A complex terrain-oriented unmanned aerial vehicle group formation dynamic obstacle avoidance control method

Abstract

The application discloses a complex-terrain-oriented dynamic obstacle avoidance control method for UAV formation, which aims to solve the problems of the existing technology, such as the conflict between formation keeping and dynamic obstacle avoidance, rough terrain modeling and weak dynamic obstacle coping ability. The method comprises the following steps: a terrain perception and modeling module fuses the onboard LiDAR / IMU / GPS data to generate a local octree map; a formation control module adaptively adjusts the formation according to the terrain complexity; a dynamic obstacle avoidance planning module adopts an improved artificial potential field method or model predictive control, fuses the terrain threat degree and dynamic obstacle information, and plans a local obstacle avoidance trajectory; a distributed cooperative decision module is based on a distributed model predictive control (DMPC) framework, and eliminates trajectory conflicts through neighbor communication and consistency iteration; and a formation reorganization module quickly recovers the expected formation after obstacle avoidance. The application realizes the unification of the formation keeping and dynamic obstacle avoidance of the UAV formation in complex terrain, and significantly improves the flight safety and environmental adaptability.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) control technology, specifically a dynamic obstacle avoidance control method for UAV swarm formations in complex terrain. Background Technology

[0002] In recent years, with the rapid development of unmanned aerial vehicle (UAV) technology, UAV swarms have shown great application potential in fields such as military reconnaissance, geological exploration, disaster relief, and agricultural plant protection. Especially in complex terrain environments such as mountains, canyons, urban canyons, and jungles, a single UAV often struggles to complete its mission due to limited visibility and insufficient payload capacity. However, UAV swarm formations can significantly improve mission efficiency and reliability through collaborative perception and division of labor. Therefore, multi-UAV swarm control for complex terrain has become one of the current research hotspots in UAV technology.

[0003] Currently, there are numerous research findings on UAV formation control. For example, the leader-follower method maintains formation by having the following UAVs track the pose of the leader UAV; the virtual structure method treats all UAVs as part of a rigid body, tracking a virtual reference point; and the behavior-based method decomposes the formation task into a weighted combination of multiple basic behaviors. In obstacle avoidance, common methods include artificial potential field methods, vector field histogram (VFH) methods, rapidly expanding random trees (RRT), and model predictive control (MPC). Some studies have also attempted to combine formation control with obstacle avoidance, for example, by simultaneously incorporating formation constraints and obstacle avoidance constraints into the MPC framework.

[0004] However, existing technologies still have significant shortcomings when applied to complex terrain environments. First, most methods treat terrain as simple geometric obstacles (such as spheres or cylinders), failing to handle continuously undulating real-world terrain surfaces, easily leading to drones crashing into mountains or getting stuck in valleys. Second, existing methods are weak in handling dynamic obstacles (such as moving vehicles, birds, or other drones), often requiring global replanning and exhibiting poor real-time performance. More importantly, in existing methods, formation maintenance and dynamic obstacle avoidance are often two relatively independent objectives, which are prone to conflict in complex terrain, leading to formation disarray, inter-drone collisions, or obstacle avoidance failures. Achieving a balance between formation stability and safe obstacle avoidance in highly complex environments with multiple dynamic obstacles is a pressing technical challenge that needs to be addressed. Summary of the Invention

[0005] This application provides a dynamic obstacle avoidance control method for UAV swarm formation in complex terrain. By deeply integrating multiple aspects such as terrain perception, formation adaptive adjustment, distributed collaborative obstacle avoidance planning, and rapid formation reorganization, it can effectively solve the problems of formation and obstacle avoidance conflict, insufficient dynamic obstacle response, and rough terrain modeling in the background technology under complex terrain.

[0006] To achieve the above objectives, this application provides the following technical solution: a dynamic obstacle avoidance control method for UAV swarm formations in complex terrain, comprising: a terrain perception and modeling module for constructing a local 3D terrain mesh map; a formation control module for generating the desired formation and desired position; a dynamic obstacle avoidance planning module for fusing terrain and obstacle information and planning local obstacle avoidance trajectories using an improved artificial potential field method or MPC; and a distributed collaborative decision-making module for resolving trajectory conflicts based on a consensus algorithm. This solution, through a perception-planning-coordination-control closed loop, enables the UAV swarm to autonomously avoid static terrain and dynamic obstacles while maintaining formation integrity, effectively improving flight safety and mission completion in complex environments.

[0007] Furthermore, the terrain perception and modeling module generates local octree maps by fusing GPS / IMU / LiDAR data and adaptively adjusts the update frequency, thereby improving the real-time performance and accuracy of terrain modeling during high-speed flight.

