Road-air cooperative unmanned vehicle emergency correction method and system

Abstract

The embodiment of the invention discloses a road-air cooperative unmanned vehicle emergency correction method and system, and the method comprises the steps: collecting environment data through a laser radar carried by an unmanned aerial vehicle, and achieving the time-space calibration of a sensor through the combination of RTK positioning and PTP time synchronization; an SLAM technology is adopted to fuse IMU pre-integration, a laser odometer and RTK constraint to construct a three-dimensional point cloud map; and filtering and downsampling the point cloud, converting the point cloud into a two-dimensional drivable area map through orthogonal projection, and transmitting the two-dimensional drivable area map to the unmanned vehicle. The unmanned vehicle calculates the distances between the left and right vertexes of the unmanned vehicle and the boundary of the drivable area in real time, and when any distance is smaller than a threshold value, a reverse steering torque is requested from the electric power steering system to achieve deviation correction. The method gets rid of dependence on illumination and lane lines, multiplexes the laser radar to reduce the cost, is suitable for complex scenes such as ports and disaster areas, and effectively reduces the collision risk.

Description

Technical Field

[0001] This invention relates to the field of unmanned vehicle technology, specifically to an emergency deviation correction method and system for unmanned vehicles using road-air cooperation. Background Technology

[0002] With the development of autonomous driving technology, the active safety functions of driverless vehicles have become a research focus. Existing emergency lane departure correction schemes generally rely on visual sensors: a front-facing camera identifies lane lines, an MCU calculates the lateral distance between the vehicle and the lane lines, and when it detects that the vehicle is about to deviate from the lane, it controls the electric power steering (EPS) system to implement reverse torque correction. This type of scheme is effective to some extent in public roads with clear lane lines and sufficient lighting, but it has significant limitations:

[0003] Poor scene adaptability: It is only suitable for structured roads with clear lane markings, and it is completely ineffective in unstructured scenarios such as factory areas, ports, and disaster areas where there are no lane markings.

[0004] Low environmental robustness: The camera is susceptible to lighting conditions (such as at night, in tunnels), weather interference (rain, snow, fog) and dirt, which may cause it to become unusable.

[0005] Redundancy in hardware costs: Although advanced autonomous driving systems are equipped with LiDAR, the correction function still requires additional cameras, resulting in a waste of sensor resources and increased costs.

[0006] While research has been conducted in recent years, the inherent limitations of in-vehicle perception have not yet been resolved: it is only applicable to scenarios with clear lane markings and relatively simple operating conditions. It lacks the concept of freespace and is not suitable for all roads without lane markings. Furthermore, cameras are overly sensitive to lighting conditions, rendering the function ineffective in low-light environments. Meanwhile, for complex scenarios, advanced autonomous driving solutions often employ LiDAR (Light Detection and Ranging) solutions. Adding cameras for a single function, on top of existing LiDAR, would increase costs. Summary of the Invention

[0007] To address this issue, this invention provides an emergency lane correction method and system for unmanned vehicles using a road-air cooperative approach, thereby solving the technical problem that existing technologies rely on front-facing cameras to identify lane lines, which are only applicable to roads with clear lane lines and simple operating conditions, and cannot be applied to complex or special scenarios.

[0008] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0009] According to a first aspect of the present invention, an emergency deviation correction method for unmanned vehicles using road-air cooperation is provided, the method comprising:

[0010] S1. Spatiotemporal calibration of UAV sensors:

[0011] The time reference for the UAV domain controller is provided by the PPS pulses of RTK and GPRMC data, and the network latency is eliminated by using the PTP protocol.

[0012] Calculate the external parameters between the UAV's lidar and inertial navigation system based on the hand-eye calibration principle to achieve spatial synchronization;

[0013] S2, UAV SLAM mapping:

[0014] The drone is equipped with a lidar to collect environmental point cloud data, and integrates IMU pre-integration constraints, lidar odometry constraints, closed-loop constraints and RTK position constraints;

[0015] The pose graph was optimized using the GTSAM optimizer to generate a 3D point cloud map in the global coordinate system and saved as a PCD file.

[0016] S3, 3D point cloud to 2D drivable area map:

[0017] The point cloud data of the PCD file is filtered and downsampled, and then converted into a two-dimensional plane through orthogonal projection.

