Low-altitude unmanned aerial vehicle logistics intelligent path planning and scheduling method

Abstract

The invention relates to the technical field of unmanned aerial vehicle navigation, and discloses a low-altitude unmanned aerial vehicle logistics intelligent path planning and scheduling method, which comprises the following steps: acquiring urban environment space data, then establishing an unmanned aerial vehicle dynamics model, satisfying a given physical constraint condition, constructing a path cost function, and establishing a path cost function based on an optimal control theory. And performing time-space discrete processing on the trajectory optimization problem, and solving by adopting a numerical method to obtain an optimal state trajectory and control input, thereby obtaining a discrete trajectory, generating a continuous control instruction sequence for the unmanned aerial vehicle flight control system, obtaining a control trajectory, and taking the control trajectory as a task scheduling basis. And distributing logistics tasks among the plurality of unmanned aerial vehicles according to a preset priority strategy. A dual-stage task scheduling structure is introduced, an adaptive unmanned aerial vehicle set is screened firstly, and then path cost minimum matching is executed, so that effective compression of a scheduling solution space and remarkable improvement of matching efficiency are realized, and a scheduling effect still having high real-time performance and high feasibility in a multi-task concurrent scene is obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of unmanned aerial vehicle navigation, in particular to a low-altitude unmanned aerial vehicle logistics intelligent path planning and scheduling method. BACKGROUND

[0002] With the rapid development of global e-commerce and the increasing demand for timeliness of logistics distribution by consumers, traditional logistics modes are facing many challenges, such as traffic congestion leading to delayed distribution, rising labor costs, etc. Under this background, low-altitude unmanned aerial vehicle logistics, as an innovative logistics distribution method, gradually shows great potential.

[0003] Unmanned aerial vehicles have the advantages of not being restricted by ground traffic, being able to flexibly traverse complex terrain, etc., and can realize rapid distribution of "the last kilometer" or even "door-to-door", effectively improving logistics efficiency. Especially in remote areas, mountainous areas and areas with poor transportation, unmanned aerial vehicle logistics can fill the gap of traditional logistics distribution and provide more convenient services for users.

[0004] However, to realize the efficient operation of low-altitude unmanned aerial vehicle logistics, intelligent path planning and scheduling are key links. On the one hand, unmanned aerial vehicle flight is restricted by factors such as battery endurance, load capacity, weather conditions, etc. How to plan the shortest and safest flight path while meeting these constraints is a problem to be solved. On the other hand, when there are multiple unmanned aerial vehicles simultaneously performing distribution tasks, reasonable scheduling of unmanned aerial vehicles is essential to avoid flight conflicts and optimize resource allocation, which is crucial to improve the operation efficiency of the entire logistics system. The existing path planning and scheduling methods have defects such as high computational complexity, poor real-time performance, and insufficient adaptability when dealing with complex and variable low-altitude flight environments and large-scale unmanned aerial vehicle collaborative operations, making it difficult to meet the needs of actual logistics distribution. Therefore, it is of great practical significance and application value to develop an intelligent path planning and scheduling method for low-altitude unmanned aerial vehicle logistics. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides an intelligent path planning and scheduling method for low-altitude unmanned aerial vehicle logistics, which solves the problem that the single-stage centralized scheduling mode is generally used in the prior art, all tasks and all available unmanned aerial vehicles are combined and matched without screening in the task allocation process, there is a lack of initial adaptability judgment, which easily leads to an increase in computational dimension, a long solving time, and an unscheduled result that cannot be updated in real time.

[0006] To achieve the above purpose, the present application realizes the following technical scheme: an intelligent path planning and scheduling method for low-altitude unmanned aerial vehicle logistics, comprising the following steps:

[0007] Obtain urban environment spatial data, and construct a cost distribution function including obstacle regions and risk regions based on the urban environment spatial data, which is used to represent flight cost of each position point;

[0008] Then, a UAV dynamics model is established, which includes position, velocity and acceleration state variables, and takes acceleration vector as control input, and satisfies given physical constraint conditions;

[0009] A path cost function is constructed, which is used to comprehensively evaluate control energy consumption, spatial environment risk and obstacle avoidance cost of a flight path;

[0010] Based on optimal control theory, a trajectory optimization model is established, which takes the path cost function as a target and the dynamics model as a constraint, and a flight control law is derived;

[0011] The trajectory optimization problem is processed by time and space discretization, and a numerical method is used to solve the optimal state trajectory and control input, so as to obtain a discrete trajectory;

[0012] The discrete trajectory is interpolated to generate a continuous control instruction sequence for a UAV flight control system, and a control trajectory is obtained;

[0013] The control trajectory is used as a task scheduling basis, and a logistics task is allocated among multiple UAVs according to a preset priority strategy.

[0014] Preferably, the cost distribution function is composed of an obstacle repulsion potential field and a risk potential field, the value of the obstacle repulsion potential field gradually increases when approaching the obstacle region, and the risk potential field is assigned with a corresponding cost weight according to the known high-risk region in the space.

