Path planning method for cooperative search of multiple unmanned aerial vehicles

Abstract

A path planning method for multi-unmanned aerial vehicle collaborative search comprises the steps that firstly, an unmanned aerial vehicle kinematics model, a sensor model and an environment cognition map are established, and a target existence probability and environment uncertainty map is updated in real time based on the Bayesian theorem; secondly, constructing a model-based prediction control path decision framework, predicting a future multi-step path through rolling optimization, and generating a control sequence; in the optimization solution, a double-population three-layer competition mechanism is adopted, particle population individuals are divided into a loser, a candidate and a winner, and differential updating strategies are adopted to balance the global exploration and local utilization ability of particles; meanwhile, a feedback mechanism based on particle evolution quality is introduced, the size of the sub-population is dynamically adjusted, and population diversity is maintained in combination with periodic recombination. According to the method, the solving quality of the search path planning problem based on model prediction control is remarkably improved, meanwhile, the high search coverage rate is guaranteed, and the method is suitable for multi-unmanned-aerial-vehicle cooperative search scenes such as disaster rescue and environment monitoring.

Description

Technical Field

[0001] This invention belongs to the field of multi-UAV collaboration and relates to a path planning method for multi-UAV collaborative search. It is a collaborative search path planning method based on model predictive control and is suitable for collaborative search tasks of multiple UAVs in complex environments. Background Technology

[0002] With advancements in aviation technology and artificial intelligence, multi-UAV swarms play a crucial role in search tasks. These systems offer advantages such as high mobility, low cost, and wide coverage, enabling efficient searches in dynamic environments. However, existing multi-UAV path planning methods, within a model predictive control framework, face challenges related to dimensionality expansion and local optima. Traditional algorithms often balance short-term and long-term gains by adjusting the prediction step size, but as the number of UAVs and the prediction step size increase, the optimization dimensionality rises sharply, leading to a decline in solution quality. Furthermore, multi-UAV path planning is prone to getting trapped in local optima, resulting in low search efficiency.

[0003] Furthermore, while existing optimization algorithms such as particle swarm optimization (PSO) or genetic algorithms have applications, they rely on basic evolutionary operations, making them prone to getting trapped in local optima and unable to handle high-dimensional problems. Although these methods possess some optimization capabilities, they neglect group collaboration effects in multi-UAV scenarios, such as slow convergence due to uneven resource allocation. These problems lead to slow search processes, low collaboration efficiency, and an inability to adapt to the demands of complex and dynamic environments. Especially in dynamic and complex environments, traditional methods struggle to form efficient global search and local utilization strategies, necessitating an optimization mechanism based on a feedback-based dual-population three-layer competition to improve algorithm performance. The limitations of existing technologies, including the challenge of dimensional expansion and getting trapped in local optima, highlight the necessity for improving system efficiency. Summary of the Invention

[0004] To address the issue of poor path planning quality and reduced search efficiency caused by high-dimensional optimization in multi-UAV cooperative search, this invention provides a path planning method for multi-UAV cooperative search. It achieves rational path planning and accelerated search through a feedback dual-population three-layer competitive swarm optimization algorithm, providing a more efficient framework for large-scale path planning. This method considers not only individual paths but also the group cooperation effect, improving the system's global optimization performance. The main innovation of this invention lies in using a feedback mechanism as the basis for resource allocation, combined with a three-layer competition strategy, to ensure the balance and diversity of the optimization process, thereby accelerating the search process and improving cooperation efficiency. Furthermore, this invention considers robustness in practical deployment and provides a scalable framework to support the transition from simulation to real-world operation.

[0005] The specific technical solution of this invention is as follows:

[0006] A path planning method for multi-UAV cooperative search includes the following steps:

[0007] Step 1: Environmental Modeling and Uncertainty Map Creation

[0008] Based on the grid division of the search area, a UAV kinematic model, sensor model, and environmental perception map are established. The search value of each grid is quantified from the uncertainty map, and Bayes' theorem is used to update the target existence probability and uncertainty metric to ensure real-time reflection of environmental changes.

[0009] Step 2: Path Decision Framework Based on Model Predictive Control

[0010] Based on the UAV and environment model established in step 1, a model predictive control framework is further adopted to transform the multi-UAV cooperative search problem into a sequence optimization task within a finite time domain. At each decision time, the future control input sequence is predicted, and a cumulative search utility function is constructed as the optimization objective. The search path is updated in real time through rolling solution.

