Unmanned aerial vehicle conflict detection and release method based on improved ORCA

Abstract

The invention discloses an unmanned aerial vehicle conflict detection and release method based on improved ORCA. The method comprises the following steps: step 1, acquiring operation data of an unmanned aerial vehicle; acquiring position and speed information of all unmanned aerial vehicles which are being simulated at present from the Airsim simulation platform; step 2, unmanned aerial vehicle conflict detection; based on a preset safety interval standard, performing conflict grading on the flight situation: (2a) modeling an unmanned aerial vehicle conflict detection model; (2b) conflict grading identification; step 3, conflict resolution of the unmanned aerial vehicle is carried out; (3a) executing corresponding classification release logic according to different conflict levels output in the conflict detection stage; (3b) returning and updating an instruction; according to the method, through an affine transformation technology, track deviation in the conflict resolution process is reduced radically and geometrically. And the calculation efficiency of the algorithm is obviously improved. Finally, a simulation experiment proves that the improved algorithm effectively solves the problem of detouring a far way while ensuring safe avoidance.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) flight and air traffic safety technology, specifically relating to a UAV conflict detection and resolution algorithm based on an improved ORCA algorithm. Background Technology

[0002] Unmanned aerial vehicle (UAV) air traffic management is a core technology for ensuring low-altitude safety and efficiency. Its key lies in enabling UAVs to autonomously perceive their surroundings, predict potential conflicts, and generate safe escape strategies, thereby replacing complete reliance on manual monitoring and significantly improving operational efficiency and safety. Conflict detection and escape is a fundamental task of autonomous UAV flight. Its goal is to identify aircraft with conflict risks from real-time situational information and calculate feasible maneuvers to avoid collisions. Thanks to the continuous development of optimization algorithms and distributed control theory over the past decade, computing systems can fully leverage their advantages in real-time processing and multi-constraint solving, allowing escape algorithms to learn and evolve from a large number of conflict scenarios and dynamic models without relying entirely on pre-set fixed rules. By introducing and optimizing algorithmic frameworks based on velocity barriers or optimization search, the system can handle more complex and high-density dynamic conflict scenarios, thus meeting the high safety and real-time requirements of actual airspace operations. With this advantage, conflict detection and escape technology plays a crucial role in urban logistics, dense formation flying, air traffic control, and emergency rescue.

[0003] Conflict resolution algorithms can be broadly categorized into deterministic algorithms based on geometric rules and intelligent algorithms based on optimization search. Geometric rule-based algorithms not only make relatively idealistic assumptions about the maneuver model but are also prone to oscillations or no solutions in multi-aircraft conflict scenarios, making them difficult to apply directly to dynamic and complex real-world airspace environments. With the rapid development of optimization algorithms and computing power, conflict resolution algorithms based on optimization search have demonstrated greater flexibility and optimality. From the perspective of solution mechanisms, optimization search algorithms can be divided into centralized optimization algorithms and distributed optimization algorithms.

[0004] Centralized optimization algorithms typically begin by minimizing the global cost function. After the state information of all aircraft is collected, a central solver generates multiple alternative trajectories or speed adjustment schemes for each aircraft within a unified spatiotemporal scope. These schemes constitute a vast solution space. The central solver needs to perform conflict detection and cost evaluation for each scheme, placing a huge computational burden on the entire system during the planning and replanning phases, and significantly consuming communication and computing resources. Furthermore, because conflict resolution requires extremely high real-time performance, and the time consumed by global optimization increases dramatically with the number of aircraft, it is difficult to apply online in larger-scale scenarios. Distributed optimization algorithms can better alleviate these problems. Each aircraft independently calculates its optimal or safer escape maneuvers based only on local neighbor information, eliminating the need for a central node to process all information. This makes the system architecture more lightweight and scalable, distributing the computational burden and making it more adaptable to large-scale, dynamic UAV operating environments.

