A method for solving multi-robot collision by using virtual force field

Abstract

This invention discloses a method for resolving multi-aircraft conflicts using a virtual force field, specifically relating to the fields of visual measurement and flight control, to address the issues of delayed response and unstable avoidance in multi-aircraft conflict resolution. The method acquires aircraft state information and environmental perception data through airborne visual sensors, constructs a virtual force field model including the aircraft, target points, and obstacles, and divides the decision responsibility areas of the aircraft based on the virtual force field. Within the decision responsibility area, an adaptive virtual decision circle is constructed for the main aircraft, and discrete virtual conflict probe points are generated on the circle. A forward-looking conflict assessment mechanism is established by combining the predicted trajectories of neighboring aircraft. Furthermore, by synthesizing classical potential field forces and flow field guiding forces based on the curvature of relative motion trajectories, a heading command that satisfies the smoothness constraints of aircraft dynamics is generated, thereby achieving safe and stable collision avoidance operation of multiple aircraft in complex scenarios.

Description

Technical Field

[0001] This invention relates to the field of visual measurement and flight control technology, and more specifically, to a method for resolving multi-aircraft conflicts using a virtual force field. Background Technology

[0002] In application scenarios where multiple aircraft collaborate to perform missions, common operating environments include low-altitude logistics transportation, urban aerial inspection, disaster monitoring formation flights, and multi-UAV collaborative search in complex airspace. In these scenarios, there are numerous aircraft, limited operating space, diverse flight mission objectives, and a lack of unified centralized scheduling mechanisms among the aircraft. Each aircraft needs to maintain autonomy while ensuring safe operation.

[0003] When multiple aircraft fly simultaneously in the same or adjacent airspace, potential conflict risks can easily arise due to intersecting flight paths, speed differences, varying mission priorities, or complex distribution of environmental obstacles. This risk is particularly pronounced in densely built-up urban areas, canyon terrain, or emergency scenarios. Existing multi-aircraft collision avoidance or conflict resolution solutions largely rely on fixed safety distances, pre-set rules, or centralized planning methods. When faced with dynamically changing traffic density, irregular obstacle distributions, and temporary mission adjustments by aircraft, these solutions often struggle to respond flexibly and promptly, leading to problems such as conservative avoidance, path oscillations, or getting stuck in local decision-making deadlocks.

[0004] Furthermore, in high-density flight scenarios, decision-making methods based solely on geometric relationships or instantaneous distances are insufficient to fully reflect the future evolution of conflicts, leading to frequent course corrections by aircraft within a short period, impacting overall operational efficiency and flight stability. A conflict resolution technology is needed that combines real-time visual perception, relative motion trends among multiple aircraft, and future conflict evolution characteristics to maintain the aircraft's autonomous decision-making capabilities while improving the safety, smoothness, and adaptability of multiple aircraft in complex dynamic scenarios. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a method for resolving multi-machine conflicts using a virtual force field to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for resolving multi-machine conflicts using a virtual force field includes the following steps: S1. Use airborne visual sensors to measure and acquire information on the initial state, target point and environmental obstacles of all aircraft, establish a virtual force field model and mark the corresponding decision responsibility area; S2. Based on the scope of the decision responsibility area and the geometric characteristics of the internal conflicts, construct a local virtual decision circle for the main aircraft. The radius of the circle is adjusted in the opposite direction to the local traffic density and approach speed. S3. On the virtual decision circle, discrete virtual conflict probe points are generated based on the predicted trajectories of neighboring machines, and a prospective conflict cost function containing the time dimension is associated with each probe point. S4. For each virtual conflict probe point, calculate the integrated virtual force in parallel at the current moment and in the finite future time domain. The integrated virtual force is composed of the classical potential force and a flow field guiding force component based on the curvature of the relative motion trajectory. S5. Evaluate the combined virtual force corresponding to all virtual conflict probe points, introduce a random perturbation factor to perturb the combined virtual force, and select the probe point with the largest resultant force vector norm after perturbation as the current optimal escape point. S6. Based on the current optimal escape point, solve for the heading angle sequence that satisfies the aircraft dynamics smoothness constraint, and take the first term of the sequence as the instantaneous heading command.

[0007] As a further aspect of the present invention, in step S1, establishing a virtual force field model and labeling the corresponding decision responsibility area specifically includes: The real-time position, velocity vector and coordinates of preset target points of all aircraft are obtained by measuring the airborne visual sensors, and the spatial distribution model of obstacles is constructed by fusing environmental perception data. Based on the spatial distribution model, corresponding virtual mass parameters are assigned to each aircraft, target point and obstacle entity. The virtual mass parameters of the aircraft are dynamically determined according to the mission priority of its current flight phase. Using the spatial coordinates of all assigned virtual mass parameters, a global virtual potential field is generated. This potential field forms a gravitational potential well at the target point coordinates and a repulsive potential peak at the coordinates of the spacecraft and the obstacle. Based on the spatial gradient distribution of the global virtual potential field, a decision responsibility region centered on the current position is defined for each aircraft. The boundary of the decision responsibility region is defined by an equipotential surface whose magnitude of the virtual potential field gradient exceeds a preset threshold value.

[0008] As a further aspect of the present invention, the construction of the spatial distribution model of the obstacles specifically includes: The airborne vision sensor acquires a continuous field-of-view image sequence that includes other aircraft and environmental scenes, identifies the depth information of obstacles within the field of view, and generates corresponding three-dimensional point cloud data. The contour features of adjacent aircraft are identified from the image sequence using a target recognition algorithm. Based on the pixel displacement of the contour features between consecutive frames and the time interval known to the sensor, the relative position change rate of other aircraft relative to the local aircraft is calculated to obtain the relative velocity vector. By integrating relative velocity vectors, 3D point cloud data, and neighboring aircraft status data obtained through inter-aircraft communication links, a spatial distribution model is generated that includes the real-time positions and velocity vectors of all aircraft and the geometric contours of obstacles.