[0011] Step 3: Design of a feedback-based two-population three-layer competitive group optimization algorithm

[0012] To solve the path planning optimization function in step 2 and obtain the UAV control input sequence, this algorithm employs an update strategy different from the traditional particle swarm optimization algorithm. The steps are as follows: First, all particles are divided into two subpopulations. Then, particles within each subpopulation are further divided into three types: winners, candidates, and losers, with differentiated update strategies applied to each: the winner subpopulation focuses on local development, the candidate subpopulation maintains its exploration capabilities, and the loser subpopulation introduces random perturbations to enhance diversity. Through particle evolution quality evaluation, a feedback mechanism is introduced to dynamically adjust the size of each subpopulation, achieving efficient allocation of computational resources. To prevent resource monopolization by any subpopulation leading to premature convergence, a periodic recombination strategy is used to periodically reset the population structure, maintaining overall diversity.

[0013] Step 4: Path Decision and Execution

[0014] The first element of the control input sequence obtained from the algorithm in step 3 is used as the path decision at the current moment to drive the UAV to perform flight maneuvers. At the same time, the environmental perception map is updated based on real-time sensor data, triggering the next round of model predictive control rolling optimization, forming a closed-loop path planning process.

[0015] Furthermore, in step 1, the environmental uncertainty measure is calculated using an entropy function to enhance sensitivity to unknown areas. Map updates are based on a sensor distance attenuation model, adaptively adjusting probability thresholds and optimizing boundaries according to UAV speed and environmental complexity to flexibly respond to dynamic changes. Step 1 provides accurate environmental quantification by fusing location, speed, and detection probability information, supporting large-scale cluster real-time deployment and laying a reliable foundation for subsequent path planning.

[0016] Furthermore, in step 2, a penalty term is introduced for control sequences that violate kinematic constraints or have negative search utility function values ​​within the prediction time domain to suppress the generation of invalid paths, such as decisions that lead to search duplication or inefficiency, thereby improving the coordination efficiency of the group search. This mechanism achieves dynamic adjustment of the global path through constraint handling techniques in the rolling optimization process, promotes the complementarity of multi-UAV cooperative paths, supports the evaluation of enhanced algorithm robustness through prediction sequences in high-dimensional search spaces, and improves the overall search performance of the system.

[0017] Furthermore, in step 3, the subpopulation size is dynamically optimized based on iterative feedback. Initially, a large exploratory population is used to focus on global search, while later the population is reduced to emphasize precise utilization. This design gradually guides the algorithm from diversification to convergence, ensuring fair resource allocation. Simultaneously, a recombination strategy optimizes computational efficiency, supporting real-time applications and multi-task adaptation.

[0018] Furthermore, in step 4, the final path planning output is formed by combining the UAV's motion constraints, obstacle avoidance rules, and the goal of maximizing search utility. This supports verifying the algorithm's effectiveness through actual flight feedback during high-dimensional optimization iterations, ensuring the system's robustness and adaptability in dynamic environments, and improving the overall search performance of the system.

[0019] The beneficial effects of this invention are:

[0020] This invention constructs a refined path planning framework by deeply integrating feedback dual-population three-layer competitive optimization, effectively solving the problems of high-dimensional optimization and local optima, significantly improving the search speed and coverage efficiency of multiple UAVs in complex environments, and has the potential for cross-domain applications.

[0021] This invention is particularly suitable for real-time path optimization frameworks in dynamic environments. It aims to improve the search efficiency and coverage of multi-agent systems through a feedback dual-population three-layer competitive group optimization algorithm. The field of multi-UAV collaboration encompasses the intersection of aerospace, artificial intelligence, and optimization algorithms, focusing on solving the path planning and collaboration problems of multi-UAV systems in uncertain environments. Specifically, this invention proposes a novel path planning mechanism to address the search needs of multi-UAV systems in scenarios such as disaster relief, environmental monitoring, and military reconnaissance. This mechanism can handle increased complexity and local optima problems, improving the overall search performance of the system. Furthermore, this technology can be extended to other fields, such as multi-robot task planning in unmanned vehicle fleet inspection or intelligent logistics, demonstrating broad application potential. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of a multi-drone collaborative search.

[0023] Figure 2 This is a schematic diagram of the sensor coverage for a drone.

[0024] Figure 3 To predict the decision path of the drone with a step size of 3.

[0025] Figure 4 A schematic diagram of the feedback dual-population three-layer competitive group optimization algorithm.