[0005] Velocity obstacle-based optimization algorithms are an important branch of distributed escape algorithms. Compared to simpler reactive rule-based methods, they typically generate smoother and more energy-efficient escape trajectories. Reactive rule-based methods (such as the artificial potential field method) involve the UAV directly calculating and adjusting a repulsive force direction in real time based on its current relative position and velocity. Velocity obstacle-based optimization algorithms (such as the ORCA algorithm) first construct a prohibited region called a "velocity obstacle cone" in velocity space based on the UAV's own state and that of neighboring aircraft. Then, outside this region, a new velocity vector is selected that conforms to its own dynamic constraints and is closest to the original desired velocity. Overall, this type of optimization algorithm represents a systematic improvement over simple reactive behavior, providing better decisions from the perspective of avoiding future conflicts.

[0006] The ORCA algorithm is a typical example of a speed barrier optimization algorithm. In traditional speed barrier methods, the set of non-collision speeds constructed for each pair of aircraft is fixed. In multi-aircraft interactions, this can lead to unsolvable problems due to overly tight constraints, increasing the risk of negotiation failure. The ORCA algorithm employs linear optimization, based on a speed barrier cone, and uses a responsibility allocation (such as equal responsibility) to linearly approximate the boundary of the prohibited area, constructing a half-plane-shaped "non-collision speed allowable domain" for each aircraft. By solving a linear programming problem (i.e., finding the point closest to the desired speed within the intersection of multiple half-planes), it can efficiently and in real-time generate conflict-free speed vectors. The linear optimization framework based on responsibility allocation links the geometric constraints of collision avoidance with efficient speed search, significantly improving the robustness and computational efficiency of multi-aircraft cooperative decoupling. In practical applications, it achieves real-time computational performance similar to reactive algorithms while ensuring high safety. Summary of the Invention

[0007] This invention addresses the problem of efficient and safe collision detection and resolution in dense, dynamic UAV environments by providing a UAV collision detection and resolution method based on the enhanced ORCA algorithm. The implementation steps of this method are as follows:

[0008] Step 1: Data collection during drone operation;

[0009] Obtain the position and velocity information of all drones currently being simulated from the Airsim simulation platform;

[0010] Step 2: Unmanned Aerial Vehicle (UAV) Conflict Detection

[0011] Based on preset safety interval standards, conflict classification is performed on flight situations:

[0012] (2a) Modeling of UAV conflict detection model;

[0013] (2b) Conflict grading identification; determining the urgency of the conflict:

[0014] Step 3: Drone Conflict Resolution

[0015] (3a) Based on the different conflict levels output during the conflict detection phase, execute the corresponding classification and resolution logic:

[0016] (3b) Command feedback and update;

[0017] Experiments show that applying this method to UAV conflict detection and resolution algorithms can improve the accuracy of conflict detection and reduce trajectory deviation in conflict resolution, demonstrating the superiority of this method.

[0018] Advantages of this method:

[0019] By employing affine transformation techniques, the heterogeneous ellipsoidal conflict problem is equivalently transformed into a standard sphere solution. Using an ellipsoid to replace the spherical model ensures accurate alignment with physical characteristics, fundamentally reducing track deviation during conflict resolution. Secondly, to overcome the drawback of invalid calculations due to single static boundaries in far-field environments, a hierarchical conflict detection model based on 'double-domain nesting' is constructed, establishing a pre-conflict domain and an emergency conflict domain. Furthermore, for the resolution velocity calculation process, boundary projection techniques are introduced to replace the traditional iterative approximation process. By simplifying the complex problem into a linear projection onto the velocity barrier boundary, the computational efficiency of the algorithm is significantly improved. Finally, simulation experiments demonstrate that the improved algorithm effectively solves the problem of "taking a longer route" while ensuring safe avoidance. Attached Figure Description

[0020] Figure 1 A diagram illustrating the status of the drone.

[0021] Figure 2ellipsoid model diagram.

[0022] Figure 3 Diagram of a dual-domain hierarchical conflict detection model.

[0023] Figure 4 Conflict detection flowchart.

[0024] Figure 5 Two-dimensional planar diagram of pre-conflict resolution.

[0025] Figure 6 Two-dimensional plan view of emergency conflict resolution.

[0026] Figure 7 Simulation results of high Z-axis deviation ratio: (a) Original ORCA algorithm (b) Improved ORCA algorithm.

[0027] Figure 8 Simulation results for low Z-axis deviation ratio: (a) Original ORCA algorithm (b) Improved ORCA algorithm. Detailed Implementation

[0028] The present invention will be further described with reference to the accompanying drawings.