[0009] As a further aspect of the present invention, in S2, constructing a local virtual decision-making circle for the main aircraft based on the scope of the decision responsibility area and the geometric characteristics of internal conflicts specifically includes: Obtain the designated decision responsibility area of ​​the main aircraft, and count the number of other aircraft located within the decision responsibility area as a measure of local traffic density; Calculate the magnitude of the relative velocity vector between the main aircraft and every other aircraft within the decision responsibility area, and take the maximum value as the representative approach velocity; Based on the local traffic density and the representative approach speed, an adaptive radius value is calculated based on a preset inverse proportional relationship function. The virtual decision circle is constructed in a two-dimensional flight plane with the adaptive radius value as the radius and the current position of the main aircraft as the center.

[0010] As a further aspect of the present invention, in step S3, generating discrete virtual conflict probe points on the virtual decision circle based on the predicted trajectories of neighboring machines, and associating each probe point with a forward-looking conflict cost function containing a time dimension specifically includes: Obtain the real-time velocity vectors and heading angles of all neighboring aircraft of the main aircraft, and deduce the predicted trajectory coordinates of the neighboring aircraft at a series of future time points based on the uniform motion model; On the constructed virtual decision circle, initial probe points are generated based on fixed angle intervals. Based on the geometric intersection relationship between the predicted trajectory of the neighboring aircraft and the predicted trajectory of the main aircraft, the probe point density is increased in the arc segment of the circle where potential conflicts are predicted, forming a discrete set of virtual conflict probe points. For each virtual conflict probe point, a prospective conflict cost function is associated. The independent variables of the conflict cost function include the estimated time for the main aircraft to fly straight from its current position to the probe point, and the position sequence of the neighboring aircraft's predicted trajectory relative to the main aircraft within that estimated time. Based on the predicted time and position sequence, the sum of the reciprocals of the minimum relative distances between the main aircraft and each neighboring aircraft in the predicted trajectory is calculated in the time dimension and used as the output value of the prospective conflict cost function.

[0011] As a further aspect of the present invention, in step S4, the parallel calculation of the comprehensive virtual force of each virtual conflict probe point in the current moment and within a finite future time domain specifically includes: For each virtual conflict probe point, based on the virtual mass and critical distance of each entity in the established virtual force field model, the attractive force vector from its own target point and the repulsive force vector from all other aircraft and obstacles are calculated. The sum of the above vectors is used as the classical potential force at that point. Based on the obtained predicted trajectory sequence of the neighboring aircraft and the predicted trajectory of the main aircraft from its current position to the probe point, the rate of change of direction of the line connecting the consecutive position points in the relative position sequence of the two is calculated, and a flow field guiding force vector is derived based on the rate of change of direction and the magnitude of the relative motion velocity. The classical potential force vector and the flow field guiding force vector are vector synthesized to obtain the comprehensive virtual force of the virtual collision probe point at the current moment; For multiple equally spaced time points within a finite future time domain, the above calculation process is repeated, and the comprehensive virtual force obtained at each time point is weighted and accumulated according to the time distance to obtain the forward-looking comprehensive virtual force of the probe point that covers the finite future time domain.

[0012] As a further aspect of the present invention, in S5, evaluating the comprehensive virtual force corresponding to all virtual conflict probe points, introducing a random perturbation factor to perturb the comprehensive virtual force, and selecting the probe point with the largest resultant force vector norm after perturbation as the current optimal escape point specifically includes: Obtain the forward-looking comprehensive virtual force vector calculated for each virtual conflict probe point; A random perturbation vector is generated by sampling from a preset random distribution. The random perturbation vector is then vector-synthesized with the corresponding forward-looking integrated virtual force vector to generate the resultant force vector after perturbation. Calculate the Euclidean norm of the resultant force vector after all perturbations, and use it as the evaluation index for each corresponding virtual conflict probe point; Compare the evaluation metrics of all virtual collision probe points, and select the virtual collision probe point with the largest evaluation metric value as the current optimal escape point.

[0013] As a further aspect of the present invention, in step S6, based on the current optimal escape point, a heading angle sequence satisfying the aircraft dynamics smoothness constraint is solved, and the first term of this sequence is used as an immediate heading command, specifically including: The target heading angle increment is calculated based on the current position coordinates of the main aircraft and the current optimal escape point coordinates. Based on the target heading angle increment and the heading angle of the main aircraft at the current moment, the desired target heading angle is derived. By utilizing the preset maximum angular velocity limit of the aircraft and the conflict control cycle, the target heading angle is decomposed into a smooth sequence of heading angle commands within multiple consecutive control cycles, which is then output as an instant heading command.