[0026] Figure 5 The image shows the path trajectory of six drones at a search step size of 100. Detailed Implementation

[0027] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0028] This invention proposes a path planning method for multi-UAV cooperative search, which improves search efficiency and path quality through environmental modeling, model predictive control decision-making, and feedback optimization. The specific steps are as follows:

[0029] Step 1: Environmental Modeling and Uncertainty Map Creation

[0030] Multi-drone collaborative search scenarios, such as Figure 1 As shown, the first step is to establish a kinematic model and an environmental model for the UAV, serving as the quantitative basis for path planning. The search area is discretized into a grid map with a size of [missing information]. The UAV state vector is s m ( k ) = [ p m ( k ) , ψ m ( k ) ] ,in The coordinates of the drone's current position. Let be the heading angle. The equation of motion for the UAV is:

[0031] [ x m ( k + 1 ) y m ( k + 1 ) ψ m ( k + 1 ) ] = [ x m ( k ) y m ( k ) ψ m ( k ) ] + [ v m c o s ψ m ( k ) v m s i n ψ m ( k ) u m ( k ) ]

[0032] in, For drone speed, Let m be the current heading angle of the drone. The yaw angle is also the control input for the drone.

[0033] In addition, the sensor models of drones, such as Figure 2 As shown, where This is the maximum detection radius of the sensor. This represents the coverage area of ​​the sensor. The detection probability of the sensor is also included. A distance-related update method is used, and the specific formula is as follows:

[0034]

[0035] in, To maximize the detection probability, The attenuation coefficient is... This represents the distance between the drone and the target.

[0036] Environmental cognition maps include environmental uncertainty maps and target existence probability maps, with the uncertainty map employing information entropy. It is calculated in the form of, and is specifically defined as follows:

[0037]

[0038] in, This represents the probability that the target exists at coordinates (x, y) at time k.

[0039] Target Existence Probability Map The update is based on Bayes' theorem, and the specific process is as follows:

[0040]

[0041] in, Indicates the probability of drone detection. Indicates the false alarm rate. A positive result indicates a target has been detected; otherwise, it indicates no target has been detected. This step provides a reliable environmental perception foundation for subsequent path decision-making by establishing an accurate kinematic model and a real-time updated cognitive map. In practical applications, this model can be adjusted according to the maneuverability parameters of different UAV platforms, such as the difference in turning radius between fixed-wing and rotary-wing UAVs, ensuring the model's versatility and adaptability. Simultaneously, the entropy calculation of the uncertainty map effectively captures the information value of unknown areas, guiding the UAV to prioritize exploring high-uncertainty grids, thereby maximizing search gains. This step lays the foundation for subsequent path planning and supports subsequent optimization and expansion.

[0042] Step 2: Path Decision Framework Based on Model Predictive Control

[0043] In the model predictive control framework, each decision step executes only the first action in the prediction sequence, followed by rolling optimization based on the updated state. This rolling mechanism significantly reduces the computational complexity of a single optimization while ensuring real-time decision-making. In high-dimensional search spaces, model predictive control avoids the overhead of global exhaustive search through finite prediction step sizes, and combined with feedback correction mechanisms, it can effectively cope with sudden environmental changes, such as target movement or sensor noise. The possible paths of the UAV when the prediction step size is 3 are as follows: Figure 3 As shown. Therefore, consider the cooperative search objective function for predicting the step size. The design is as follows:

[0044]

[0045] in, Represents the decision variables for drones. Indicates the current moment. Indicates the prediction step size. Indicates the number of drones. This represents the current sensor coverage area of ​​the drone.

[0046] Step 3: Design of a feedback-based two-population three-layer competitive group optimization algorithm (e.g.) Figure 4 (As shown)

[0047] Unlike the single update iteration method in traditional particle swarm optimization, the proposed algorithm divides the entire population into two subpopulations, and further divides each subpopulation into three layers: losers, candidates, and winners, and adopts different update methods for different particles.

[0048] First, the update formula for the loser particles is:

[0049]

[0050] in, For the speed of the losers, and For subpopulation The positions of the winners and losers. for An internal random number, where a = 1, 2, or 3. The average position of the candidates in the opposing subpopulation. This is the adjustment coefficient.

[0051] Secondly, the update formula for candidate particles is:

[0052]

[0053] in, For the candidate's speed, For subpopulation The position of the candidates It represents the average position of the winner in the opposing subpopulation.

[0054] The final winner particle update formula is:

[0055]

[0056] in, and The winner's speed and position are respectively. , For subpopulation The number of particles From a fitness value better than Randomly selected from the -th particles. The optimal particle position among the winners of the two subpopulations.

[0057] The feedback strategy dynamically adjusts the subpopulation size based on the particle mass before and after the update to improve search efficiency. The updated subpopulation size is determined by the evolutionary quality of the particles.