[0029] Step 1: Data collection during drone operation;

[0030] This algorithm uses the Airsim simulation platform, which provides APIs to obtain the drone's position and speed information in real time. For simplicity, we'll use two drones as an example here. Figure 1 As shown, two drones and The parameters, whose state parameters are the velocity vector, are... , With position vector , .

[0031] Step two, drone conflict detection;

[0032] (2a) Modeling of UAV conflict detection model;

[0033] Choose "ellipsoid" as the geometric bounding box model for the drone (e.g., Figure 2As shown in the figure, this is mainly because the physical structure of a quadrotor UAV has significant geometric differences along the horizontal and vertical axes. An ellipsoidal model, by adjusting the semi-axis length parameter, can accurately fit the rotor plane and fuselage height, achieving a compact representation of the space occupied by the UAV. Compared to the spherical modeling in the traditional ORCA algorithm, the ellipsoidal model more accurately reflects the actual asymmetric fuselage structure of the UAV, improving space utilization. In handling conflict detection and resolution between multiple UAVs, heterogeneous space mapping modeling based on affine transformation directly solves for tangents or minimum distances in the original space, involving complex second-order constraints. Therefore, the idea of ​​affine transformation is introduced to transform the heterogeneous expansion problem into an isotropic problem in a regularized space. To simplify computation, an affine transformation matrix is ​​defined. and its inverse matrix Scale normalization is performed on each axis:

[0034]

[0035] Heterogeneous conflict modeling between multiple drones, specifically for two heterogeneous drones. and Their ellipsoid radii are respectively and This paper calculates the maximum ratio of the lengths of each axis. :

[0036]

[0037] This yields the equivalent ellipsoid radius after superposition. Through the matrix The mapping simplifies the originally heterogeneous collision detection problem into a regularized space with radius r. The problem of determining the equivalent sphere.

[0038] (2b) Conflict grading identification; determining the urgency of the conflict:

[0039] like Figure 2 As shown, with the UAV's own center of mass as the origin, two nested risk boundaries are defined in the three-dimensional airspace:

[0040] 1. Outer pre-collision domain: A blue spherical region centered on the UAV. This domain corresponds to a relatively long time window. Its physical meaning is that, considering the current relative motion state, in the future... Within a given timeframe, there is a "warning buffer zone" that indicates a potential collision risk. Its main function is to detect potential threats at a distance in advance, triggering early, smooth track coordination or speed fine-tuning, with the aim of resolving conflicts with minimal energy cost.

[0041] 2. Inner Emergency Conflict Domain: A red ellipsoidal region centered on the drone (the semi-axis length can be set in different directions, forward / backward and left / right, depending on the drone's maneuverability). This domain corresponds to a shorter, more urgent time window. It defines the "last line of defense" that must be taken immediately with strong evasive maneuvers, focusing on high-urgency, close-range collision risks, and ensuring that highly certain and large-amplitude escape maneuvers can be triggered in critical moments.

[0042] These two risk domains work together to form a detection mechanism of "early warning and tiered response." Real-time kinematic analysis is required to achieve accurate judgment across these two domains. (As above) Figure 3 As shown, consider two drones. and Its state parameters are the velocity vector. , With position vector , Calculate relative velocity = relative position The known time window for the pre-collision domain is... The time window for emergency conflict domains is ,when This indicates that the two aircraft are in a close approach posture, a necessary prerequisite for a potential conflict. Based on this, the logic for determining a graded conflict is as follows:

[0043] Pre-collision determination: When the two aircraft are in a state of mutual approach ( Furthermore, at the current relative speed, the time required to reach each other's positions is less than the warning time window. When this condition is met, it is considered a pre-conflict. This condition can be quantified as follows: and .

[0044] No-conflict determination: If the first determination is not a pre-conflict, then there is no conflict;

[0045] Emergency Conflict Detection: When two aircraft not only approach each other, but their relative positions also pass through a time window. After mapping, the modulus of the constructed transformation matrix 𝑻 (which contains information about the shape and relative motion of the ellipsoidal domain) is less than a set safety threshold (e.g., 1+). ,in (This refers to the safety margin coefficient), which determines whether an emergency conflict is imminent. This condition can be quantified as: .

[0046] The decision process flowchart is as follows: Figure 4 As shown.