[0014] The technical effects and advantages of the method for resolving multi-machine conflicts using a virtual force field according to the present invention are as follows: By introducing a collaborative mechanism of decision responsibility region and local virtual decision circle during multi-aircraft operation, the conflict handling behavior of each aircraft is confined to a local spatial range directly related to its operational state. This avoids the computational burden and decision lag of traditional global collision avoidance strategies, thereby improving real-time response capabilities in high-density flight scenarios. By deploying discrete virtual conflict probe points on the virtual decision circle and constructing a forward-looking conflict cost function with a time dimension in conjunction with neighboring aircraft predicted trajectories, aircraft can consider not only current spatial relationships when making heading decisions but also comprehensively assess the development trend of potential conflicts over a future period, effectively reducing flight oscillations caused by frequent heading corrections within short periods. Synthesizing classical potential field forces with flow field guiding forces based on the curvature of relative motion trajectories allows the generated integrated virtual force to simultaneously reflect spatial safety constraints and motion trend guidance, helping aircraft complete avoidance maneuvers along smoother directions with a greater tendency to escape conflict. Introducing a controlled random perturbation mechanism to micro-perturb the integrated virtual force effectively avoids the problem of getting stuck in local extrema or decision deadlock in multi-aircraft symmetrical scenarios, thereby improving the robustness and collaborative stability of the overall system in complex dynamic environments. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a method for resolving multi-machine conflicts using a virtual force field according to the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1 Figure 1 This invention presents a method for resolving multi-machine conflicts using a virtual force field, comprising the following steps: S1. Use airborne visual sensors to measure and acquire information on the initial state, target point and environmental obstacles of all aircraft, establish a virtual force field model and mark the corresponding decision responsibility area; S2. Based on the scope of the decision responsibility area and the geometric characteristics of the internal conflicts, construct a local virtual decision circle for the main aircraft. The radius of the circle is adjusted in the opposite direction to the local traffic density and approach speed. S3. On the virtual decision circle, discrete virtual conflict probe points are generated based on the predicted trajectories of neighboring machines, and a prospective conflict cost function containing the time dimension is associated with each probe point. S4. For each virtual conflict probe point, calculate the integrated virtual force in parallel at the current moment and in the finite future time domain. The integrated virtual force is composed of the classical potential force and a flow field guiding force component based on the curvature of the relative motion trajectory. S5. Evaluate the combined virtual force corresponding to all virtual conflict probe points, introduce a random perturbation factor to perturb the combined virtual force, and select the probe point with the largest resultant force vector norm after perturbation as the current optimal escape point. S6. Based on the current optimal escape point, solve for the heading angle sequence that satisfies the aircraft dynamics smoothness constraint, and take the first term of the sequence as the instantaneous heading command.

[0018] In step S1, a virtual force field model is established and the corresponding decision responsibility area is marked.

[0019] During flight, the main aircraft continuously utilizes its onboard visual sensor to continuously observe the airspace within a certain field of view in front of and to the sides of the current flight direction. The onboard visual sensor acquires a continuous sequence of images containing other aircraft and natural or artificial environmental elements at a fixed frame rate. Each frame retains complete spatial perspective information and timestamp information for subsequent spatial reconstruction processing. For the acquired continuous image sequence, the environmental region and moving target region in the image are first distinguished. Depth estimation is performed on regions with significant texture changes, continuous edges, and no rapid changes with the viewing angle to extract the corresponding obstacle depth information. During the depth estimation process, a multi-frame disparity consistency constraint is used to align the disparity changes of the same obstacle in the continuous field of view, thereby obtaining a stable depth distribution result. After obtaining the depth information corresponding to each pixel, the two-dimensional image coordinates and depth values ​​are combined and mapped to the spatial coordinate system according to the sensor's own imaging geometry, generating three-dimensional point cloud data reflecting the shape contour and spatial occupancy of obstacles point by point. The generated 3D point cloud data not only includes the geometry of the obstacle surface, but also preserves the spatial distribution of the obstacle relative to the main aircraft at the current moment.

[0020] After constructing the 3D point cloud of environmental obstacles, further analysis and processing are performed on moving targets in the continuous field-of-view image sequence to obtain relative motion information of adjacent aircraft. Specifically, in each frame, target regions with obvious rigid body features, continuous contours, and overall displacement over time are identified, and corresponding aircraft contour features are extracted. These contour features include edge segments, corner distribution, and scale range of the aircraft shape, used to establish stable feature correspondences between adjacent frames. Between consecutive frames, the pixel position changes of the same aircraft contour feature are matched, recording its displacement direction and amount in the image coordinate system. Combined with the known sampling time interval of the airborne vision sensor, this pixel displacement is converted into a relative position change rate per unit time. Simultaneously, combined with the 3D point cloud information obtained in the previous step, the 2D pixel displacement is mapped to 3D spatial coordinates, thereby calculating the relative velocity vector of adjacent aircraft relative to the host aircraft. This relative velocity vector fully reflects the direction and speed of the neighboring aircraft's motion in space.

[0021] After acquiring the relative velocity vectors of adjacent aircraft and the 3D point cloud data of environmental obstacles, information from different sources is fused to construct a complete spatial distribution model. Specifically, the relative velocity vectors obtained from visual calculations are first bound to the corresponding spatial positions of the aircraft, forming a neighboring aircraft state description with the main aircraft as the reference coordinate system. Simultaneously, state data periodically broadcast by neighboring aircraft is received via an inter-aircraft communication link. This state data includes at least the current position and velocity direction information measured by the neighboring aircraft itself. The data acquired through the communication link is compared with the visual calculation results for consistency. When there is a slight deviation in spatial position or velocity direction, the visual calculation results are used as the primary basis for correction and fusion, thereby avoiding the adverse effects of communication delays on spatial judgment. Subsequently, the fused multi-aircraft state information and the aforementioned obstacle 3D point cloud data are uniformly mapped to the same spatial coordinate system. The geometric contours of the obstacles, the spatial positions of the aircraft, and their respective motion directions are organized holistically to generate a spatial distribution model containing the real-time positions and velocity vectors of all aircraft and the geometric shapes of the obstacles.