[0015] Preferably, the UAV dynamics model is a four-dimensional state system, including position, velocity and acceleration state variables, and the control input is a two-dimensional acceleration vector, and the model has the following form:

[0016]

[0017] wherein x(t), y(t) are positions, v x (t), v y (t) are velocities, a x (t), a y (t) are control accelerations.

[0018] Preferably, the path cost function is an integral functional on a time interval, and the integral integrand includes a weighted combination of the square norm of the control input, the value of the obstacle potential field function and the value of the risk function, and the path cost function is a continuous functional, and the formula is as follows:

[0019]

[0020] wherein:

[0021] x(t) represents a state trajectory;

[0022] u(t) represents a control trajectory;

[0023] Φ(x(t)) is a potential field cost function;

[0024] R(x(t)) is a risk cost function;

[0025] α1, α2, α3 are objective function weighting coefficients.

[0026] Preferably, the flight control law is obtained by constructing a Hamiltonian function and applying the Pontryagin minimum principle, wherein the Hamiltonian function contains the inner product of the state control dynamics term and the co-state variable of the path cost function.

[0027] Preferably, the space-time discretization adopts a pseudospectral method to construct path nodes, and a high-order orthogonal polynomial node is used to construct a discrete model, which is converted into a nonlinear constraint optimization problem for solving.

[0028] Preferably, the continuous control instruction sequence is generated by polynomial interpolation or spline interpolation, and the interpolation result has first-order and above derivative continuity within the flight time interval and satisfies the unmanned aerial vehicle dynamics constraint condition.

[0029] Preferably, the task scheduling is based on the path cost function value, the distance information between the flight start and end points, and the logistics task priority information, and the objective function of minimizing the overall scheduling cost is used to allocate tasks to multiple unmanned aerial vehicles.

[0030] Preferably, the flight control trajectory is updated in real time, and when the environmental cost function changes or the task demand is adjusted, the optimizer is quickly solved to reconstruct the trajectory for dynamic response.

[0031] Preferably, the task scheduling algorithm adopts a two-stage structure, the first stage selects an adaptive unmanned aerial vehicle set, and the second stage performs optimal matching based on the path cost to satisfy the time window constraint on the basis of the global cost minimum.

[0032] The present application provides a low-altitude unmanned aerial vehicle logistics intelligent path planning and scheduling method.

[0033] 1、The present application introduces a two-stage task scheduling structure, selects an adaptive unmanned aerial vehicle set first, and then performs path cost minimum matching, effectively compresses the scheduling solution space, significantly improves the matching efficiency, and obtains a scheduling result with high real-time performance and high feasibility in a multi-task concurrent scenario.

[0034] 2、The present application realizes the dynamic response of the path to the sudden environmental change and the task adjustment by constructing the environment perception type path cost function and reconstructing the flight trajectory in real time, and obtains the high robustness effect that the flight control is continuously stable and is not interrupted by the sudden situation.

[0035] 3、The present application realizes the multi-factor trade-off decision of the task allocation by fusing the path cost, the time window deviation and the task priority in the scheduling target function, and obtains the comprehensive optimization effect that the high-priority task is responded in time and the system overall resource is efficiently utilized.

[0036] 4、The present application realizes the feedback linkage of the real-time update of the scheduling result to the control layer trajectory by the closed-loop coupling design of the path optimization module and the task scheduling module, and obtains the system operation effect that the whole process of task execution is highly coordinated and the self-adaptive ability is strong. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 The method flow chart of the low-altitude unmanned aerial vehicle logistics intelligent path planning and scheduling method of the present application. DETAILED DESCRIPTION

[0038] The technical solutions of the present application will be described clearly and completely below with reference to the drawings of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0039] Please refer to the drawings of the present application Figure 1 The embodiment of the present application provides a low-altitude unmanned aerial vehicle logistics intelligent path planning and scheduling method, which comprises the following steps:

[0040] Obtain the urban environment space data, construct the cost distribution function containing the obstacle area and the risk area based on the urban environment space data, and use it to represent the flight cost of each position point;

[0041] Then, the unmanned aerial vehicle dynamics model is established, which includes the position, velocity and acceleration state variables, and uses the acceleration vector as the control input to meet the given physical constraint condition;

[0042] The path cost function is constructed to comprehensively evaluate the control energy consumption, the spatial environment risk and the obstacle avoidance cost of the flight path;

[0043] Based on the optimal control theory, the trajectory optimization model is established with the path cost function as the target and the dynamics model as the constraint, and the flight control law is derived;

[0044] The trajectory optimization problem is discretized in time and space, and the optimal state trajectory and control input are obtained by numerical methods, thus obtaining the discrete trajectory.

[0045] The discrete trajectory is interpolated to generate a continuous control command sequence for the UAV flight control system, thus obtaining the control trajectory.

[0046] The control trajectory is used as the basis for task scheduling, and logistics tasks are allocated among multiple drones according to a preset priority strategy.

[0047] The cost distribution function consists of an obstacle repulsion potential field and a risk potential field. The value of the obstacle repulsion potential field gradually increases when it is close to the obstacle, and the risk potential field assigns corresponding cost weights based on the known high-risk areas in the space.