[0018] Generate a grayscale image in PGM format, verify it, and then convert it into a black and white BMP format map of the drivable area.

[0019] S4. Autonomous vehicle boundary crossing correction control:

[0020] Transform the map coordinates of the drivable area to the vehicle coordinate system and calculate the distances between the left and right vertices of the vehicle and the boundary of the drivable area;

[0021] When the distance to the left vertex is less than the threshold, request right-turn torque from EPS; when the distance to the right vertex is less than the threshold, request left-turn torque from EPS.

[0022] Furthermore, a time reference is provided for the UAV domain controller using RTK PPS pulses and GPRMC data, and the PTP protocol is used to eliminate network latency, including:

[0023] Parse UTC timestamps using RTK's GPRMC data;

[0024] Align the rising edge of the PPS pulse with the second pulse of the GPRMC, and distribute the time base to the lidar and inertial navigation sensors via the PTP protocol.

[0025] Furthermore, based on the hand-eye calibration principle, the external parameters between the UAV's lidar and inertial navigation system are calculated to achieve spatial synchronization, including:

[0026] While the UAV is in motion, inertial navigation recursive trajectory and lidar recursive trajectory are collected respectively;

[0027] The rotation matrix and translation vector of the lidar relative to the inertial navigation coordinate system are calculated using the hand-eye calibration principle.

[0028] Furthermore, the pose graph is optimized using the GTSAM optimizer to generate a 3D point cloud map in the global coordinate system and save it as a PCD file, including:

[0029] Accurate pose information is obtained by utilizing the pre-integration constraints of the IMU, the constraints of the lidar odometry, the closed-loop constraints, and the position constraints provided by RTK:

[0030] The pre-integration constraint of the IMU is to calculate the relative pose change by integrating the angular velocity and acceleration;

[0031] The lidar odometry constraint is to obtain the pose transformation matrix based on continuous frame point cloud matching.

[0032] The closed-loop constraint is to identify loop closures by matching scene feature points and then add pose constraints.

[0033] After obtaining accurate pose information, the point cloud data acquired by the LiDAR is converted to the global coordinate system and saved as a PCD file.

[0034] Furthermore, the PCD file point cloud data is filtered and downsampled, and then converted into a two-dimensional plane through orthogonal projection, including:

[0035] Set the resolution of the 2D grid and project the 3D point cloud onto the XY plane along the Z-axis.

[0036] Take the maximum value of the point cloud height within each grid to generate an elevation grayscale image.

[0037] Furthermore, the coordinates of the drivable area map are transformed to the vehicle coordinate system, including:

[0038] Based on the RTK positioning data and heading angle of the unmanned vehicle, a transformation matrix from the vehicle coordinate system to the geodetic coordinate system is constructed;

[0039] The coordinates of the drivable area map are transformed to the vehicle coordinate system through inverse matrix operations.

[0040] Further, the distances between the left and right vertices of the vehicle and the boundaries of the drivable area are calculated, including:

[0041] Based on the vehicle size parameters, with the center of the rear axle of the vehicle as the origin, calculate the coordinates of the left and right vertices in the vehicle body coordinate system;

[0042] The Euclidean distance formula is used to calculate the distance from the vertex to the nearest point on the boundary of the drivable area.

[0043] Furthermore, the EPS torque request in step S4 is a preset fixed value or a variable dynamically adjusted based on vehicle speed.

[0044] Furthermore, between steps S2 and S3, the drone also transmits the PCD format point cloud map to the unmanned vehicle in real time via a 5G or dedicated wireless communication link.

[0045] According to a second aspect of the present invention, a road-air cooperative emergency deviation correction system for unmanned vehicles is provided, the system comprising:

[0046] The UAV sensor spatiotemporal calibration module is used to perform the following steps:

[0047] The time reference for the UAV domain controller is provided by the PPS pulses of RTK and GPRMC data, and the network latency is eliminated by using the PTP protocol.

[0048] Calculate the external parameters between the UAV's lidar and inertial navigation system based on the hand-eye calibration principle to achieve spatial synchronization;

[0049] The UAV SLAM mapping module is used to perform the following steps:

[0050] The drone is equipped with a lidar to collect environmental point cloud data, and integrates IMU pre-integration constraints, lidar odometry constraints, closed-loop constraints and RTK position constraints;

[0051] The pose graph was optimized using the GTSAM optimizer to generate a 3D point cloud map in the global coordinate system and saved as a PCD file.