[0008] Furthermore, the formation control module includes a formation adaptive adjustment unit that dynamically shrinks the formation or switches the column formation when the terrain complexity exceeds the limit, thereby enhancing the formation's ability to pass through narrow terrain.

[0009] Furthermore, the dynamic obstacle avoidance planning module introduces an adaptive repulsive force gain based on terrain threat level and a dynamic collision avoidance potential field based on relative velocity into the improved artificial potential field method, which solves the local minima problem and the dynamic obstacle avoidance problem of the traditional potential field method.

[0010] Furthermore, the distributed collaborative decision-making module adopts the distributed model predictive control (DMPC) framework, which only requires a small amount of information from neighboring UAVs to solve for a globally consistent trajectory through local optimization, thereby reducing the requirements for computing power and communication bandwidth.

[0011] Furthermore, after obstacle avoidance, the formation reorganization and recovery module uses an optimization assignment algorithm to generate a non-intersecting reorganization path from the disordered state to the desired formation, thus achieving rapid self-recovery of the formation.

[0012] Furthermore, the onboard computing unit adopts a high-performance embedded AI platform (such as Jetson AGX Orin), and the communication uses Wi-Fi Mesh that supports the MAVLink protocol, ensuring the real-time operation of complex algorithms and reliable inter-machine communication.

[0013] Furthermore, dynamic obstacle detection combines visual sensors with deep learning models such as YOLOv8 and performs Kalman filtering state prediction on the detected target, enabling the drone to avoid high-speed moving obstacles in advance.

[0014] Furthermore, terrain complexity is quantified by the height variance and average slope of the local grid map, serving as a trigger condition for adaptive formation adjustment, thus tightly coupling formation behavior with terrain features.

[0015] Furthermore, the adaptive repulsive force gain function of the improved artificial potential field method dynamically changes with terrain complexity and the reciprocal of distance, realizing the risk avoidance logic that the more dangerous the terrain and the closer the distance, the stronger the repulsive force, thus balancing smoothness and safety.

[0016] Compared with the prior art, the beneficial effects of this application are: 1. Strong environmental adaptability: By introducing terrain complexity assessment and octree maps, it can effectively handle continuous undulating real terrain and adaptively adjust the formation accordingly, which significantly improves the survivability of UAV swarms in complex environments such as mountains and jungles.

[0017] 2. Good obstacle avoidance and formation coordination: This application deeply integrates dynamic obstacle avoidance planning with distributed collaborative decision-making, and adopts the improved potential field method and DMPC framework. While avoiding static terrain and dynamic obstacles, the consensus algorithm ensures that no collisions occur within the formation, thus avoiding the contradiction of choosing between obstacle avoidance and formation preservation in traditional methods.

[0018] 3. High real-time performance and robustness: Based on a distributed architecture, each drone only needs to perceive locally and communicate with its neighbors, resulting in a low computational burden. Furthermore, the rapid reorganization module can tolerate some formation disturbances caused by obstacle avoidance, making the system robust overall and suitable for practical deployment. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall process of the method described in this application.

[0020] Figure 2 This is an example of a local octree map generated by the terrain perception and modeling module in this application.

[0021] Figure 3 This is a schematic diagram of multi-machine trajectory conflict resolution in the distributed collaborative decision-making module of this application.

[0022] The diagram shows: 1. Terrain perception and modeling module; 2. Formation control module; 3. Dynamic obstacle avoidance planning module; 4. Distributed collaborative decision-making module; 5. Formation reorganization and recovery module. Detailed Implementation

[0023] The following will refer to the accompanying drawings (within) of the embodiments of this application. Figures 1-3Taking the embodiments of this application as an example, the technical solutions in the embodiments are clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0024] In the description of this application, any descriptions of orientation, such as up, down, front, back, left, right, etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are merely for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. When a feature is referred to as being set, fixed, or connected to another feature, it can be directly set, fixed, or connected to the other feature, or it can be indirectly set, fixed, or connected to the other feature.

[0025] Please see Figures 1 to 3 This application provides the following technical solution: a dynamic obstacle avoidance control method for UAV swarm formation in complex terrain.

[0026] Specifically, the method includes the following steps and modules: (I) Terrain Perception and Modeling Module (1): Each UAV is equipped with an airborne LiDAR (such as Livox Mid-360), an inertial measurement unit (IMU, such as ADIADIS16470), and GPS (such as U-blox NEO-M9N). Multi-source data is fused in real time using a tightly coupled SLAM algorithm (such as LIO-SAM) to generate a local 3D octomap. The map resolution is set to 0.2m, and the update frequency is dynamically adjusted according to the UAV's current ground speed: when the speed > 10m / s, the update frequency is 20Hz; when the speed < 5m / s, the update frequency is 5Hz. Simultaneously, for each voxel in the map, its height variance is calculated as a terrain complexity index, and areas with a slope > 30 are marked as non-flyable areas.