[0048] Specifically, by introducing a cost distribution function, the urban space is mapped into a continuous cost field, thus providing basic environmental information support for path generation and optimization. The cost distribution function consists of two core sub-functions: an obstacle repulsion potential field and a risk potential field. These two potential fields exert influence on the path cost from different dimensions, forming a synthetic spatial potential energy structure for use by the path planning module. The obstacle repulsion potential field is used to simulate the influence of obstacles on the flight path. As the UAV path approaches the obstacle boundary, the value of this potential field gradually increases, thereby producing a "repulsion" effect on obstacles during the path optimization process.

[0049] The obstacle repulsion potential field Φ(x) can be constructed based on the distances to the obstacle boundaries. Let any point in space be x = (x, y), and the set of obstacles be... use Let x represent the minimum Euclidean distance between any point x in space and the set of obstacles. Then we have:

[0050]

[0051] Where, ||xo||2 is the Euclidean distance (2-norm) from point x to a point o in the set of obstacles; To find the distance from x to the nearest point among all obstacle boundary points, let's denot it as...

[0052] The barrier repulsive potential field can take the following form:

[0053]

[0054] in:

[0055] d0 represents the potential field's influence threshold, and k is the obstacle strength adjustment coefficient, used to control the amplitude of the obstacle's "repulsive force." The region near the obstacle edge experiences this force. The potential field value will rise sharply, and far away from the obstacle Φ(x)→0, this feature makes the path naturally away from the dense area of obstacles. To avoid the function in the numerical calculation of singular values, the exponential decay processing can be carried out:

[0056]

[0057] Where, α i and β i are the weight coefficient and influence strength related to the obstacle; o i is the coordinate of the ith obstacle;‖x-o i ‖ 2 is the square of the Euclidean distance from point x to obstacle o i .

[0058] The mechanism of risk potential field R(x) is different from the obstacle repulsion potential field. Its core goal is to reflect the information of high-risk areas in space preset by humans. These areas include but are not limited to airport no-fly zones, areas prone to disturb the public, accident-prone areas, or above sensitive facilities.

[0059] The risk potential field is modeled as a regional cost function with discrete cost levels. There are M high-risk areas R j in space, then:

[0060]

[0061] Where:

[0062] r j is the cost value corresponding to the jth risk area;

[0063] χ j (x) is an indicator function, which is 1 if x∈R j , otherwise 0.

[0064] To improve continuity and differentiability, the above indicator function is fuzzified and approximated using a smooth function, for example:

[0065]

[0066] Where:

[0067] c j is the center point of the region;

[0068] d j is the radius of the risk area;

[0069] κ j controls the transition slope.

[0070] Through this continuous approximation method, the risk potential field exhibits smooth transition boundaries in space, not only preserving the influence of risk areas, but also making the overall cost function have good numerical characteristics.

[0071] As an actual application example, in a certain logistics scenario, the airspace above the city center hospital is designated as a first-level high-risk area, with a cost setting of r1=100, while the airspace above the industrial park is set as a second-level area, with a cost setting of r2=50. The flight controller will preferentially plan to avoid the r1 area, while trying to avoid r2 within an acceptable cost range.

[0072] Finally, the complete cost distribution function can be represented as:

[0073] C(x) = Φ(x) + R(x)

[0074] This function is embedded in the optimization objective in the form of an integral as a path cost item in the path optimization model, used to constrain path selection behavior. The derivability of its mathematical form and the fine level of spatial expression provide analytical support for the path planning solver constructed by the invention. Through this combined modeling method, the cost distribution function not only has realistic interpretability, but also exhibits excellent stability and convergence characteristics in numerical calculation.

[0075] The unmanned aerial vehicle dynamics model is a four-dimensional state system, with four-dimensional state variables (x, y, v x , v y ), and a two-dimensional acceleration vector as control input, and has the following form:

[0076]

[0077] where x(t), y(t) are positions, v x (t), v y (t) are velocities, and a x (t), a y (t) are control accelerations.

[0078] Specifically, the unmanned aerial vehicle dynamics model is set as a four-dimensional state system, including two-dimensional position state x(t), y(t) and two-dimensional velocity state v x (t), v y (t), and the control input is a two-dimensional acceleration vector a x (t), a y (t).

[0079] The mathematical expression of this model in the time continuous domain is as follows:

[0080]

[0081] Wherein: x(t), y(t): represent the two-dimensional position of the unmanned aerial vehicle at time t;

[0082] v x (t), v y (t): represent the speed of the unmanned aerial vehicle in the horizontal direction and the vertical direction;

[0083] a x (t), a y (t): control input, respectively, represent the acceleration in two directions.

[0084] The state vector and the control vector can be expressed as:

[0085]

[0086] The system state equation can be simplified to a linear form:

[0087]

[0088] Wherein, the matrix A and B are respectively:

[0089]

[0090] The structure of this linear time-invariant system provides modeling convenience for subsequent trajectory optimization and controller design, and can also adapt to the trajectory tracking structure of most flight control platforms;

[0091] As an option, in the context where speed saturation or acceleration limit needs to be considered, the constraint condition: ‖v(t)‖≤v max ,‖u(t)‖≤a max can be added to the model.