[0022] In the completed spatial distribution model, the various entity types included in the model are clearly identified, including each aircraft entity participating in cooperative flight, the preset target point entities corresponding to each aircraft, and all obstacle entities identified in the model. For different entity types, virtual mass parameter allocation is performed according to their different roles in the flight conflict resolution process. Specifically, target point entities are first uniformly processed, with their virtual mass parameters set to a fixed high value to reflect the continuous and stable guiding role of the target point on the aircraft in flight decision-making. For example, in inspection or delivery missions, the termination point or intermediate track point of the current mission segment is used as the target point, and its virtual mass parameter is set to a value range significantly higher than that of ordinary obstacles. Subsequently, virtual mass parameters are allocated to obstacle entities. This allocation is determined based on the geometric dimensions and occupied area of ​​the obstacle in the spatial distribution model. The larger the occupied space and the more stable the shape of the obstacle, the higher its virtual mass parameter is set to ensure a stable and significant repulsive effect in the subsequent potential field construction. For the aircraft entity, its virtual mass parameters are dynamically determined based on the mission priority corresponding to the current flight phase. Flight phases are explicitly divided into several discrete levels through flight control logic, such as mission execution phase, normal cruise phase, and emergency avoidance phase, each corresponding to a different mission priority level. During the emergency avoidance or critical mission execution phase, the aircraft's virtual mass parameters are set to higher values ​​than in other phases to reflect its priority avoidance weight in the potential field; during the normal cruise phase, relatively lower virtual mass parameters are used.

[0023] After allocating virtual mass parameters for various entities, a global virtual potential field is generated based on the spatial coordinates of all entities in the spatial distribution model. Specifically, the unified spatial coordinate system used by the spatial distribution model is used as the reference coordinate system for potential field construction, and all entities with allocated virtual mass parameters are mapped to this coordinate system one by one. During potential field generation, different potential field rules are applied to different entity types: for target point entities, a potential well structure with significant attraction characteristics is generated at their spatial coordinate positions. This potential well exhibits a gravitational tendency that gradually weakens with distance in space, ensuring that the aircraft always has a driving force towards the target point under the influence of the potential field; for aircraft and obstacle entities, a repulsive potential peak is generated at their corresponding spatial coordinate positions. This peak forms a spatial repulsion region around the entity, causing other aircraft to experience a significant repulsion tendency when approaching this region. In actual construction, the potential field influences generated by each entity are not isolated but are superimposed according to their spatial positional relationships, thus forming a global virtual potential field covering the entire spatial distribution model. The global virtual potential field is continuously distributed in space without any abrupt breaks, and can simultaneously reflect the guidance trend of the target point and the safety constraint trend of the aircraft and obstacles.

[0024] After obtaining the global virtual potential field, the decision responsibility area is delineated for each aircraft. Using the real-time position of the corresponding aircraft in the spatial distribution model as the center point, the spatial gradient distribution of the global virtual potential field is sampled in the surrounding space. The spatial gradient reflects the intensity of the potential field's change in different directions, characterizing the degree of change in spatial constraints and guidance trends near that position. In the actual delineation process, a gradient magnitude threshold is pre-set to define the significance of potential field changes. This threshold is set based on flight safety experience; for example, a gradient change level that will not cause significant heading correction under normal flight conditions can be selected as a reference. Using this threshold as the criterion, the search expands outward from the aircraft's current position. When the potential field gradient magnitude first exceeds the threshold, the corresponding position is considered the boundary point of the decision responsibility area. By performing consistent gradient detection in all directions around the aircraft, a closed region centered on the aircraft's current position and surrounded by equipotential surfaces that satisfy the gradient conditions is ultimately formed, defined as the aircraft's decision responsibility area. Spatial changes within this region are considered to have a direct impact on the aircraft's current decision-making, while potential field changes outside this region are not considered in this round of decision-making. This method of delineating the decision-making responsibility area based on potential field gradients allows for dynamic adjustments to the decision-making responsibility area in response to changes in the spatial environment and entity distribution, ensuring that the aircraft's decision-making scope always matches its current risk and constraint environment.

[0025] In S2, a local virtual decision-making circle is constructed for the main aircraft based on the scope of the decision responsibility area and the geometric characteristics of internal conflicts.

[0026] After defining its own decision-making responsibility area, the primary aircraft first scans and statistically processes the spatial state within that area to obtain traffic density information reflecting the current level of local airspace congestion. Using the primary aircraft's current position as a reference, and the spatial range within its decision-making responsibility area as the statistical boundary, the positions of other aircraft entities identified in the spatial distribution model are determined one by one. When the real-time position coordinates of other aircraft are within the boundary of the decision-making responsibility area, that aircraft is included in the statistical range of local traffic density. This statistical process does not distinguish between aircraft mission types or flight altitude differences, but uniformly judges based on spatial occupancy relationships to ensure that the statistical results accurately reflect the concentration of aircraft in the airspace currently occupied by the primary aircraft. In actual operation, local traffic density is represented in integer form; for example, in low-density scenarios, the statistical result is one or two neighboring aircraft, and in high-density scenarios, the statistical result is three or more neighboring aircraft.

[0027] The relative motion states between the primary aircraft and other aircraft within the decision-making responsibility area are analyzed and processed. Specifically, for each other aircraft within the decision-making responsibility area, its relative velocity vector information relative to the primary aircraft is read. This relative velocity vector, obtained through the aforementioned visual measurement and communication fusion steps, includes a comprehensive description of the relative motion direction and speed. For each relative velocity vector, its corresponding velocity magnitude is calculated to characterize the strength of the neighboring aircraft's tendency to approach or move away from the primary aircraft at the current moment. Among all the relative velocity vectors involved in the statistics, the one with the largest velocity magnitude is selected as the representative approach velocity to reflect the most pressing approach situation faced by the primary aircraft within the current decision-making responsibility area. By selecting the maximum value, it ensures that subsequent decision-making processes focus on the relative motion relationships that have the greatest impact on the safety of the primary aircraft, without being influenced by multiple low-speed or moving-away aircraft.