[0027] (ii) Formation Control Module (2): Adopts a virtual navigator-follower structure. The virtual navigator flies according to the globally planned path, and each following UAV calculates its desired position relative to the virtual navigator according to the desired formation (e.g., five UAVs in a V-shape). This module has a built-in formation adaptive adjustment unit: when the terrain complexity index (height variance) within 10m ahead is >0.8m as learned from the terrain perception module, it is judged as narrow and complex terrain. The unit automatically reduces the lateral spacing of the formation from the default 1.5 times the wingspan to 0.8 times the wingspan and switches the formation to a single-file column mode to improve passability. When the terrain complexity returns to <0.3m for 2 seconds, the original formation is gradually restored.

[0028] (III) Dynamic obstacle avoidance planning module (3): Local obstacle avoidance planning is carried out using the improved artificial potential field method.

[0029] Gravitational field: The direction from the current drone position to the desired position given by the formation module.

[0030] Repulsive field: including static terrain repulsion, dynamic obstacle repulsion, and machine-to-machine repulsion.

[0031] The module ultimately outputs a smooth local obstacle avoidance trajectory.

[0032] (iv) Distributed Collaborative Decision Module (4): To avoid conflicts between obstacle avoidance trajectories of multiple drones, this module adopts a distributed model predictive control (DMPC) framework. Each drone only exchanges its predicted trajectory (10 time steps long) with its neighbors within a communication range (within 200 meters) via Wi-Fi Mesh (MAVLink protocol). Each drone solves a local optimization problem: minimizing its own deviation from the desired position, the deviation from the predicted trajectory of its neighbors, and the control input, while satisfying obstacle avoidance constraints and the minimum safe distance (1.5 meters) constraint. This problem is solved online using sequential quadratic programming (SQP) at a frequency of 50Hz. After 2-3 rounds of communication / optimization iterations, the cluster achieves a consistent trajectory, i.e., global conflict-free operation.

[0033] (V) Formation Reorganization and Recovery Module (5): After all UAVs have passed through the obstacle zone (based on feedback from the virtual navigator), this module obtains the current actual position and desired formation position of each UAV, and uses the Hungarian algorithm to solve an assignment problem that minimizes the total movement distance, matching each UAV to a desired position without overlap. Subsequently, a B-spline path is generated for each UAV that avoids the current position of other UAVs, guiding it to complete formation reorganization within 3 seconds.

[0034] When using: First, before takeoff, the ground station pre-stores the mission area map (if any) into each UAV. After takeoff, each UAV runs the terrain perception and modeling module (1) in real time to continuously build a local octree map. The formation control module (2) guides the cluster flight according to the preset V-shaped formation and global path. When the complexity of the terrain ahead is perceived to increase, the formation adaptive adjustment unit automatically shrinks and switches to column mode. At the same time, the dynamic obstacle avoidance planning module (3) calculates the local obstacle avoidance trajectory every 20ms. If a static terrain obstacle (such as a ridge) or a dynamic obstacle (such as a bird) is detected, an avoidance repulsion force is immediately generated. In order to avoid new collisions caused by inconsistent avoidance directions among multiple UAVs, the distributed collaborative decision-making module (4) reaches a consensus within 40ms through inter-UAV communication, so that each UAV flies along the negotiated conflict-free trajectory. Once all UAVs have passed through the obstacle area, the formation reorganization module (5) quickly restores the scattered UAVs to the desired formation and continues to perform subsequent tasks. The whole process is fully automated and requires no ground intervention.

[0035] It is worth noting that the hardware selection disclosed in this embodiment is only one specific example. The core chip of the airborne computing unit can be the NVIDIA Jetson AGX Orin 64GB module, whose computing power is sufficient to run the aforementioned SLAM, YOLOv8, and MPC optimization algorithms in real time. The communication module can use the ESP32-S3 to build a Wi-Fi Mesh network with a communication frequency of 2.4GHz and a transmit power of 20dBm. Under actual testing conditions, the packet loss rate within 200 meters is less than 1%, meeting the requirements of collaborative control. The lidar can also be OusterOS0-64 or Hesai XT32M, and the vision sensor can be Intel RealSense D455 or STEREOLABSZED2. The control algorithm uses the Robot Operating System (ROS2) as the software framework and is implemented using C++ programming. The MPC optimization relies on the open-source library OSQP for efficient solution. The start, stop, and parameter configuration of each module can be managed through the control switch group on the airborne ground station or remote controller. The specific logic programming adopts the state machine method commonly used in the prior art.