[0092] The above constraint can effectively avoid generating physically unachievable control input and improve the feasibility of trajectory execution. If the unmanned aerial vehicle has angle control or yaw attitude control, it can also be extended to a six-dimensional state system based on the original four-dimensional system, and the heading angle θ(t) and angular velocity ω(t) are introduced. However, considering that most short-range point-to-point logistics tasks only need to consider two-dimensional translational control, the present application adopts a four-dimensional modeling method, taking into account simplicity and practicality. The dynamic model is used as a hard constraint condition for trajectory optimization to ensure that the optimization result meets the physical execution feasibility within the entire time interval. In the subsequent path cost function construction and optimal control derivation steps, the model will run through the whole process as the system dynamic basis of the path generation algorithm.

[0093] The path cost function is an integral functional on the time interval, and the integral term contains the weighted combination of the square norm of the control input, the obstacle potential function value and the risk function value. The path cost function is a continuous functional, and the formula is as follows:

[0094]

[0095] wherein:

[0096] x(t) represents a state trajectory;

[0097] u(t) represents a control trajectory;

[0098] Φ(x(t)) is a potential field cost function;

[0099] R(x(t)) is a risk cost function;

[0100] α1, α2, α3 are objective function weighting coefficients.

[0101] Specifically, after completing the above path cost function construction considering environmental factors and dynamic modeling, the application further constructs a path cost function to comprehensively measure the overall cost of the flight trajectory in terms of control energy consumption, space obstacle avoidance, and risk area avoidance. The path cost function is the core objective function of the subsequent trajectory optimization model and plays a decisive role in generating the final flight path.

[0102] Generally, to achieve smooth control, obstacle avoidance, risk avoidance, and task adaptability, the path cost function should have the following characteristics: good derivability, clear physical meaning of cost items, adjustable weighting structure, and strong constraint driving ability.

[0103]

[0104] The path cost function is a continuous function with good mathematical properties. Especially under the condition that the state and control input are continuous and derivable functions, the functional can realize efficient iteration of first or second order gradient method in the optimization solver. As an option, the control input term adopts the square form of Euclidean norm:

[0105] The main role of its introduction is to suppress excessive control pulses and improve the energy efficiency and dynamic smoothness of the path. Specifically, the obstacle term Φ(x(t)) rapidly rises when close to the obstacle boundary, and its gradient direction is towards the direction away from the obstacle, making the path optimization process automatically avoid the obstacle region. The risk term R(x(t)) provides an avoidance driving force for artificially delineated areas such as no-fly zones or restricted air routes. By adjusting the proportional coefficient relationship of α1: α2: α3, different weights can be set according to the flight task requirements.

[0106] If the task emphasizes safety, let α2, α3 be much larger than α1, so that the path is significantly away from the high-risk area.

[0107] If the task emphasizes flight efficiency, appropriately increase a1, and reduce the path distortion caused by unnecessary avoidance areas;

[0108] If there are multiple target task requirements in the area, a multi-point cost term can be introduced in the integral function, but this extension does not affect the main structure of the current cost function. The path cost function can also introduce a terminal state penalty term, for example:

[0109]

[0110] Where x goal is the target point state, which is used to enhance the end point approach accuracy, especially suitable for high precision landing control scenarios. The path cost function constructed finally is used as the objective function of the optimization problem, and together with the system dynamics model, it constitutes the mathematical basis of the constrained optimization problem, providing theoretical support and calculation basis for the subsequent trajectory generation and control scheduling of the invention.

[0111] The flight control law is obtained by constructing the Hamilton function and applying the Pontryagin minimum principle, wherein the Hamilton function contains the inner product of the costate variable of the path cost function and the state control dynamics term.

[0112] Specifically, after the path cost function is constructed and combined with the dynamics modeling, in order to realize the real-time calculation of the optimal control input, a control law solving method based on the Pontryagin minimum principle is introduced. By constructing the Hamilton function and solving the necessary condition system, the functional relationship between the state trajectory and the control trajectory can be derived, so as to complete the analytical expression or numerical construction of the control law;

[0113] Generally, the Pontryagin minimum principle is suitable for the optimal problem of the system controlled by differential constraints, and is suitable for the continuous time trajectory optimization problem proposed in the invention. The core idea is to make the objective functional (i.e. the path cost function) take the minimum value under the constraint condition of the state evolution equation;

[0114] In this embodiment, the state variable constructed is:

[0115] The control variable is:

[0116]

[0117] The system dynamics relationship is:

[0118]

[0119] The path cost function is:

[0120]

[0121] On the basis of the above, the costate variable is introduced Construct the Hamiltonian function H:

[0122]

[0123] Where:

[0124] α1‖u(t)‖ 2 : control energy term;

[0125] Φ(x(t)): environmental cost function (such as obstacles / risk);

[0126] R(x(t)): other cost terms (such as route deviation, etc.);

[0127] Costate variable (dual variable);

[0128] Ax(t) + Bu(t): system dynamics constraints;

[0129] Where the first half is the cost term, and the second half is the inner product of the state constraint term and the costate variable, which reflects the sensitivity of the state evolution to the target value.