[0028] After obtaining the local traffic density and representative approach speed, a virtual decision circle is constructed based on their combined relationship. A set of inverse proportionality rules is pre-defined to describe the changing pattern of the decision radius. These rules explicitly state that when the local traffic density or representative approach speed increases, the radius of the virtual decision circle decreases accordingly; when both are at low levels, the radius increases accordingly. This inverse proportionality is set empirically. For example, in low-density, low-approach-speed scenarios, the decision radius is set to a larger range to improve the flexibility of heading selection; in high-density or high-approach-speed scenarios, the decision radius is reduced to a smaller range to enhance the fine control of local avoidance. Based on this inverse proportionality rule, combined with the currently obtained statistical values ​​of local traffic density and representative approach speed, a uniquely determined adaptive radius value is calculated. Subsequently, using this adaptive radius value as the circle radius and the current position coordinates of the main aircraft as the center, a virtual decision circle is constructed in a two-dimensional flight plane.

[0029] In step S3, discrete virtual conflict probe points are generated on the virtual decision circle based on the predicted trajectories of neighboring machines, and a prospective conflict cost function containing a time dimension is associated with each probe point.

[0030] After constructing the virtual decision circle, the main aircraft uniformly acquires and organizes the motion states of its neighboring aircraft. It reads the real-time velocity vectors and corresponding heading angles of all identified neighboring aircraft from the spatial distribution model; this information reflects the direction and intensity of motion of the neighboring aircraft at the current moment. Based on this, a uniform motion assumption is used to predict the trajectory of the neighboring aircraft in the near future. This uniform motion assumption is explicitly limited to a finite time range, such as a few seconds, to ensure that the prediction results are consistent with the actual flight state. During the prediction process, starting from the current moment, a series of future time points are generated at fixed time intervals, and the predicted spatial position coordinates of the neighboring aircraft are calculated at each time point, thus forming a sequence of predicted trajectories for the neighboring aircraft. Subsequently, initial probe points are generated at fixed angular intervals on the constructed virtual decision circle. These angular intervals are set empirically, for example, by dividing the circle into several directional sectors to ensure uniform coverage of the circle's directions. Next, a geometric relationship analysis is performed between the predicted trajectory of the neighboring aircraft and the predicted trajectory of the main aircraft while maintaining its current heading. If it is found that the main aircraft, moving in that direction, has a spatial intersection or significant approach tendency with the predicted trajectory of the neighboring aircraft, this arc segment is determined to be a potential conflict arc segment. Within the potential conflict arc segment, the probe point deployment density is increased by reducing the angle interval, while the original probe point density is maintained in arc segments without conflict tendency.

[0031] After generating a set of virtual conflict probe points, a forward-looking conflict cost function is constructed for each probe point. First, the estimated time required for the primary aircraft to travel from its current position to the corresponding probe point along a straight path is determined. This estimated time is calculated using the primary aircraft's current speed and the spatial distance between the probe point and the current position, and is limited to the time range of the predicted trajectories of neighboring aircraft to ensure time scale consistency. Then, within this estimated time range, the spatial position sequences for the corresponding time periods in all predicted trajectories of neighboring aircraft are extracted and synchronized with the predicted position of the primary aircraft traveling in a straight line within that time period, forming a set of relative position sequences indexed by time. The input variables of the forward-looking conflict cost function explicitly include two aspects: the estimated time length for the primary aircraft to reach the probe point, and the relative position changes between the primary aircraft and each neighboring aircraft within that time span. By simultaneously incorporating time factors and spatial relative relationships into the cost function definition, each probe point not only reflects the immediate safety in a certain spatial direction but also reflects the evolution trend of potential conflicts over a future period while moving along that direction. This cost function is output numerically for subsequent quantitative comparisons between different probe points.

[0032] After determining the input elements of the forward-looking conflict cost function, the output value of the cost function is calculated. For each virtual conflict probe point, the predicted trajectories of the main aircraft and each neighboring aircraft are analyzed time-by-time within its corresponding predicted time range. At each predicted time point, the relative distance between the predicted position of the main aircraft and the predicted position of the neighboring aircraft is calculated, and the minimum value of this relative distance is recorded over the entire predicted time span to reflect the likelihood of the closest approach in that direction. Subsequently, the reciprocal of the minimum relative distance is taken, so that a smaller distance corresponds to a higher risk. For the case of multiple neighboring aircraft, the reciprocal of the minimum relative distance obtained for each neighboring aircraft is accumulated over time using a direct summation method to ensure that the risk level can be effectively superimposed when multiple neighboring aircraft exist simultaneously. Through the above calculation method, a unique forward-looking conflict cost output value is generated for each virtual conflict probe point. The larger the output value, the higher the overall risk of close approach or conflict in the future predicted time when moving along the direction of the probe point.

[0033] In S4, for each virtual conflict probe point, its comprehensive virtual force at the current moment and in the finite future time domain is calculated in parallel.