[0058]

[0059] in, The total number of particles, and These represent the number of particles in the two subpopulations, respectively. and These are the average fitness values ​​of the two subpopulations. and The degree of improvement in the fitness values ​​of particles in the two populations. and The proportion of improved particles in the two populations.

[0060] The constraints are:

[0061]

[0062] The periodic recombination strategy reorders particles every fixed number of generations and evenly divides them into two subpopulations, maintaining diversity. This design gradually guides the shift from exploration to utilization, ensuring fair resource allocation, while optimizing computational efficiency through the recombination strategy to support real-time applications.

[0063] The core of the proposed algorithm lies in the competition and cooperation mechanism between two populations. Losers improve their optimization quality by learning from winners within their own population and opposing candidates; candidates balance information between winners and opposing winners in their own population, promoting information flow between populations; and winners conduct refined searches under the guidance of elite particles. This three-layer structure avoids the single update mode of traditional particle swarm optimization algorithms, significantly improving the algorithm's performance in high-dimensional model prediction and control problems. The feedback mechanism comprehensively evaluates the efficiency of subpopulations based on three dimensions, enabling dynamic adjustment of computational resources. For example, a larger scale is allocated to exploratory subpopulations in the early stages of the search, while later resources are allocated to exploitative subpopulations. Periodic reorganization prevents premature population convergence and ensures global diversity. This algorithm performs exceptionally well when handling high-dimensional optimization variables and is suitable for solving real-time path planning problems involving multiple UAVs.

[0064] Step 4: Path Decision and Execution

[0065] The algorithm proposed in step 3 is used to solve the search path planning problem based on model predictive control, generating the optimal control input sequence within the future step size range. The first element in the sequence As the path decision result at the current decision time k, the m-th UAV is controlled to perform corresponding heading adjustments and translational movements. Simultaneously, based on real-time sensor observation data, the target existence probability map and environmental uncertainty map are updated, incorporating a feedback adjustment mechanism to trigger the rolling optimization process in the next decision cycle k+1, forming a closed-loop path planning and execution flow. In this embodiment, the search area is discretized into 50 × 50 grids within a 5 km × 5 km square search area. There are a total of 6 UAVs, with a prediction step size h=3. The total simulation time corresponds to 100 decision steps. Finally, at the 100th step, the UAV's search trajectory is as follows... Figure 5 As shown.

Claims

1. A path planning method for cooperative search among multiple unmanned aerial vehicles (UAVs), characterized in that, Includes the following steps: Step 1: Environmental Modeling and Uncertainty Map Creation Based on the grid division of the search area, a UAV kinematic model, sensor model, and environmental cognition map are established; the search value of each grid is quantified from the uncertainty map, and the target existence probability and uncertainty measure are updated using Bayes' theorem to ensure real-time reflection of environmental changes; Step 2: Path Decision Framework Based on Model Predictive Control A model predictive control framework is adopted to transform the multi-UAV cooperative search problem into a sequence optimization task within a finite time domain. At each decision time, the future control input sequence is predicted, and a cumulative search utility function is constructed as the optimization objective. The search path is updated in real time through rolling solution. Step 3: Design of a feedback-based two-population three-layer competitive group optimization algorithm To solve the path planning optimization function in step 2 and obtain the control input sequence of the UAV, the following steps are taken: First, all particles are divided into two subpopulations, and then the subpopulations are further divided into three subpopulations: winners, candidates, and losers. Differentiated update strategies are implemented for each subpopulation: the winner subpopulation focuses on local development, the candidate subpopulation maintains its exploration capability, and the loser subpopulation introduces random perturbations to enhance diversity. Through particle evolution quality assessment, a feedback mechanism is introduced to dynamically adjust the size of each subpopulation, achieving efficient allocation of computing resources. To prevent resource monopoly by any subpopulation from leading to premature convergence, a periodic recombination strategy is adopted to periodically reset the population structure and maintain overall diversity. Step 4: Path Decision and Execution The first element of the control input sequence obtained by solving the algorithm in step 3 is used as the path decision at the current moment to drive the UAV to perform flight actions; at the same time, the environmental cognition map is updated based on real-time sensor data, triggering the next round of model predictive control rolling optimization, forming a closed-loop path planning process.

2. The path planning method for multi-UAV cooperative search according to claim 1, characterized in that, In step 1, the environmental uncertainty measure is calculated using an entropy function to enhance sensitivity to unknown areas; the map update is based on a sensor distance attenuation model, adaptively adjusting the probability threshold, optimizing the boundary according to the UAV speed and environmental complexity, and flexibly responding to dynamic changes.