[0130] According to the requirements of the Pontryagin minimum principle, the optimal control law must satisfy the following necessary conditions:

[0131] State equation:

[0132]

[0133] Costate equation (negative value of the derivative of Hamiltonian function with respect to state):

[0134]

[0135] Optimality condition (make H minimum with respect to u(t)):

[0136]

[0137] Further, the explicit relationship between the control input and the costate can be derived:

[0138]

[0139] This analytical expression provides a formal control law for the flight controller, which can directly generate a control input sequence under the premise that the costate trajectory is known. The acquisition of the costate trajectory requires simultaneous solution with the state trajectory, i.e., forming a two-point boundary value problem;

[0140] Considering that the above continuous control law is difficult to be directly used for embedded implementation, a discrete pseudospectral method can also be used to discretize the state and co-state system, a nonlinear optimizer can be combined for numerical solution, and a model predictive control idea can be used to update the optimal control input at the current time at each time step, which has higher robustness and real-time performance in actual deployment.

[0141] The control law derivation process provides strong theoretical support for path generation and flight control execution of the application, ensures that the generated trajectory meets dynamic constraints while having optimization, and enables the path control to have certain adaptive ability and real-time adjustment space.

[0142] Space-time discretization uses a pseudospectral method to construct path nodes, constructs a discrete model through high-order orthogonal polynomial nodes, and converts it into a nonlinear constraint optimization problem for solution.

[0143] Specifically, in the actual solution process, in order to realize numerical optimization and deployability, the continuous optimization problem needs to be discretized in space and time to form a solvable finite-dimensional optimization model.

[0144] Generally, the conventional equidistant dispersion method has problems such as unstable solution accuracy and poor convergence when facing high-dimensional systems, especially in curved regions of the path, numerical divergence or discontinuous state transition often occurs.

[0145] Therefore, in the embodiment, a pseudospectral method is used for joint space-time discretization, and a non-uniform node distribution is constructed through high-order orthogonal polynomials to effectively improve numerical stability and global approximation ability.

[0146] The entire time domain t∈[0,T] is mapped to the standard interval τ∈[-1,1], and N+1 Legendre Gauss Lobatto (LGL) nodes are selected on the interval:

[0147] τ0=-1<τ1<…<τ N =1

[0148] On these nodes, the state variable x(t) and the control input u(t) are interpolated and modeled. As an option, Lagrange polynomial basis functions are used to construct the state trajectory:

[0149]

[0150] Where x i is the state value at the i-th node, L i (τ) is the i-th Lagrange interpolation basis function, which satisfies L i (τ j )=δ ijTo construct the dynamic constraints, the differential matrix D ij is introduced, which represents the derivative approximation of the polynomial at each node, i.e.,

[0151]

[0152] Combining the time mapping relationship , the expression of the true time and the cycle number can be obtained:

[0153]

[0154] Thus, the continuous-time dynamic equation is converted into the difference constraints between nodes, i.e.,

[0155]

[0156] where f(·) is the system dynamic function term, representing the right side expression of . Similarly, the path cost function is also reconstructed on the nodes by numerical integration. For example, the weighted coefficients w j under the LGL collocation points can be used for weighted summation:

[0157]

[0158] Finally, after unifying the above state dynamic constraints, path cost terms, and boundary conditions (such as the starting and ending state constraints), a nonlinear constraint optimization problem is constructed;

[0159] Optimization variables: state sequence {x i} at the nodes, control sequence {u i};

[0160] Optimization objective: weighted summation form of the weighted cost function;

[0161] Constraint conditions: node dynamic difference constraints + starting and ending state constraints + control / velocity / acceleration, etc. physical amplitude conditions;

[0162] To ensure the smoothness of the trajectory, the terminal state is set as a soft constraint or an additional penalty term is introduced into the cost function, so as to avoid the discontinuous jump phenomenon at the tail of the trajectory;

[0163] In addition, the high-order global approximation property of the pseudospectral method makes it still maintain high precision even in the case of a small number of nodes, which is particularly significant in real-time embedded deployment;

[0164] The final state trajectory and control trajectory not only satisfy the system dynamics and physical constraints, but also have high numerical precision and path smoothness, which can be directly transmitted to the flight controller for trajectory tracking execution.

[0165] The continuous control instruction sequence is generated by polynomial interpolation or spline interpolation, the interpolation result has first-order and above derivative continuity in the time-of-flight interval, and meets the unmanned aerial vehicle dynamics constraint conditions.

[0166] Specifically, the trajectory generated by the pseudospectral method is only defined on a set of sparse discrete time nodes. Directly using these discrete points as control inputs will cause problems such as time discontinuity, control mutation or trajectory jump during execution, especially in the scenario where the dynamic response time of the unmanned aerial vehicle is short. Such discontinuous control input may cause flight instability or even attitude collapse, therefore, in the embodiment, the method of polynomial interpolation or spline interpolation is used to convert the discrete control input sequence {u i} output by the optimizer into a continuous control function u(t) defined on the entire flight interval t∈[0,T]. The function is not only mathematically continuous and derivable, but also needs to meet the structure requirements and constraint conditions of the original dynamics model. A cubic spline is used to perform piecewise interpolation on each control variable dimension, and the interpolation function is as follows:

[0167]