[0034] For each generated virtual conflict probe point, based on the previously established virtual force field model, the classical potential force experienced by that probe point at the current moment is calculated point by point. The scope of entities involved in the potential field interaction is clearly defined, including the target point entity of the main aircraft itself, other aircraft entities within the decision-making responsibility area, and all obstacle entities identified in the spatial distribution model. For the target point entity, a continuous and stable attractive force is formed at its spatial coordinate position. The strength of the attractive force is determined by the virtual mass parameter corresponding to the target point and gradually decreases with the change in spatial distance between the probe point and the target point. For other aircraft entities and obstacle entities, a repulsive force is formed at their spatial coordinate positions. The strength of the repulsive force is constrained by the virtual mass parameter of the corresponding entity and a set critical distance. The critical distance is set based on flight safety experience, for example, selecting a safe distance that requires advance avoidance under normal flight conditions as a reference. When the distance between the probe point and an entity is less than this critical distance, the repulsive force is significantly enhanced; when the distance is greater than this critical distance, the repulsive force gradually weakens. For each probe point, the directions of the attractive forces from the target point and the repulsive forces from various aircraft and obstacles are calculated. All attractive and repulsive forces are then combined in terms of direction and intensity to obtain the classical potential force vector corresponding to that probe point at the current moment. This classical potential force fully reflects the combined influence of spatial static constraints and target guidance on the probe point's direction.

[0035] After completing the classical potential force calculation, a flow field guiding force calculation process based on relative motion trends is introduced for each virtual conflict probe point. First, the previously generated sequence of predicted trajectories for neighboring aircraft and the predicted trajectory of the main aircraft flying in a straight line from its current position to the probe point are obtained. These predicted trajectories are described in discrete time points, with consistent time intervals between each point, for example, by selecting a fixed time step for trajectory sampling. Then, the relative positions of the main aircraft's predicted trajectory and the neighboring aircraft's predicted trajectory at the same time index are calculated time-by-time, forming a sequence of relative positions arranged chronologically. In this sequence, the relative position vectors of adjacent time points are connected, and the directional change between adjacent connecting lines is calculated to characterize the degree of curvature of the relative motion trajectories of the main aircraft and the neighboring aircraft in that probe direction. A larger rate of directional change indicates a significant trajectory intersection or detour trend in space. Combining this rate of directional change with the relative velocity within the corresponding time period, a flow field guiding force vector with a clear direction is generated. The flow field guiding force does not depend on static distance constraints, but reflects the dynamic changing trend of relative motion geometry. Its direction is conducive to reducing the probability of trajectory intersection and alleviating future motion conflicts in the spatial direction, thus supplementing the characterization of the decision value of the probe point.

[0036] After obtaining the classical potential force vector and the flow field guiding force vector respectively, the two types of forces are vector-synthesized to obtain the comprehensive virtual force of each virtual conflict probe point at the current moment. During the synthesis process, their respective directional characteristics are preserved, and a comprehensive result reflecting spatial constraints, target guidance, and relative motion trends is formed through vector superposition. After obtaining the comprehensive virtual force at the current moment, to introduce a forward-looking judgment on the future conflict evolution, the calculation process is extended within a finite time domain. Specifically, a finite prediction time domain length is pre-set, for example, covering several consecutive control cycles, and multiple equally spaced time points are selected within this time domain. At each time point, the corresponding classical potential force, flow field guiding force, and comprehensive virtual force are recalculated in the same manner as described above. Subsequently, the comprehensive virtual forces at each time point are weighted according to their time proximity, with higher weights assigned to comprehensive virtual forces closer to the current moment and relatively lower weights assigned to comprehensive virtual forces further back in time, for example, by using a decreasing weight sequence for accumulation. By accumulating the comprehensive virtual forces at each time point according to their weights, the forward-looking comprehensive virtual force of the virtual conflict probe point is finally obtained. This forward-looking comprehensive virtual force reflects both the current and future comprehensive hedging trends in terms of both numerical value and direction.

[0037] In step S5, the comprehensive virtual force corresponding to all virtual conflict probe points is evaluated, a random perturbation factor is introduced to perturb the comprehensive virtual force, and the probe point with the largest resultant force vector norm after perturbation is selected as the current optimal escape point.

[0038] For each probe point, random perturbation processing is performed. First, the forward-looking integrated virtual force vector corresponding to each virtual conflict probe point is obtained one by one. This vector comprehensively reflects multiple factors such as target guidance, spatial constraints, and future relative motion trends. Then, a random perturbation vector is generated by sampling from a preset random distribution to slightly perturb the forward-looking integrated virtual force. The random distribution is set to a zero-mean, finite-amplitude distribution. The perturbation amplitude is constrained by flight control experience, for example, set to a small fraction of the forward-looking integrated virtual force amplitude, to ensure that the perturbation does not change the overall decision-making direction. The direction of the random perturbation vector is uniformly distributed in space, and its magnitude is limited by the aforementioned amplitude, thus ensuring that in scenarios where multiple probe points are symmetrical or approximately symmetrical, the numerical consistency of the resultant force direction can be effectively broken. After obtaining the random perturbation vector, it is vector-synthesized with the forward-looking integrated virtual force vector of the corresponding probe point to generate the perturbated resultant force vector. While maintaining the original overall trend, this resultant force vector introduces controlled uncertainty, causing numerically distinguishable differences between different probe points. This avoids decision-making stagnation or repeated oscillations in complex symmetrical conflict scenarios. Through the aforementioned explicit random perturbation synthesis method, it is ensured that each virtual conflict probe point corresponds to a uniquely determined resultant force vector after perturbation.

