[0066] S2.0, each UAV and the central command station rasterize the space to be explored into three-dimensional voxel grids, each grid can store an exploration state (reached or not reached), and all grids are initialized to not reached.

[0067] S2.1, each drone node runs its own SLAM algorithm and updates its status through IMU sensor data. Status transmission to Where k refers to the current moment and k-1 refers to the previous moment.

[0068] S2.2. Each UAV node matches the point cloud observation data to its own local map M locali Observe and update to optimize its own posture information and the transformation matrix And use this posture information to control the posture of the UAV; Refers to the transfer matrix from the i-th UAV body coordinates to the global coordinate system at the current moment k.

[0069] S2.3. Each drone node maintains and splices its own local map:

[0070] S2.4. Each drone node sends its own position information Transformation Matrix And the downsampled observation point cloud data Transmit to the central command station;

[0071] S2.5, the central command station receives the transformation matrix in S2.2 The new point cloud data observed by each drone Transform to the global coordinate system and stitch them into a rough real-time global map:

[0072] S2.6. The central command station will provide a rough real-time global map Map to the three-dimensional voxel grid map, and mark the mapped grids as arrived, while the rest of the grids remain in their original state;

[0073] S2.7, the central command station broadcasts the updated information of the grid that has arrived to each drone node;

[0074] S2.8, each drone node searches for nearby grids according to the distance until it finds the nearest unreached grid, then guides the drone to fly in that direction, and repeats the process from step S2.1. If all surrounding grids have been reached, the drone is controlled to return.

[0075] When searching for the nearest unreached grid, first mark all grids as unreached as described in step S2.0, and then update them iteratively step by step. The grids scanned by the point cloud are updated to the arrived state, as described in step S2.6, and the remaining grids are unreached.

[0076] In the autonomous exploration stage, each drone has flown into the confined space, and the data communication between the drone and the central command station is in an unreliable state. Therefore, the workflow of this stage must be robust to communication interruptions, so that the entire system has feasibility in engineering practice. Since the SLAM algorithms of each drone node in the exploration stage of the present invention are all locally operated and do not rely on mutual communication between machines as observation, the worst impact is only the explored situation that the drone nodes cannot synchronize the grids with each other in time. There is a possibility that a certain grid is repeatedly explored by different drones, but it will not affect the SLAM algorithm work of a certain node, and will not cause the node SLAM algorithm to diverge and the drone to lose control. That is, for the present invention, the most undesirable situation is that the communication of steps S2.4 and S2.7 fails, and the grid is repeatedly explored. At present, other collaborative SLAM algorithms all have a strong dependence on real-time communication. When each drone node performs point cloud observation and update as in S2.2, it needs to rely on point cloud information transmitted in real time from other nodes. Once a communication failure occurs, the SLAM algorithm cannot work, resulting in divergence of drone positioning and control.

[0077] 3. Offline precise modeling stage: Figure 4 As shown, the following algorithms are all run on the central command station.

[0078] S3.1. The central command station controls each UAV node to return to the initial position, and transmits or copies the original sensor data of each UAV to the central command station through a data link or a removable storage medium;

[0079] S3.2, according to the initial posture observed in S1.3, assign initial values ​​to each sub-SLAM algorithm;

[0080] S3.3. Splice the initial observation point clouds of each node as the initial accurate global map:

[0081] S3.4, in turn according to the sampling time sequence of each node, perform the update step for each sub-SLAM algorithm according to the IMU data of each sub-SLAM node, Status transmission to

[0082] S3.5, in turn, according to the sampling time sequence of each node, perform the observation step for each sub-SLAM algorithm, match the observed point cloud to the refined global map, and optimize the pose accuracy in the state variable. The difference from S2.2 is that the point cloud observation will be performed on the entire refined global map M globalP Therefore, it is possible to better match the point cloud data perceived by multiple drones;

[0083] S3.6. Attach a detailed global map at each step:

[0084] S3.7, perform loop detection and graph optimization to further improve pose and modeling accuracy;

[0085] S3.8, loop to S3.3 and restart the process until all sensor data are processed;

[0086] S3.9, output the final global map M globalP , as a refined global model.