[0168] Among them:

[0169] t i is the time corresponding to each pseudospectral node;

[0170] u i is the node control input value;

[0171] Splinelnterp represents the spline interpolation operation, and the generated result is a cubic polynomial in each sub-interval;

[0172] This method can ensure that the generated control function has at least first-order derivative continuity C 1 in the entire time interval, accurately passes through the original optimization result at the interpolation nodes, and has good fitting and local adjustment ability. As an option, when interpolating the state sequence {x i}, a quintic polynomial interpolation form can also be selected to obtain higher-order continuity including C 2 or C 3 , which is suitable for scenarios that require smoother acceleration or even jerk control;

[0173] Specifically, each dimensional state component or control input can be constructed as follows:

[0174] u k (t)=a0+a1t+a2t 2 +a3t 3 +…;

[0175] The interpolation coefficients a jThe node value and its derivative boundary condition can be determined, and the interpolation curve can be generated by least square fitting, Hermite interpolation or B-spline, and the method can be selected according to the deployment of hardware resources and path requirements.

[0176] To ensure that the interpolation curve strictly satisfies the unmanned aerial vehicle dynamics constraint conditions including speed limit and acceleration constraint in the whole trajectory interval, a soft constraint penalty term can be introduced in the interpolation solving process or a projection interpolation method can be used to correct the result, and in addition, to prevent the interpolation curve from deviating from the optimized trajectory and causing precision loss, intermediate nodes including Chebyshev points or non-uniform adaptive points can be inserted between control points to improve the interpolation precision, and finally the continuous control trajectory u(t) can be directly used as the input flow of the flight control system, and the unmanned aerial vehicle can be driven to fly along the optimized path in real time through a low-delay digital controller or a PWM speed regulator. The whole control process not only retains the global optimality of the pseudospectral method, but also combines the local continuity advantage of the interpolation method, and a good balance between control precision and flight feasibility is achieved.

[0177] The task scheduling is based on the path cost function value, the distance information between the flight start and end points and the logistics task priority information, and the task allocation is performed on the multiple unmanned aerial vehicles by minimizing the objective function of the overall scheduling cost.

[0178] Specifically, the logistics tasks in the urban environment have obvious heterogeneity. For example, the spatial distance between the starting point and the target point is different, the task priority is inconsistent, and the flight cost is different due to the spatial risk distribution and the path congestion degree. Therefore, the scheduling mode based on only the task order or distance cannot effectively meet the global optimal goal of the system.

[0179] Therefore, a comprehensive scheduling objective function is constructed in the embodiment, multiple factors are considered, and the optimal allocation of the multiple unmanned aerial vehicle tasks is realized by minimizing the overall scheduling cost.

[0180] The scheduling input information includes the following three types:

[0181] The path cost function value J ij represents the complete path cost value corresponding to the i-th unmanned aerial vehicle performing the j-th task, which is derived from the output of the path optimization module of the application;

[0182] The distance weight D ij represents the Euclidean distance between the current position of the unmanned aerial vehicle and the starting point of the task, which is used as a priori estimation of the execution cost;

[0183] The task priority weight P j is set by the task submission and represents the importance, urgency or time sensitivity of the task, and a scheduling cost function C ij: C ij = λ1J ij + λ2D ij - λ3P j wherein:

[0184] λ1, λ2, λ3 are weighting coefficients;

[0185] path cost term J ij represents higher consumption cost;

[0186] distance term D ij represents higher flight preparation cost;

[0187] priority term P j represents higher task importance, thus the negative sign is introduced to promote priority matching.

[0188] The optimization goal of task scheduling is to minimize the total scheduling cost of all assignment combinations:

[0189]

[0190] The constraints include:

[0191] Each UAV can be assigned to at most one task (or upper limit according to capability);

[0192] Each task must be assigned to exactly one UAV;

[0193] Optional: additional constraints such as UAV remaining battery, task time window limit, task area limit, etc.

[0194] wherein x ij = 1 means assigning the jth task to the ith UAV, x ij = 0 means not assigning;

[0195] This problem form is equivalent to the weighted bipartite graph matching problem or mixed integer linear programming problem, and in small and medium-sized task scheduling scenarios, ready solvers such as Gurobi, CPLEX can be directly used for solving; to improve the scheduling real-time performance and system flexibility, heuristic algorithms or distributed scheduling strategies can also be used, for example, based on distributed auction or multi-agent game model, to realize local optimal but globally approximately optimal task dynamic assignment, in the heuristic version, all tasks can be sorted by priority first, then match the UAV with the minimum cost for each task, and the subsequent tasks exclude the assigned UAVs;

[0196] In addition, to deal with the uncertainty of task arrival, a rolling horizon optimization is adopted to dynamically update the scheduling matching relationship within a certain time interval, ensuring that new tasks can be inserted into the allocation in the future window while maintaining the stability of the execution of existing tasks. The final scheduling result is passed into the path optimization module to reconstruct the path and update the control command, thereby realizing the linkage control mechanism from the task level to the trajectory level.

[0197] The flight control trajectory is updated in real time. When the environmental cost function changes or the task demand is adjusted, the trajectory is reconstructed by a fast solver for dynamic response.