[0039] After generating the resultant force vectors of all virtual conflict probe points after perturbation, each probe point is uniformly evaluated and ranked to determine the current optimal escape point. For each virtual conflict probe point, the spatial norm of its corresponding resultant force vector after perturbation is calculated. This norm characterizes the strength of the overall escape trend in the direction of that probe point. The norm calculation process uses a unified Euclidean distance metric, regardless of the type of force source, focusing only on the overall magnitude of the resultant force vector, thus ensuring consistency in evaluation standards among different probe points. After completing the norm calculations for all probe points, the evaluation indicators of each probe point are compared centrally and ranked according to their numerical values. In the ranking results, the virtual conflict probe point with the largest evaluation indicator value is determined to be the spatial direction most conducive to the main aircraft's rapid escape from the potential conflict zone in the current moment and within the finite future time domain. This probe point is then defined as the current optimal escape point, and its corresponding spatial direction is used to guide the main aircraft's subsequent heading decisions. By employing the maximum resultant force norm as the selection criterion, the selected escape point not only satisfies the avoidance constraint but also demonstrates a clear tendency to escape conflict, which is beneficial for forming decisive and stable collision avoidance behavior in high-density, multi-objective interaction scenarios. Furthermore, due to the aforementioned introduction of a controlled random perturbation mechanism, even when the original resultant forces of multiple probe points are close, a unique optimal escape point can be stably selected, avoiding uncertainties or frequent switching in decision-making outcomes.

[0040] In step S6, based on the current optimal escape point, a heading angle sequence that satisfies the aircraft dynamics smoothness constraint is solved, and the first term of the sequence is used as the instantaneous heading command.

[0041] After determining the current optimal escape point, the main aircraft calculates the corresponding target heading angle based on its current spatial state. This is achieved by acquiring the real-time position coordinates of the main aircraft at the current moment and simultaneously acquiring the coordinates of the current optimal escape point in the same spatial coordinate system. Then, starting from the main aircraft's current position, a spatial line is constructed pointing towards the optimal escape point. The direction corresponding to this line in the two-dimensional flight plane is the desired flight direction. By performing angle analysis on this spatial line direction, the heading angle increment between the main aircraft's current heading and the desired direction is obtained. This heading angle increment reflects the turning amplitude required for the main aircraft to adjust from its current heading to the escape direction. After obtaining the heading angle increment, it is combined with the main aircraft's current heading angle to derive the desired target heading angle. This target heading angle serves as the direct output of the collision avoidance decision, explicitly indicating the spatial direction the main aircraft should ideally point in.

[0042] After obtaining the desired target heading angle, it is smoothed to meet aircraft dynamics constraints and avoid abrupt control commands. During implementation, a maximum permissible angular velocity limit for the aircraft is pre-set, determined based on the aircraft's structural characteristics and flight safety requirements; for example, it is set as the maximum allowable heading angle range per unit time. Simultaneously, the length of the conflict control cycle is defined, which is a fixed time interval for updating flight control commands. Based on this, the maximum allowable heading adjustment within a single control cycle is calculated according to the difference between the target heading angle and the current heading angle, combined with the maximum angular velocity limit. Subsequently, the target heading angle is decomposed into a heading angle sequence that gradually approaches the target heading angle across multiple consecutive control cycles, according to the aforementioned maximum adjustment constraint. This heading angle sequence changes continuously over time, and the heading change between adjacent cycles is always limited by the maximum angular velocity requirement, thus forming a smooth and executable turning process. In actual output, only the first item in the heading angle sequence is sent as an immediate heading command to the aircraft actuators; subsequent heading angles are re-evaluated and updated in the next control cycle. This decomposition and step-by-step output method ensures that the aircraft's movements are smooth and its response is controllable when performing collision avoidance maneuvers. At the same time, it can continuously correct its heading commands according to environmental changes, maintaining the stability and safety of the collision avoidance process.

[0043] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0044] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0045] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0046] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0047] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0048] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0049] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0050] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0051] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for solving multi-robot collision using virtual force field, characterized in that, Includes the following steps: S1. Use airborne visual sensors to measure and acquire information on the initial state, target point and environmental obstacles of all aircraft, establish a virtual force field model and mark the corresponding decision responsibility area; S2. Based on the scope of the decision responsibility area and the geometric characteristics of the internal conflicts, construct a local virtual decision circle for the main aircraft. The radius of the circle is adjusted in the opposite direction to the local traffic density and approach speed. S3. On the virtual decision circle, discrete virtual conflict probe points are generated based on the predicted trajectories of neighboring machines, and a prospective conflict cost function containing the time dimension is associated with each probe point. S4. For each virtual conflict probe point, calculate the integrated virtual force in parallel at the current moment and in the finite future time domain. The integrated virtual force is composed of the classical potential force and a flow field guiding force component based on the curvature of the relative motion trajectory. S5. Evaluate the combined virtual force corresponding to all virtual conflict probe points, introduce a random perturbation factor to perturb the combined virtual force, and select the probe point with the largest resultant force vector norm after perturbation as the current optimal escape point. S6. Based on the current optimal escape point, solve for the heading angle sequence that satisfies the aircraft dynamics smoothness constraint, and take the first term of the sequence as the instantaneous heading command.

2. The method for solving multi-robot collision using virtual force field according to claim 1, wherein, In step S1, establishing the virtual force field model and labeling the corresponding decision-making responsibility area specifically includes: The real-time position, velocity vector and coordinates of preset target points of all aircraft are obtained by measuring the airborne visual sensors, and the spatial distribution model of obstacles is constructed by fusing environmental perception data. Based on the spatial distribution model, corresponding virtual mass parameters are assigned to each aircraft, target point and obstacle entity. The virtual mass parameters of the aircraft are dynamically determined according to the mission priority of its current flight phase. Using the spatial coordinates of all assigned virtual mass parameters, a global virtual potential field is generated. This potential field forms a gravitational potential well at the target point coordinates and a repulsive potential peak at the coordinates of the spacecraft and the obstacle. Based on the spatial gradient distribution of the global virtual potential field, a decision responsibility region centered on the current position is defined for each aircraft. The boundary of the decision responsibility region is defined by an equipotential surface whose magnitude of the virtual potential field gradient exceeds a preset threshold value.