[0052] The 3D point cloud to 2D drivable area map module is used to perform the following steps:

[0053] The point cloud data of the PCD file is filtered and downsampled, and then converted into a two-dimensional plane through orthogonal projection.

[0054] Generate a grayscale image in PGM format, verify it, and then convert it into a black and white BMP format map of the drivable area.

[0055] The autonomous vehicle boundary crossing correction control module is used to perform the following steps:

[0056] Transform the map coordinates of the drivable area to the vehicle coordinate system and calculate the distances between the left and right vertices of the vehicle and the boundary of the drivable area;

[0057] When the distance to the left vertex is less than the threshold, request right-turn torque from EPS; when the distance to the right vertex is less than the threshold, request left-turn torque from EPS.

[0058] The embodiments of the present invention have the following advantages:

[0059] This invention utilizes a drone equipped with a LiDAR to collect environmental data, combining RTK positioning and PTP time synchronization for sensor spatiotemporal calibration. SLAM technology is employed to fuse IMU pre-integration, LiDAR odometry, and RTK constraints to construct a 3D point cloud map. The point cloud is then filtered and downsampled before being converted into a 2D drivable area map via orthogonal projection and transmitted to the drone. The drone calculates the distances between its left and right vertices and the drivable area boundaries in real time. When any distance falls below a threshold, it requests reverse steering torque from the electric power steering system to correct its course. This invention eliminates reliance on lighting and lane markings, reuses LiDAR to reduce costs, and is suitable for complex scenarios such as ports and disaster areas, effectively reducing collision risks. Attached Figure Description

[0060] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0061] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0062] Figure 1 A schematic diagram of the logical structure of an emergency deviation correction system for unmanned vehicles with road-air cooperation provided in an embodiment of the present invention;

[0063] Figure 2 A flowchart illustrating an emergency deviation correction method for unmanned vehicles using road-air cooperation, provided as an embodiment of the present invention;

[0064] Figure 3 The 3D point cloud map is provided in an emergency correction method for unmanned vehicles with road-air cooperation in an embodiment of the present invention.

[0065] Figure 4 The 2DPGM diagram is provided in an embodiment of the present invention for an emergency correction method for unmanned vehicles using road-air cooperation.

[0066] Figure 5 The image shown is a black-and-white BMP image used in an emergency correction method for unmanned vehicles with road-air cooperation provided in an embodiment of the present invention. Detailed Implementation

[0067] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] ERKS (Emergency Road Keep System): Emergency Road Keeping System.

[0069] SLAM (Simultaneous Localization and Mapping): Simultaneous localization and mapping allows autonomous vehicles, drones, robots, and other devices to build maps in unknown environments while simultaneously determining their own location.

[0070] PTP (Precision Time Protocol) is a protocol used to achieve precise time synchronization in computer networks. Based on the IEEE 1588 standard, it provides sub-microsecond synchronization accuracy and is suitable for scenarios with stringent time synchronization requirements.

[0071] PPS (Pulse Per Second): The number of pulses per second is a unit for measuring the frequency of a signal or pulse, and is commonly used in fields such as communication, measurement, and testing.

[0072] GPRMC (Global Positioning System Recommended Minimum data) is a standard format defined in the NMEA 0183 protocol for transmitting basic GPS position, velocity, and time information.

[0073] Freespace: Driveable area.

[0074] EPS (Electric Power Steering): Electric power steering system.

[0075] In current solutions for emergency lane correction functions in autonomous vehicles, the primary sensor used is a front-facing camera. This camera identifies the lane lines in the vehicle's lane and provides lane line attribute information. Simultaneously, the MCU (Microcontroller Unit) calculates the distance between the vehicle and the lane lines after acquiring this information. When the autonomous vehicle fails to follow the lane direction and the distance between the vehicle and the lane lines falls below a certain threshold, the MCU controls the steering wheel to rotate in the opposite direction. This function ensures that if the autonomous vehicle deviates from its lane due to external factors, it can be pulled back into its lane by controlling the steering wheel, thus reducing the risk of collision.