[0047] Step 3: Unmanned Aerial Vehicle (UAV) Conflict Resolution;

[0048] (3a) Based on the different conflict levels output during the conflict detection phase, execute the corresponding classification and resolution logic:

[0049] Conflict-free: When there is no conflict, the drone only needs to execute the predetermined instructions.

[0050] Pre-conflict resolution, such as Figure 5 As shown, drones A and B are simplified to point mass models, and their relative position vectors are... The relative velocity vector is Assuming the UAV exhibits asymmetric maneuverability in the horizontal direction, it can be equivalently represented as an ellipsoid. For efficient computation, a predefined mapping transformation matrix is ​​first used... Map the ellipse to a radius of The standard circle, whose inverse transformation is denoted as .

[0051] The specific calculation process is as follows:

[0052] (1);

[0053] (2)

[0054] set up: Then we have:

[0055] (3)

[0056] (4)

[0057] (5)

[0058] We can obtain the following by substituting formulas (1) and (2):

[0059] (6)

[0060] According to the quadratic formula, we can obtain The solution is:

[0061] (7)

[0062] when Pick Time represents separation The furthest one Therefore, it is effective due to the presence of rays. Take only Therefore, according to It can be obtained unit vector

[0063] Therefore, we can obtain... :

[0064] (8)

[0065] If Through the inverse reflection transformation matrix Transforming back to the original physical space yields the pre-collision domain. Half-plane boundary points :

[0066] (9)

[0067] Normal vector of a half-plane for:

[0068] (10)

[0069] Emergency conflict resolution, such as Figure 6 As shown, the specific calculation process is as follows:

[0070] (11)

[0071] (12)

[0072] (13)

[0073] Unit vector Therefore, the vector can be calculated.

[0074] (14)

[0075] If Through the inverse reflection transformation matrix Transforming back to the original physical space yields the pre-collision domain. Half-plane boundary points :

[0076] (15)

[0077] Normal vector of a half-plane for:

[0078] (16)

[0079] Finally, the final speed of liberation was determined using linear programming.

[0080] (3b) Command feedback and update

[0081] The speed command after release is sent to the simulation platform for execution, and the loop returns to step one to perform closed-loop detection.

[0082] Experiments and Analysis

[0083] 1. Experimental conditions

[0084] The hardware testing platform for this invention consists of an AMD Ryzen 78845H CPU with a clock speed of 3.80 GHz, 32 GB of memory, and an Nvidia GeForce GTX 4070 Laptop graphics card. The software platform is a Windows 11 operating system, the development environment is PyCharm, the programming language is Python, and the simulation platform is AirSim.

[0085] 2. Experimental Data

[0086] The proposed simulation parameters are shown in Table 1 below:

[0087] Table 1

[0088]

[0089] 3. Results Analysis

[0090] Experimental results comparison Figure 7 , Figure 8 As shown. By comparing the performance of the traditional sphere obstacle avoidance method and the improved ellipsoidal obstacle avoidance algorithm of this invention under different deviation ratios, the following conclusions can be drawn:

[0091] The conservative nature of traditional spherical obstacle avoidance methods: Figure 7 and Figure 8 The upper path and the corresponding data curve ( Figure 7 a, Figure 8 a) It can be seen that, because the traditional method simplifies the UAV into an isotropic sphere, its obstacle avoidance constraints do not change with the relative orientation of the UAV. Experimental data shows that regardless of the deviation ratio between the two UAVs on the z-axis, the shortest obstacle avoidance distance of the sphere model remains relatively stable at around 4 meters (e.g., 4.02 meters and 3.69 meters). This "one-size-fits-all" modeling approach leads to overly conservative obstacle avoidance constraints in complex environments, resulting in a serious waste of airspace resources.

[0092] Improved Adaptive Performance of the Ellipsoid Decomposition Method: The Improved Algorithm Proposed in this Patent ( Figure 7 , Figure 8 The path below (using ellipsoidal modeling and affine transformation) achieves accurate perception of the "wide and thin" physical characteristics of the UAV, and its decoupling maneuverability exhibits significant adaptive features:

[0093] High z-axis deviation ratio scenarios: such as Figure 7 As shown, when the deviation between the two drones accounts for a high proportion along the z-axis, the improved algorithm can take advantage of the drone's smaller envelope in the vertical direction. In this case, the shortest escape distance is significantly reduced and approaches 2 meters. This demonstrates that the improved algorithm, while ensuring absolute safety, significantly improves the utilization rate of the three-dimensional airspace by compressing ineffective safety redundancy.