3. The method for solving multi-robot collision using virtual force field according to claim 2, wherein, The construction of the spatial distribution model of the obstacles specifically includes: The airborne vision sensor acquires a continuous field-of-view image sequence that includes other aircraft and environmental scenes, identifies the depth information of obstacles within the field of view, and generates corresponding three-dimensional point cloud data. The contour features of adjacent aircraft are identified from the image sequence using a target recognition algorithm. Based on the pixel displacement of the contour features between consecutive frames and the time interval known to the sensor, the relative position change rate of other aircraft relative to the local aircraft is calculated to obtain the relative velocity vector. By integrating relative velocity vectors, 3D point cloud data, and neighboring aircraft status data obtained through inter-aircraft communication links, a spatial distribution model is generated that includes the real-time positions and velocity vectors of all aircraft and the geometric contours of obstacles.

4. The method for solving multi-robot collision using virtual force field according to claim 1, wherein, In step S2, constructing a local virtual decision-making circle for the main aircraft based on the scope of the decision-making responsibility area and the geometric characteristics of internal conflicts specifically includes: Obtain the designated decision responsibility area of ​​the main aircraft, and count the number of other aircraft located within the decision responsibility area as a measure of local traffic density; Calculate the magnitude of the relative velocity vector between the main aircraft and every other aircraft within the decision responsibility area, and take the maximum value as the representative approach velocity; Based on the local traffic density and the representative approach speed, an adaptive radius value is calculated based on a preset inverse proportional relationship function. The virtual decision circle is constructed in a two-dimensional flight plane with the adaptive radius value as the radius and the current position of the main aircraft as the center.

5. The method for solving multi-robot collision using virtual force field according to claim 1, wherein, In step S3, discrete virtual conflict probe points are generated on the virtual decision circle based on the predicted trajectories of neighboring machines, and a prospective conflict cost function containing a time dimension is associated with each probe point. Specifically, this includes: Obtain the real-time velocity vectors and heading angles of all neighboring aircraft of the main aircraft, and deduce the predicted trajectory coordinates of the neighboring aircraft at a series of future time points based on the uniform motion model; On the constructed virtual decision circle, initial probe points are generated based on fixed angle intervals. Based on the geometric intersection relationship between the predicted trajectory of the neighboring aircraft and the predicted trajectory of the main aircraft, the probe point density is increased in the arc segment of the circle where potential conflicts are predicted, forming a discrete set of virtual conflict probe points. For each virtual conflict probe point, a prospective conflict cost function is associated. The independent variables of the conflict cost function include the estimated time for the main aircraft to fly straight from its current position to the probe point, and the position sequence of the neighboring aircraft's predicted trajectory relative to the main aircraft within that estimated time. Based on the predicted time and position sequence, the sum of the reciprocals of the minimum relative distances between the main aircraft and each neighboring aircraft in the predicted trajectory is calculated in the time dimension and used as the output value of the prospective conflict cost function.

6. The method for solving multi-robot collision using virtual force field according to claim 1, wherein, In step S4, for each virtual conflict probe point, the parallel calculation of its comprehensive virtual force in the current moment and the finite future time domain specifically includes: For each virtual conflict probe point, based on the virtual mass and critical distance of each entity in the established virtual force field model, the attractive force vector from its own target point and the repulsive force vector from all other aircraft and obstacles are calculated. The sum of the above vectors is used as the classical potential force at that point. Based on the obtained predicted trajectory sequence of the neighboring aircraft and the predicted trajectory of the main aircraft from its current position to the probe point, the rate of change of direction of the line connecting the consecutive position points in the relative position sequence of the two is calculated, and a flow field guiding force vector is derived based on the rate of change of direction and the magnitude of the relative motion velocity. The classical potential force vector and the flow field guiding force vector are vector synthesized to obtain the comprehensive virtual force of the virtual collision probe point at the current moment; For multiple equally spaced time points within a finite future time domain, the above calculation process is repeated, and the comprehensive virtual force obtained at each time point is weighted and accumulated according to the time distance to obtain the forward-looking comprehensive virtual force of the probe point that covers the finite future time domain.

7. The method for solving multi-robot collision using virtual force field according to claim 1, wherein, In step S5, the comprehensive virtual force corresponding to all virtual conflict probe points is evaluated. A random perturbation factor is introduced to perturb the comprehensive virtual force, and the probe point with the largest norm of the resultant force vector after perturbation is selected as the current optimal escape point. Specifically, this includes: Obtain the forward-looking comprehensive virtual force vector calculated for each virtual conflict probe point; A random perturbation vector is generated by sampling from a preset random distribution. The random perturbation vector is then vector-synthesized with the corresponding forward-looking integrated virtual force vector to generate the resultant force vector after perturbation. Calculate the Euclidean norm of the resultant force vector after all perturbations, and use it as the evaluation index for each corresponding virtual conflict probe point; Compare the evaluation metrics of all virtual collision probe points, and select the virtual collision probe point with the largest evaluation metric value as the current optimal escape point.

8. The method for solving multi-robot collision using virtual force field according to claim 1, wherein, In step S6, based on the current optimal escape point, a heading angle sequence that satisfies the aircraft dynamics smoothness constraint is calculated, and the first term of this sequence is used as the instantaneous heading command. Specifically, this includes: The target heading angle increment is calculated based on the current position coordinates of the main aircraft and the current optimal escape point coordinates. Based on the target heading angle increment and the heading angle of the main aircraft at the current moment, the desired target heading angle is derived. By utilizing the preset maximum angular velocity limit of the aircraft and the conflict control cycle, the target heading angle is decomposed into a smooth sequence of heading angle commands within multiple consecutive control cycles, which is then output as an instant heading command.

Images