[0087] In summary, the confined space autonomous exploration drone system of the present invention can, under the control of the central command station, have multiple drones autonomously locate and map in the confined space through the onboard visual sensors and lidar sensors, autonomously plan exploration trajectories, traverse the confined space, and establish a rough real-time map on the central command station. After the drones return, the raw sensor data stored by multiple drones are merged and imported into the central command station, and an accurate map model is calculated and established, which can realize autonomous navigation and execution of tasks in complex, narrow or dangerous environments, and is used for emergency search and rescue, military operations command, making up for the shortcomings of the existing technology.

[0088] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the present invention in any form, and any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A multi-machine collaborative autonomous exploration drone system for confined spaces, characterized in that: include: Several UAV nodes and a central command station communicate with each other, and the central command station runs a multi-node real-time online collaborative SLAM algorithm and an offline high-precision SLAM modeling algorithm; the UAV node includes: a flight controller, an actuator, several visual sensors and lidar sensors, a high-performance processor, a memory and a multi-point communication device, the visual sensor is used to collect image information of the surrounding environment, the lidar sensor is used to collect point cloud data of the surrounding environment, the high-performance processor runs the local SLAM mapping and traversal exploration algorithm of each UAV node, and is connected to the visual sensor, lidar sensor, memory and multi-point communication device, and the high-performance processor controls the flight controller, and then sends instructions to the actuator.

2. The multi-machine collaborative confined space autonomous exploration UAV system according to claim 1 is characterized in that: The visual sensor and the lidar sensor are connected to the high-performance processor via Ethernet, MIPI or USB high-speed data interface.

3. The multi-machine collaborative confined space autonomous exploration UAV system according to claim 1 is characterized in that: The memory is used to store the original data information of various sensors carried by the drone, and the multi-point communication device is used to establish data communication between each drone and the central command station.

4. The exploration method of a multi-machine collaborative confined space autonomous exploration UAV system according to claim 1 is characterized in that: It consists of three stages: (1) The initial stage is used to detect the initial posture of each UAV and initialize the initial value of each SLAM node; (2) Autonomous exploration and coarse modeling stage, each UAV node conducts autonomous exploration and builds a real-time coarse model on the central command station. This step is robust to communication loss between nodes. (3) Offline precise modeling stage: after returning, each drone node transmits or copies the data to the central command station. Build accurate models offline.

5. The exploration method of a multi-machine collaborative confined space autonomous exploration UAV system according to claim 4 is characterized in that: In the initial stage, the drones gather near the central node to wait for takeoff, and one or more laser cooperation targets are deployed in the environment.

6. The exploration method of a multi-machine collaborative confined space autonomous exploration UAV system according to claim 5 is characterized in that: The specific workflow of the initial stage is as follows: S1.

1. Keep each drone node stationary for a few seconds to obtain the zero bias and gravity observation of the IMU sensor data on the drone. The synchronized drone node will accumulate the visual and lidar observation data for a few seconds, and integrate to obtain the initial observation P0 of the surrounding environment point cloud data. i , where the subscript 0 represents the initial moment, and the superscript i represents the UAV node number. The UAVs are arranged in the order in which they are powered on, and the body coordinate system of the UAV numbered 0 when it is powered on is the global coordinate system; S1.2, each drone node downsamples its point cloud observation of the surrounding environment Transmit to the central command station; S1.

3. The central command station collects the point cloud observation data sampled by each drone. Point cloud observation data sampled by the first drone Match each other to get the relative position of each drone before departure and the transformation matrix in Refers to the position vector of the i-th UAV in global coordinates at the initial moment, Refers to the attitude angle quaternion of the i-th UAV in the global coordinate system at the initial moment, Refers to the transfer matrix from the i-th UAV body coordinates to the global coordinate system at the initial moment; S1.