[0094] Scenarios with low z-axis deviation ratio: such as Figure 8 As shown, when the z-axis deviation ratio is small, the collision mainly occurs on the horizontal detection surface. At this time, the shortest release distance of the improved algorithm rises back to close to 4 meters (the actual measurement is 3.62 meters), which is consistent with the spherical model and ensures a safe distance during horizontal maneuvers.

[0095] Meanwhile, a series of experiments were designed to address the conflict resolution between the two drones. The experimental results are shown in Table 2. The first three columns of the table represent the target trajectories of the two drones. , , The distance on the axis, the last two columns are the shortest distance between the two drones when the original ORCA algorithm and the improved ORCA algorithm are running.

[0096] Table 2

[0097]

[0098] As shown in Table 2, the experiment recorded in detail the two drones at different relative positions ( shaft and Axis coordinate changes, Comparison of the shortest distances for conflict resolution between the original ORCA algorithm and the improved ORCA algorithm when the axis is fixed at 0.

[0099] The data trends show that the original ORCA algorithm's performance was relatively conservative and simplistic, regardless of... Regardless of how the axial deviation changes, its shortest release distance remains stable at around 4 meters (fluctuating only between 3.986 meters and 4.023 meters), which is consistent with the characteristics of a sphere safety model with a fixed radius.

[0100] In contrast, the improved ORCA algorithm demonstrates greater adaptability. With The shaft deviation ratio gradually increases (from 0.0 to 1.0) and With the reduction of the axial component, the shortest distance of the improved algorithm shows a significant decreasing trend, gradually decreasing from approximately 3.9 meters initially to approximately 2.0 meters. This result indicates that the improved algorithm can dynamically adjust the safety envelope based on the vertical offset, effectively shortening unnecessary avoidance distances in scenarios with large vertical height differences, thereby improving airspace utilization while ensuring safety.

[0101] In summary, this invention proposes a collision avoidance method combining dual-domain hierarchical collision detection and ellipsoidal maneuver constraint optimization to effectively address the problem of real-time and reliable collision avoidance for UAVs in dense dynamic airspace. First, a dual-domain hierarchical collision detection model is constructed. By introducing a pre-collision domain and an emergency collision domain into traditional relative motion analysis, early identification and risk quantification of potential collisions are achieved, providing accurate input for subsequent collision avoidance decisions. Second, an efficient collision avoidance solution method based on an ellipsoidal model is designed. Through affine transformation, the anisotropic maneuver constraint boundary is converted into a standard sphere for optimization, followed by an inverse transformation to restore it. This method strictly conforms to the actual physical characteristics of UAVs while ensuring the real-time performance and reliability of the solution. Simulation results demonstrate that the proposed method effectively improves the success rate and trajectory smoothness of collision avoidance, and outperforms traditional detection methods based on fixed thresholds and ORCA collision avoidance models with spherical assumptions in terms of both computational efficiency and safety.

Claims

1. A method for UAV conflict detection and resolution based on improved ORCA, characterized in that, The implementation steps of this method are as follows: Step 1: Collect UAV operational data; Obtain the position and velocity information of all drones currently being simulated from the Airsim simulation platform; Step 2: Unmanned Aerial Vehicle (UAV) Conflict Detection; Based on preset safety interval standards, conflict classification is performed on flight situations: (2a) Modeling of UAV conflict detection model; (2b) Conflict grading identification; determining the urgency of the conflict: Step 3: Unmanned Aerial Vehicle (UAV) Conflict Resolution; (3a) Based on the different conflict levels output during the conflict detection phase, execute the corresponding classification and resolution logic: (3b) Command feedback and update.

2. The UAV conflict detection and resolution method based on improved ORCA according to claim 1, characterized in that, In step one, the position and speed information of the drones are obtained in real time through the API provided by the simulation platform. (Two drones...) and The parameters, whose state parameters are the velocity vector, are... , With position vector , .