[0036] Although embodiments of this application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic obstacle avoidance control method for UAV swarm formation in complex terrain, characterized in that, include: Terrain perception and modeling module (1) is used to collect terrain data by airborne radar or depth camera and construct a local three-dimensional terrain grid map; The formation control module (2) is used to generate the desired formation of the UAV group and the desired position of each UAV based on the virtual structure method or the pilot-follower method. The dynamic obstacle avoidance planning module (3) is used to integrate the local three-dimensional terrain grid map and dynamic obstacle detection information, and to plan a local obstacle avoidance trajectory for each UAV using an improved artificial potential field method or model predictive control (MPC) algorithm. The distributed collaborative decision-making module (4) is used to share the local obstacle avoidance trajectory and status information planned by each UAV through the communication link between the UAVs, and to resolve the trajectory conflict collaboratively based on the consensus algorithm to generate the conflict-free final control command. The terrain perception and modeling module (1), formation control module (2), dynamic obstacle avoidance planning module (3) and distributed collaborative decision-making module (4) are connected electrically or through communication in sequence and run in the UAV onboard computing unit or ground station control system.

2. The method for dynamic obstacle avoidance control of UAV swarm formation in complex terrain according to claim 1, characterized in that: The terrain perception and modeling module (1) is specifically used to fuse data from the global positioning system, inertial measurement unit and airborne lidar in real time, generate a local octree map with terrain slope and roughness information, and adaptively adjust the map update frequency according to the flight speed of the UAV, so as to ensure the real-time performance and accuracy of terrain modeling under high-speed flight.

3. The method for dynamic obstacle avoidance control of UAV swarm formation in complex terrain according to claim 1, characterized in that: The formation control module (2) includes a formation adaptive adjustment unit. When the complexity of the terrain ahead is detected to exceed a preset threshold, the formation adaptive adjustment unit is used to dynamically shrink the formation spacing or switch to special formations such as column / dispersion to cope with complex terrain, thereby improving the formation's passability in narrow valleys or jungle environments.

4. The method for dynamic obstacle avoidance control of UAV swarm formation in complex terrain according to claim 1, characterized in that: The improved artificial potential field method adopted by the dynamic obstacle avoidance planning module (3) adds an adaptive repulsive gain based on terrain threat level and a dynamic obstacle avoidance potential field based on relative velocity to the traditional gravitational field and repulsive field. This effectively solves the problem that the traditional potential field method is prone to getting into local minima and cannot avoid moving obstacles in complex terrain.

5. The method for dynamic obstacle avoidance control of UAV swarm formation in complex terrain according to claim 1, characterized in that: The distributed collaborative decision-making module (4) adopts a framework based on distributed model predictive control. Each UAV only needs to exchange predicted trajectories and control quantities with neighboring UAVs within the communication range. By iteratively solving local optimization problems with consistency constraints, it achieves global conflict-free formation obstacle avoidance, reduces computational complexity, and improves the system's communication anti-interference capability.

6. The method for dynamic obstacle avoidance control of UAV swarm formation in complex terrain according to claim 1, characterized in that: It also includes a formation reorganization and recovery module (5). After the UAV group passes through the obstacle area as a whole, the formation reorganization and recovery module (5) is used to generate a non-intersecting, short-distance reorganization path from the current disordered state to the desired formation using an optimization assignment algorithm based on the actual position and speed deviation of each UAV. This achieves efficient self-recovery of the formation after obstacle avoidance.

7. The method for dynamic obstacle avoidance control of UAV swarm formation in complex terrain according to claim 1, characterized in that: The core processor of the onboard computing unit is an NVIDIA Jetson AGX Orin or an embedded AI computing platform with equivalent computing power. The communication link adopts a Wi-Fi Mesh or ZigBee wireless module that supports the MAVLink protocol, ensuring the reliability of the computing power requirements for algorithm operation and low-latency communication between the machines.

8. The method for dynamic obstacle avoidance control of UAV swarm formation in complex terrain according to claim 1, characterized in that: The dynamic obstacle detection information is obtained by combining airborne vision sensors with real-time target detection algorithms such as YOLOv8, and the detected dynamic obstacles are predicted by Kalman filtering to generate evasive maneuvers in advance.

9. A dynamic obstacle avoidance control method for UAV swarm formation in complex terrain according to claim 1, characterized in that: The terrain complexity is defined by calculating the height variance and average slope of the local grid map. When the height variance is greater than 0.5m or the average slope is greater than 30, it is considered as complex terrain, triggering the formation adaptive adjustment function of the formation control module (2).

Images