3. The path planning method for multi-UAV cooperative search according to claim 1, characterized in that, In step 2, a penalty term is introduced to suppress invalid path generation for control sequences that violate kinematic constraints or have negative search utility function values ​​in the prediction time domain.

4. The path planning method for multi-UAV cooperative search according to claim 1, characterized in that, In step 3, the subpopulation size is dynamically optimized based on iterative feedback. In the early stage, a large exploratory population is used to focus on global search, and in the later stage, the population is reduced to emphasize precise utilization.

5. The path planning method for multi-UAV cooperative search according to claim 1, characterized in that, In step 4, the final path planning output is formed by combining the UAV's motion constraints, obstacle avoidance rules, and the goal of maximizing search utility; this supports the verification of the algorithm's effectiveness through actual flight feedback during high-dimensional optimization iterations.

6. A path planning method for multi-UAV cooperative search according to any one of claims 1-5, characterized in that, The specific steps are as follows: Step 1: Environmental Modeling and Uncertainty Map Creation First, it is necessary to establish a kinematic model and an environmental model for the UAV as the quantitative basis for path planning; then, the search area is discretized into a grid map with a size of [missing information]. The UAV state vector is s m ( k ) = [ p m ( k ) , ψ m ( k ) ] ,in The coordinates of the drone's current position. Let be the heading angle; the equation of motion for the UAV is: [ x m ( k + 1 ) y m ( k + 1 ) ψ m ( k + 1 ) ] = [ x m ( k ) y m ( k ) ψ m ( k ) ] + [ v m c o s ψ m ( k ) v m s i n ψ m ( k ) u m ( k ) ] ； in, For drone speed, Let m be the current heading angle of the drone. This is the yaw angle, which is also the control input for the drone; For the sensor model of the drone, among which This is the maximum detection radius of the sensor. This represents the coverage area of ​​the sensor; where the sensor's detection probability is... A distance-related update method is used, and the specific formula is as follows: ； in, To maximize the detection probability, The attenuation coefficient is... The distance between the drone and the target; Environmental cognition maps include environmental uncertainty maps and target existence probability maps, with the uncertainty map employing information entropy. It is calculated in the form of, and is specifically defined as follows: ； in, This represents the probability that the target exists at coordinates (x, y) at time k. The probability update is based on Bayes' theorem, and the specific process is as follows: ； in, Indicates the probability of drone detection. Indicates the false alarm rate. This indicates that a target has been detected; otherwise, it indicates that no target has been detected. Step 2: Path Decision Framework Based on Model Predictive Control Under the model predictive control framework, each decision step executes only the first action in the prediction sequence, followed by rolling optimization based on the updated state; a collaborative search objective function considering the prediction step size is used. The design is as follows: ； in, Represents the decision variables for drones. Indicates the current moment. Indicates the prediction step size. Indicates the number of drones. This refers to the current sensor coverage area of ​​the drone; Step 3: Design of a feedback-based two-population three-layer competitive group optimization algorithm The entire population is evenly divided into two subpopulations, and each subpopulation is further divided into three layers: losers, candidates, and winners. Different update methods are used for different particles. First, the update formula for the loser particles is: ； in, For the speed of the losers, and For subpopulation The positions of the winners and losers; for An internal random number, where a = 1, 2, or 3. The average position of the candidates in the opposing subpopulation. This is the adjustment coefficient; Secondly, the update formula for candidate particles is: ； in, For the candidate's speed, For subpopulation The position and speed of the candidates The average position of the winner in the opposing subpopulation; The final winner particle update formula is: ； in, and The winner's speed and position are respectively. , For subpopulation The number of particles; From a fitness value better than Randomly selected from the -th particles. The optimal particle position among the winners of the two-sub population; The feedback strategy dynamically adjusts the subpopulation size based on the particle mass before and after the update to improve search efficiency. The updated subpopulation size is determined by the evolutionary quality of the particles. ； in, The total number of particles, and These represent the number of particles in the two subpopulations, respectively. and These are the average fitness values ​​of the two subpopulations. and The degree of improvement in the fitness values ​​of particles in the two populations. and The proportion of improved particles in the two populations; The constraints are: ； Step 4: Path Decision and Execution The algorithm proposed in step 3 is used to solve the search path planning problem based on model predictive control, generating the optimal control input sequence within the future step size range. The first element in the sequence As the path decision result at the current decision moment k, the m-th UAV is controlled to perform the corresponding heading adjustment and translational movement. At the same time, the target existence probability map and environmental uncertainty map are updated based on real-time sensor observation data, and a feedback adjustment mechanism is incorporated to trigger the rolling optimization process of the next decision cycle k+1, forming a closed-loop path planning and execution process.

Images