3. The UAV conflict detection and resolution method based on improved ORCA according to claim 1, characterized in that, Step two includes: (2a) Modeling of UAV conflict detection model; An ellipsoid is chosen as the geometric bounding box model for the UAV. By adjusting the semi-axis length parameter, the ellipsoid model accurately fits the rotor plane and fuselage height, achieving a compact representation of the space occupied by the UAV. In handling conflict detection and resolution among multiple UAVs, heterogeneous space mapping modeling based on affine transformation directly solves for tangents or minimum distances in the original space, involving complex second-order constraints. The concept of affine transformation is introduced to transform the heterogeneous expansion problem into an isotropic problem in a regularized space. An affine transformation matrix is ​​defined. and its inverse matrix Scale normalization is performed on each axis: ； Heterogeneous conflict modeling between multiple drones, specifically for two heterogeneous drones. and Their ellipsoid radii are respectively and ; Calculate the maximum ratio of the lengths of each axis. : ； This yields the equivalent ellipsoid radius after superposition. ; through matrix The mapping simplifies the originally heterogeneous collision detection problem into a regularized space with radius r. The problem of determining the equivalent sphere; (2b) Conflict grading identification; determining the urgency of the conflict: Using the drone's own center of mass as the origin, define two nested risk boundaries in the three-dimensional airspace: 1) Outer pre-collision domain: a spherical region centered on the UAV; the time window corresponding to this spherical region. Considering the current relative motion state, in the future Within a certain timeframe, it may enter the warning buffer zone for collision risk; 2) Inner Emergency Conflict Domain: An ellipsoidal region centered on the UAV; this ellipsoidal region corresponds to a time window. , < ; Consider two drones and Its state parameters are the velocity vector. , With position vector , ; Calculate relative velocity = relative position The known time window for the pre-conflict domain is... The time window for emergency conflict domains is ,when This indicates that the two aircraft are in a close approach posture, which is a necessary prerequisite for a potential conflict; the logic for determining a graded conflict is as follows: Pre-collision determination: When two aircraft are in a state of near proximity, Furthermore, at the current relative speed, the time required to reach each other's positions is less than the warning time window. When a conflict is identified, it is considered a pre-conflict; the conditions can be quantified as follows: and ; No-conflict determination: If the first determination is not a pre-conflict, then there is no conflict; Emergency Conflict Detection: When two aircraft not only approach each other, but their relative positions also pass through a time window. Constructed transformation matrix After mapping, if its magnitude is less than the set safety threshold, it is determined to be an emergency conflict; this condition can be quantified as follows: .

4. The UAV conflict detection and resolution method based on improved ORCA according to claim 1, characterized in that, Step three includes: (3a) Based on the different conflict levels output during the conflict detection phase, execute the corresponding classification and resolution logic: Conflict-free: When there is no conflict, the drone only needs to execute the predetermined instructions; Pre-collision resolution simplifies UAVs A and B into point mass models, with their relative position vectors being... The relative velocity vector is Assuming the UAV exhibits asymmetric maneuverability in the horizontal direction, it can be equivalent to an ellipsoid. First, a predefined mapping transformation matrix is ​​used. Map the ellipse to a radius of The standard circle, whose inverse transformation is denoted as . , , ; The specific calculation process is as follows: （1）； （2）； set up: Then we have: （3）； （4）； （5）； We can obtain the following by substituting formulas (1) and (2): （6）； According to the quadratic formula, we can obtain The solution is: （7）； when Pick Time represents separation The furthest one Therefore, it is effective due to the radiation. Take only Therefore, according to It can be obtained unit vector ; Therefore, we can obtain... : （8）； If Transformation matrix via inverse reflection Transforming back to the original physical space yields the pre-collision domain. Half-plane boundary points : （9）； Normal vector of a half-plane for: （10）； The specific calculation process for resolving emergency conflicts is as follows: （11）； （12）； （13）； Unit vector Therefore, the vector can be calculated. ; （14）； If Transformation matrix via inverse reflection Transforming back to the original physical space yields the pre-collision domain. Half-plane boundary points : （15）； Normal vector of a half-plane for: （16）； Finally, the final speed of liberation was determined using linear programming. (3b) Command feedback and update The speed command after release is sent to the simulation platform for execution, and the loop returns to step one to perform closed-loop detection.

Images