[0076] Existing solutions are only suitable for scenarios with clear lane markings and relatively simple operating conditions. They lack the concept of Freespace and are not applicable to all roads without lane markings. Furthermore, the cameras are overly sensitive to lighting conditions, rendering the function ineffective in low light. Meanwhile, for complex scenarios, advanced autonomous driving solutions often employ LiDAR (Light Detection and Ranging) systems. Adding cameras for a single function, on top of existing LiDAR, would increase costs.

[0077] To address the aforementioned technical issue that relying on front-facing cameras to identify lane lines only works on roads with clear lane lines and simple conditions, and cannot be applied to complex or special scenarios.

[0078] This invention proposes a solution for emergency deviation correction within a drivable area through collaborative operation of an electric multi-rotor drone and an unmanned vehicle (UAV), applicable to complex or special scenarios such as factory areas, disaster zones, and ports. The solution involves the electric multi-rotor drone pre-collecting point cloud data of the environment ahead of the UAV using LiDAR and transmitting it to the UAV. The UAV then processes the point cloud map and converts it into a 2D map of the drivable area. While the UAV is moving towards its target point, if factors such as LiDAR malfunction or control inaccuracies cause the vehicle to deviate from the drivable area, the UAV can automatically correct its course to avoid a collision.

[0079] refer to Figure 1 This invention discloses an emergency deviation correction system for unmanned vehicles that combines road and air traffic. The system includes: a UAV sensor spatiotemporal calibration module 1; a UAV SLAM mapping module 2; a 3D point cloud to 2D drivable area map module 3; and an unmanned vehicle boundary crossing correction control module 4.

[0080] Corresponding to the aforementioned road-air cooperative emergency deviation correction system for unmanned vehicles, this invention also discloses a road-air cooperative emergency deviation correction method for unmanned vehicles. The following details the road-air cooperative emergency deviation correction method for unmanned vehicles disclosed in this invention, in conjunction with the road-air cooperative emergency deviation correction system described above.

[0081] refer to Figures 2 to 5This invention discloses an emergency deviation correction method for unmanned vehicles using road-air cooperation, the method comprising:

[0082] S1. Spatiotemporal calibration of UAV sensors:

[0083] The time reference for the UAV domain controller is provided by the PPS pulses of RTK and GPRMC data, and the network latency is eliminated by using the PTP protocol.

[0084] Calculate the external parameters between the UAV's lidar and inertial navigation system based on the hand-eye calibration principle to achieve spatial synchronization;

[0085] S2, UAV SLAM mapping:

[0086] The drone is equipped with a lidar to collect environmental point cloud data, and integrates IMU pre-integration constraints, lidar odometry constraints, closed-loop constraints and RTK position constraints;

[0087] The pose graph was optimized using the GTSAM optimizer to generate a 3D point cloud map in the global coordinate system and saved as a PCD file.

[0088] S3, 3D point cloud to 2D drivable area map:

[0089] The point cloud data of the PCD file is filtered and downsampled, and then converted into a two-dimensional plane through orthogonal projection.

[0090] Generate a grayscale image in PGM format, verify it, and then convert it into a black and white BMP format map of the drivable area.

[0091] S4. Autonomous vehicle boundary crossing correction control:

[0092] Transform the map coordinates of the drivable area to the vehicle coordinate system and calculate the distances between the left and right vertices of the vehicle and the boundary of the drivable area;

[0093] When the distance to the left vertex is less than the threshold, request right-turn torque from EPS; when the distance to the right vertex is less than the threshold, request left-turn torque from EPS.

[0094] Furthermore, a time reference is provided for the UAV domain controller using RTK PPS pulses and GPRMC data, and the PTP protocol is used to eliminate network latency, including:

[0095] Parse UTC timestamps using RTK's GPRMC data;

[0096] Align the rising edge of the PPS pulse with the second pulse of the GPRMC, and distribute the time base to the lidar and inertial navigation sensors via the PTP protocol.

[0097] Furthermore, based on the hand-eye calibration principle, the external parameters between the UAV's lidar and inertial navigation system are calculated to achieve spatial synchronization, including:

[0098] While the UAV is in motion, inertial navigation recursive trajectory and lidar recursive trajectory are collected respectively;

[0099] The rotation matrix and translation vector of the lidar relative to the inertial navigation coordinate system are calculated using the hand-eye calibration principle.