[0198] Specifically, the traditional offline planning mode generates a path only once at the initial stage of the task, and lacks the ability to adapt to environmental changes. The path control mechanism proposed in the present application has the ability of online trajectory reconstruction, and the current state and the latest environmental information are incrementally solved by a fast optimization module to complete path update without interrupting execution.

[0199] In the present embodiment, the flight controller periodically monitors the following state change sources:

[0200] The spatial distribution of the environmental cost function C(x) changes, including the addition of obstacles and the movement of dynamic risk areas;

[0201] The task instruction between the current position of the unmanned aerial vehicle and the target point is changed, such as target position switching, time window adjustment, deviation during path execution, or exceeding the acceptable trajectory error range;

[0202] The task reassignment signal or interruption command issued by the upper scheduling system.

[0203] When any of the above events occurs, the controller will trigger the path reconstruction process, and the steps are as follows:

[0204] First, update the current starting point state: obtain the system state vector at the current time as the starting state of path reconstruction;

[0205] Second, call the optimizer, with x curr as the initial state, the updated target state x goal as the end condition, and the latest cost distribution function C'(x) = Φ'(x) + R'(x) to reconstruct the path optimization problem;

[0206] Third, real-time solving is performed using a fast path optimization algorithm. As an option, a pre-configured template based on the pseudospectral method can be used to incrementally solve the trajectory with a lower dimension node, significantly reducing the calculation time.

[0207] Fourthly, the new optimal control input trajectory u * (t) is interpolated into a continuous trajectory sequence and is input into the flight control system to replace and execute. * (t) is interpolated into a continuous trajectory sequence and is input into the flight control system to replace and execute.

[0208] The path updating module sets a maximum calculation time threshold (such as 200 ms), if the upper limit cannot be solved, the current trajectory is retained or a backup strategy (such as the shortest obstacle avoidance detour) is used to ensure control continuity. In order to speed up the solving speed, the path optimization model can also be pre-trained in the system with a higher scheduling frequency, for example, a neural network is used to fit the trajectory mapping, so that the initial guess is output through the neural network in the trajectory reconstruction stage, the convergence speed of the optimizer is improved, and the path update can also be embedded in the model predictive control structure, with a limited prediction window for trajectory rolling update, only the trajectory in a short period of time is optimized in each period, and the next period is rolling prediction, forming a closed loop correction. Under the conditions of road construction, temporary gathering area, sudden no-fly instruction and other urban dynamic conditions, the system can make a quick response at the trajectory level, so as to realize the synchronous guarantee of flight safety, control stability and task completion rate.

[0209] The task scheduling algorithm adopts a two-stage structure, the first stage selects an adaptive unmanned aerial vehicle set, and the second stage performs optimal matching based on path cost, which is used to schedule the result to meet the time window constraint on the basis of the minimum global cost.

[0210] Specifically, to cope with the efficient scheduling demand of multiple unmanned aerial vehicles in complex urban logistics task scenarios, the two-stage scheduling structure is introduced in the task allocation module. The structure design aims to improve the scheduling efficiency, reduce the combination complexity, and guarantee the feasibility and global superiority of the scheduling result under the multiple constraints of path cost, task time window and flight resources. Unlike the traditional single assignment strategy, the structure divides the scheduling process into two logical stages, coarse screening and fine matching, to realize the step-by-step convergence process from schedulability judgment to cost-optimal allocation.

[0211] In this embodiment, the scheduling process includes the following two stages:

[0212] The first stage: the goal of this stage is to eliminate individuals that do not have the ability to execute the current task from the entire unmanned aerial vehicle set, thereby reducing the subsequent optimization scale and improving the overall scheduling calculation efficiency.

[0213] Generally, the adaptability judgment is based on the following indicators:

[0214] Position proximity: the distance D between the current position of the unmanned aerial vehicle and the starting point of the task ij Whether it is within the preset threshold d max ;

[0215] Flight accessibility: whether the current remaining power or endurance is sufficient to complete the whole flight of the task;

[0216] Load limit: whether the weight of the scheduled task item matches the current load capacity of the UAV;

[0217] Feasible time window overlap: whether the time window required by the current task intersects with the estimated arrival window of the UAV;

[0218] Control conflict avoidance: whether the UAV has been allocated with other non-interruptible tasks in the current scheduling period, assuming that task T j is the set of all UAVs, and the filtered adaptive set is:

[0219]

[0220] If , the task enters the suspension queue or is processed by the backup aircraft pool.

[0221] Second stage: after completing the adaptive filtering, enter the scheduling core stage: select an individual for each task from the candidate UAVs with the optimal path cost and meeting the time window requirements;

[0222] First, calculate the comprehensive scheduling cost C ij for each candidate combination (i, j), including:

[0223] Optimal path cost J ij (from the current location to the task starting point and then to the target point);

[0224] Any buffer waiting cost (if arriving in advance needs to hover);

[0225] Time window deviation penalty term, such as early / late arrival cost;

[0226] UAV priority scheduling weight (used to adjust critical node resource occupation);

[0227] The matching objective function is as follows:

[0228]

[0229] Where:

[0230] x ij represents whether task j is assigned to UAV i;

[0231] Δt ij represents the deviation value of the actual arrival time from the center of the time window;

[0232] ρ is the time penalty weight coefficient.