4. The central command station sends the relative position of each drone before departure back to each drone node to initialize the initial position of each SLAM node state variable.

7. The exploration method of a multi-machine collaborative confined space autonomous exploration UAV system according to claim 4 is characterized in that: The specific workflow of the autonomous exploration phase is as follows: S2.0, each UAV and the central command station rasterize the space to be explored into three-dimensional voxel grids, and initialize all grids as unreached; S2.1, each drone node runs its own SLAM algorithm and updates its status through IMU sensor data. Status transmission to Where k refers to the current moment, and k-1 refers to the previous moment; S2.

2. Each UAV node matches the point cloud observation data to its own local map M locali Observe and update to optimize its own posture information and the transformation matrix And use this posture information to control the posture of the drone. Refers to the transfer matrix from the i-th UAV body coordinates to the global coordinate system at the current moment k; S2.

3. Each drone node maintains and splices its own local map: S2.

4. Each drone node sends its own position information Transformation Matrix And the downsampled observation point cloud data Transmit to the central command station; S2.5, the central command station receives the transformation matrix in S2.2 The new point cloud data observed by each drone Transform to the global coordinate system and stitch them into a rough real-time global map: S2.

6. The central command station will provide a rough real-time global map Map to the three-dimensional voxel grid map, and mark the mapped grids as arrived, while the rest of the grids remain in their original state; S2.7, the central command station broadcasts the updated information of the grid that has arrived to each drone node; S2.

8. Each drone node searches for grids near itself according to the distance until it finds the nearest unreached grid, then guides the drone to fly in that direction and repeats the work from step S2.1; if all surrounding grids have been reached, the drone is controlled to return.

8. The exploration method of a multi-machine collaborative confined space autonomous exploration UAV system according to claim 7, characterized in that: Each three-dimensional voxel grid can store an exploration state, which is divided into: arrived and not arrived.

9. The exploration method of a multi-machine collaborative confined space autonomous exploration UAV system according to claim 6, characterized in that: The specific workflow of the offline precise modeling stage is as follows: S3.

1. The central command station controls each UAV node to return to the initial position, and the original sensor data of each UAV is transmitted to the central command station; S3.2, according to the initial posture observed in S1.3, assign initial values ​​to each sub-SLAM algorithm; S3.

3. Splice the initial observation point clouds of each node as the initial accurate global map: S3.4, in turn according to the sampling time sequence of each node, perform the update step for each sub-SLAM algorithm according to the IMU data of each sub-SLAM node, Status transmission to S3.5, perform the observation step for each sub-SLAM algorithm in turn according to the sampling time sequence of each node, match the observed point cloud to the refined global map, and optimize the pose accuracy in the state variable; S3.

6. Attach a detailed global map at each step: S3.7, perform loop detection and graph optimization to further improve pose and modeling accuracy; S3.8, loop to S3.3 and restart the process until all sensor data are processed; S3.9, output the final global map M globalP , as a refined global model.

10. The exploration method of a multi-machine collaborative confined space autonomous exploration UAV system according to claim 9, characterized in that: The original sensor data of each UAV is transmitted or copied to the central command station via data links or removable storage media.

Citation Information

Patent Citations

  • Quasi-real-time processing method and device on aerial remote sensing data machine and storage medium

    CN110209847A

  • Visual sensor laser radar fused unmanned aerial vehicle positioning and mapping device and method

    CN112347840A

  • Multi-machine cooperative laser SLAM method, device and system of mobile robot

    CN114383611A

  • Mining intrinsic safety type explosion-proof laser radar slam measuring method and device

    CN116381718A

  • Multi-robot collaborative mapping method for post-disaster rescue scene

    CN116503576A

Cited By

  • Underground pipe gallery detection method based on unmanned aerial vehicle point cloud scanning

    CN119986687A