[0100] Furthermore, the pose graph is optimized using the GTSAM optimizer to generate a 3D point cloud map in the global coordinate system and save it as a PCD file, including:

[0101] Accurate pose information is obtained by utilizing the pre-integration constraints of the IMU, the constraints of the lidar odometry, the closed-loop constraints, and the position constraints provided by RTK:

[0102] The pre-integration constraint of the IMU is to calculate the relative pose change by integrating the angular velocity and acceleration;

[0103] The lidar odometry constraint is to obtain the pose transformation matrix based on continuous frame point cloud matching.

[0104] The closed-loop constraint is to identify loop closures by matching scene feature points and then add pose constraints.

[0105] After obtaining accurate pose information, the point cloud data acquired by the LiDAR is converted to the global coordinate system and saved as a PCD file.

[0106] Furthermore, the PCD file point cloud data is filtered and downsampled, and then converted into a two-dimensional plane through orthogonal projection, including:

[0107] Set the resolution of the 2D grid and project the 3D point cloud onto the XY plane along the Z-axis.

[0108] Take the maximum value of the point cloud height within each grid to generate an elevation grayscale image.

[0109] Furthermore, the coordinates of the drivable area map are transformed to the vehicle coordinate system, including:

[0110] Based on the RTK positioning data and heading angle of the unmanned vehicle, a transformation matrix from the vehicle coordinate system to the geodetic coordinate system is constructed;

[0111] The coordinates of the drivable area map are transformed to the vehicle coordinate system through inverse matrix operations.

[0112] Further, the distances between the left and right vertices of the vehicle and the boundaries of the drivable area are calculated, including:

[0113] Based on the vehicle size parameters, with the center of the rear axle of the vehicle as the origin, calculate the coordinates of the left and right vertices in the vehicle body coordinate system;

[0114] The Euclidean distance formula is used to calculate the distance from the vertex to the nearest point on the boundary of the drivable area.

[0115] Furthermore, the EPS torque request in step S4 is a preset fixed value or a variable dynamically adjusted based on vehicle speed.

[0116] Furthermore, between steps S2 and S3, the drone also transmits the PCD format point cloud map to the unmanned vehicle in real time via a 5G or dedicated wireless communication link.

[0117] This method is designed for use by autonomous vehicles in complex or special scenarios (such as factory interiors, disaster areas, and ports). It utilizes the lidar and inertial navigation systems on the drone to locate and map unknown areas, obtaining a 3D point cloud map. This 3D point cloud map is then processed into a 2D binary image: a drivable area map. When the autonomous vehicle is traveling along the desired trajectory within the drivable area, if lidar malfunctions or control errors cause the vehicle to nearly exceed the drivable area, the vehicle can be promptly corrected back into the drivable area by rotating the steering wheel.

[0118] Step S100, Spatiotemporal Calibration of UAV Sensors: This aims to ensure the accuracy and consistency of environmental perception data and positioning information. To achieve this goal, time and space synchronization between the LiDAR, inertial navigation system (INS), and RTK is required. For time synchronization, this invention uses the RTK's PPS and GPRMC to provide a precise time reference for the UAV's domain controller. Simultaneously, PTP effectively eliminates time delays caused by network transmission, thereby ensuring that the timestamps of all sensors remain consistent. For spatial synchronization, the external parameters between the LiDAR and INS are calibrated. When the UAV is in motion, the recursive trajectories of the INS and LiDAR are calculated separately, and the external parameters between the LiDAR and INS are determined using the hand-eye calibration principle.

[0119] Step S200, SLAM constructs a point cloud map: The UAV is equipped with a LiDAR to capture surrounding environmental data in real time, and then uses the SLAM algorithm to estimate its own position and construct a point cloud map of the 3D environment. To obtain high-precision point cloud data in the global coordinate system, this invention employs a GTSAM optimizer to perform detailed optimization of the pose map. This optimization process comprehensively considers various constraints, including IMU pre-integration constraints, LiDAR odometry constraints, closed-loop constraints, and position constraints provided by RTK. Once accurate pose information is obtained, the point cloud data acquired by the LiDAR can be accurately converted to the global coordinate system and saved in PCD file format for subsequent analysis and utilization.