[0233] To meet the rigid constraints of time windows, the following restrictions are added to the constraint set:

[0234]

[0235] wherein is the estimated task arrival time after the UAV i performs the task j;

[0236] The matching solution can be solved using the assignment problem framework, or converted into an integer linear programming model, and the optimizer or heuristic algorithm is called by the scheduler to quickly solve it. To avoid overloading the centralized scheduler, the two-stage process can also be modularized and decentralized to realize an edge distributed scheduling structure, that is, each station or cluster independently performs the first stage screening, and the center coordinates to perform the second stage optimization, improving the system scalability.

[0237] Finally, the scheduler issues the matching results to the path planning module, which solves the optimal trajectory for each assigned path and sends it to the flight control system, completing the full-process closed loop of scheduling and control.

[0238] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A low-altitude unmanned aerial vehicle logistics intelligent path planning and scheduling method, characterized in that, Includes the following steps: Acquire urban environmental spatial data, and construct a cost distribution function that includes obstacle areas and risk areas based on the urban environmental spatial data to represent the flight cost at each location point; Then, a UAV dynamics model is established, which includes position, velocity and acceleration state variables, and uses the acceleration vector as the control input to satisfy the given physical constraints. Construct a path cost function to comprehensively evaluate the control energy consumption, space environment risks, and obstacle avoidance costs of the flight path; Based on optimal control theory, a trajectory optimization model with path cost function as objective and dynamic model as constraint is established, and the flight control law is derived. The trajectory optimization problem is discretized in time and space, and the optimal state trajectory and control input are obtained by numerical methods, thus obtaining the discrete trajectory. The discrete trajectory is interpolated to generate a continuous control command sequence for the UAV flight control system, thus obtaining the control trajectory. The control trajectory is used as the basis for task scheduling, and logistics tasks are allocated among multiple drones according to a preset priority strategy. 2.The low-altitude unmanned aerial vehicle logistics intelligent path planning and scheduling method according to claim 1, characterized in that: The cost distribution function consists of an obstacle repulsion potential field and a risk potential field. The value of the obstacle repulsion potential field gradually increases when it approaches the obstacle, and the risk potential field assigns corresponding cost weights based on known high-risk areas in space. 3.The low-altitude unmanned aerial vehicle logistics intelligent path planning and scheduling method according to claim 1, characterized in that: The UAV dynamics model is a four-dimensional state system, including position, velocity, and acceleration state variables. The control input is a two-dimensional acceleration vector. The model has the following form: Where x(t) and y(t) are positions, v x (t),v y (t) represents velocity, a x (t),a y (t) represents the control acceleration. 4.The low-altitude unmanned vehicle logistics intelligent path planning and scheduling method according to claim 1, characterized in that: The path cost function is an integral functional over the time interval, and its integrand includes a weighted combination of the square norm of the control input, the obstacle potential field function value, and the risk function value. The path cost function is a continuous functional, and the formula is as follows: in: x(t) represents the state trajectory; u(t) represents the control trajectory; φ(x(t)) is the potential cost function; R(x(t)) is the risk cost function; α1, α2, α3 are the weighting coefficients of the objective function. 5.The low-altitude unmanned vehicle logistics intelligent path planning and scheduling method according to claim 1, characterized in that: The flight control law is obtained by constructing a Hamiltonian function and applying the Pontryagin minimum principle, wherein the Hamiltonian function contains the inner product of the path cost function and the state control dynamics term with co-state variables. 6.The low-altitude unmanned vehicle logistics intelligent path planning and scheduling method according to claim 1, characterized in that: The spatial-temporal discretization uses a pseudospectral method to construct path nodes, constructs a discrete model through high-order orthogonal polynomial nodes, and transforms it into a nonlinear constrained optimization problem for solution. 7.The low-altitude unmanned vehicle logistics intelligent path planning and scheduling method according to claim 1, characterized in that: The continuous control command sequence is generated by polynomial interpolation or spline interpolation. The interpolation result has first-order or higher derivative continuity within the flight time interval and satisfies the UAV dynamics constraints. 8.The low-altitude unmanned vehicle logistics intelligent path planning and scheduling method according to claim 1, characterized in that: The task scheduling is based on the path cost function value, the distance information between the flight start and end points, and the priority information of the logistics tasks. Multiple drones are assigned tasks by minimizing the overall scheduling cost. 9.The low-altitude unmanned vehicle logistics intelligent path planning and scheduling method according to claim 1, characterized in that: The flight control trajectory is updated in real time. When the environmental cost function changes or the mission requirements are adjusted, the trajectory is reconstructed by a fast-solving optimizer for dynamic response. 10.The low-altitude unmanned vehicle logistics intelligent path planning and scheduling method according to claim 1, characterized in that: The task scheduling algorithm adopts a two-stage structure, the first stage screens the adaptive unmanned aerial vehicle set, and the second stage performs optimal matching based on path cost, so that the scheduling result meets the time window constraint on the basis of the minimum global cost.

Images