[0120] Step S300, 3D Point Cloud to 2D Driving Area Map: To obtain a 2D driving area map, this invention needs to convert the PCD format point cloud data into a grayscale image format, that is, convert the 3D point cloud data into a 2D grayscale image. First, the PCD file is read to obtain the point cloud data; then, the point cloud data is preprocessed such as filtering and downsampling. This invention uses orthogonal projection to project the 3D point cloud onto a 2D plane; after writing the file header information (such as image width, height, grayscale level, etc.) according to the PGM format file structure, the generated grayscale image data is written to the file. Finally, OpenCV is used to read the PGM file to determine the accuracy of the conversion result, proving that the conversion is complete.

[0121] Step S400, Determine Boundary Marker Position & Control EPS Direction: After obtaining the drivable area map, coordinate transformation can be performed to convert the coordinates of the drivable area map from the geodetic coordinate system to the vehicle coordinate system. The coordinates of the left and right vertices of the vehicle can be calculated using the vehicle's dimensions. Then, combining this with the relative position of the vehicle to the drivable area, it can be calculated whether the vehicle has exceeded the drivable area. When the distance between the left vertices and the boundary of the drivable area meets the distance threshold, a right turn angle or torque is requested from the EPS. When the distance between the right vertices and the boundary of the drivable area meets the distance threshold, a left turn angle or torque is requested from the EPS.

[0122] This invention performs a 2D conversion on the 3D point cloud generated by SLAM technology and generates a usable BMP file as a drivable area map. By transforming the coordinates of the drivable area to the vehicle coordinate system to obtain the relative positional relationship between the autonomous vehicle and the drivable area, it is possible to request the opposite angle or torque from the EPS when the autonomous vehicle is about to exceed the drivable area, thereby achieving correction. This has the following advantages:

[0123] 1) The autonomous vehicle corrects its course by adjusting its position relative to the drivable area. It does not rely on factors such as lighting or lane markings and can operate at any time and place.

[0124] 2) It does not rely on visual factors, reducing the number of sensors and lowering hardware costs.

[0125] One approach is to use a camera-based sensor solution to correct deviations within the drivable area. The drivable area of ​​the road is visually extracted, and the camera's built-in computing chip calculates the relative position and angle between the drivable area and the vehicle. The MCU (Microcontroller Unit) can then use this positional relationship to perform emergency corrections when the vehicle is about to leave the drivable area. However, this camera-based solution is highly dependent on lighting conditions, and will be limited in adverse weather conditions such as rain, snow, fog, or insufficient lighting at night.

[0126] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for emergency deviation correction of unmanned vehicles using road-air cooperative systems, characterized in that, The method includes: S1. Spatiotemporal calibration of UAV sensors: The time reference for the UAV domain controller is provided by the PPS pulses of RTK and GPRMC data, and the network latency is eliminated by using the PTP protocol. Calculate the external parameters between the UAV's lidar and inertial navigation system based on the hand-eye calibration principle to achieve spatial synchronization; S2, UAV SLAM mapping: The drone is equipped with a lidar to collect environmental point cloud data, and integrates IMU pre-integration constraints, lidar odometry constraints, closed-loop constraints and RTK position constraints; The pose graph was optimized using the GTSAM optimizer to generate a 3D point cloud map in the global coordinate system and saved as a PCD file. S3, 3D point cloud to 2D drivable area map: The point cloud data of the PCD file is filtered and downsampled, and then converted into a two-dimensional plane through orthogonal projection. Generate a grayscale image in PGM format, verify it, and then convert it into a black and white BMP format map of the drivable area. S4. Autonomous vehicle boundary crossing correction control: Transform the map coordinates of the drivable area to the vehicle coordinate system and calculate the distances between the left and right vertices of the vehicle and the boundary of the drivable area; When the distance to the left vertex is less than the threshold, request right-turn torque from EPS; when the distance to the right vertex is less than the threshold, request left-turn torque from EPS.

2. The emergency deviation correction method for unmanned vehicles using road-air cooperation as described in claim 1, characterized in that, The UAV domain controller is provided with a time reference through RTK PPS pulses and GPRMC data, and the PTP protocol is used to eliminate network latency, including: Parse UTC timestamps using RTK's GPRMC data; Align the rising edge of the PPS pulse with the second pulse of the GPRMC, and distribute the time base to the lidar and inertial navigation sensors via the PTP protocol.

3. The emergency deviation correction method for unmanned vehicles using road-air cooperation as described in claim 1, characterized in that, Based on the hand-eye calibration principle, the external parameters between the UAV's lidar and inertial navigation system are calculated to achieve spatial synchronization, including: While the UAV is in motion, inertial navigation recursive trajectory and lidar recursive trajectory are collected respectively; The rotation matrix and translation vector of the lidar relative to the inertial navigation coordinate system are calculated using the hand-eye calibration principle.

4. The emergency deviation correction method for unmanned vehicles using road-air cooperation as described in claim 1, characterized in that, The pose graph is optimized using the GTSAM optimizer, generating a 3D point cloud map in the global coordinate system and saving it as a PCD file, including: Accurate pose information is obtained by utilizing the pre-integration constraints of the IMU, the constraints of the lidar odometry, the closed-loop constraints, and the position constraints provided by RTK: The pre-integration constraint of the IMU is to calculate the relative pose change by integrating the angular velocity and acceleration; The lidar odometry constraint is to obtain the pose transformation matrix based on continuous frame point cloud matching. The closed-loop constraint is to identify loop closures by matching scene feature points and then add pose constraints. After obtaining accurate pose information, the point cloud data acquired by the LiDAR is converted to the global coordinate system and saved as a PCD file.

5. The emergency deviation correction method for unmanned vehicles using road-air cooperation as described in claim 1, characterized in that, The PCD file point cloud data is filtered and downsampled, then converted into a two-dimensional plane through orthogonal projection, including: Set the resolution of the 2D grid and project the 3D point cloud onto the XY plane along the Z-axis. Take the maximum value of the point cloud height within each grid to generate an elevation grayscale image.

6. The emergency deviation correction method for unmanned vehicles using road-air cooperation as described in claim 1, characterized in that, Transform the map coordinates of the drivable area to the vehicle coordinate system, including: Based on the RTK positioning data and heading angle of the unmanned vehicle, a transformation matrix from the vehicle coordinate system to the geodetic coordinate system is constructed; The coordinates of the drivable area map are transformed to the vehicle coordinate system through inverse matrix operations.

7. The emergency deviation correction method for unmanned vehicles using road-air cooperation as described in claim 6, characterized in that, Calculate the distances from the vehicle's left and right vertices to the boundaries of the drivable area, including: Based on the vehicle size parameters, with the center of the rear axle of the vehicle as the origin, calculate the coordinates of the left and right vertices in the vehicle body coordinate system; The Euclidean distance formula is used to calculate the distance from the vertex to the nearest point on the boundary of the drivable area.

8. The emergency deviation correction method for unmanned vehicles using road-air cooperation as described in claim 1, characterized in that, The EPS torque request in step S4 is either a preset fixed value or a variable that is dynamically adjusted based on vehicle speed.

9. The emergency deviation correction method for unmanned vehicles using road-air cooperation as described in claim 1, characterized in that, Between steps S2 and S3, the drone also transmits the PCD format point cloud map to the unmanned vehicle in real time via a 5G or dedicated wireless communication link.

10. An emergency deviation correction system for unmanned vehicles with road-air cooperation, characterized in that, The system includes: The UAV sensor spatiotemporal calibration module is used to perform the following steps: The time reference for the UAV domain controller is provided by the PPS pulses of RTK and GPRMC data, and the network latency is eliminated by using the PTP protocol. Calculate the external parameters between the UAV's lidar and inertial navigation system based on the hand-eye calibration principle to achieve spatial synchronization; The UAV SLAM mapping module is used to perform the following steps: The drone is equipped with a lidar to collect environmental point cloud data, and integrates IMU pre-integration constraints, lidar odometry constraints, closed-loop constraints and RTK position constraints; The pose graph was optimized using the GTSAM optimizer to generate a 3D point cloud map in the global coordinate system and saved as a PCD file. The 3D point cloud to 2D drivable area map module is used to perform the following steps: The point cloud data of the PCD file is filtered and downsampled, and then converted into a two-dimensional plane through orthogonal projection. Generate a grayscale image in PGM format, verify it, and then convert it into a black and white BMP format map of the drivable area. The autonomous vehicle boundary crossing correction control module is used to perform the following steps: Transform the map coordinates of the drivable area to the vehicle coordinate system and calculate the distances between the left and right vertices of the vehicle and the boundary of the drivable area; When the distance to the left vertex is less than the threshold, request right-turn torque from EPS; when the distance to the right vertex is less than the threshold, request left-turn torque from EPS.

Images