Unmanned aerial vehicle path planning method and system based on gradient descent method

Abstract

The application belongs to the technical field of path planning, and provides a UAV path planning method and system based on gradient descent method, which is based on a global initial path, performs B-spline curve fitting to obtain initial control points of B-spline function, introduces a speed adjustment factor, takes the control points as decision variables, uses the gradient descent method to obtain an optimal path, uses the geometric position relationship of the control points to approximately represent the radian of the curve, so that the speed is easy to control and a large amount of calculation burden is not increased, through the speed adjustment factor, the speed dynamic adjustment suitable for the radian of the trajectory is realized, the phenomenon of sharp turning is avoided, and the planning safety is improved; meanwhile, collision detection is performed in the optimal path, if the number of collisions is greater than a preset number, a temporary intermediate target point is selected near the obstacle, the temporary intermediate target point is replaced with the target point, path planning is performed again, and a final global optimal path is obtained, the guiding direction of the UAV is changed, so that the planned path will not fall into a local minimum value.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of path planning, and particularly relates to a UAV path planning method and system based on gradient descent method. BACKGROUND

[0002] Most of the existing path planning algorithms are composed of two parts of front-end path search and back-end trajectory optimization, that is, an initial path without obstacles is searched first, and then a smooth, collision-free and dynamically feasible path is obtained through trajectory optimization. The front-end path search algorithm mainly includes heuristic A* algorithm, hybrid A* algorithm and sampling-based methods such as Probabilistic Roadmap Method (PRM) and Rapidly-Exploring Random Tree (RRT); and the trajectory optimization method mainly has soft constraint optimization and hard constraint optimization, including fourth derivative of trajectory polynomial (minimum snap), Bezier curve optimization, B-spline curve optimization and quadratic programming, and the optimal solution of the trajectory is obtained through numerical solution methods such as gradient descent method and Newton method. In these algorithms, the gradient-based method is widely used, that is, the size and direction of the gradient are estimated by constructing the Euclidean Signed Distance Field (ESDF), and the optimal path is obtained through numerical optimization.

[0003] The inventors find that, although these optimization algorithms have high computational efficiency, there are two problems, on the one hand, the planning strategy based on ESDF and gradient descent method will often fall into local optimal solution, thereby directly leading to the failure of planning, especially in the presence of U-shaped obstacles, L-shaped obstacles or large obstacle environment; in order to solve this problem, the existing work searches a guide path, adds the distance cost from the current point to the guide path point in the objective function, or inserts an intermediate path point to pull the current planning path out of the local minimum point, although a feasible path can be successfully planned through these ways, but the final planning path may also produce large turning, sudden increase in speed. On the other hand, the dynamic adjustment of speed still lacks consideration, one way to improve flight efficiency is to make the unmanned aerial vehicle fly at the maximum expected speed, but it will greatly increase the collision probability in places with large corners; in addition, too high speed may cause deviation in state estimation, perception and control of the unmanned aerial vehicle, causing the unmanned aerial vehicle to deviate from the planned trajectory; and too conservative speed will greatly reduce the flight efficiency and cause unnecessary resource and energy consumption; the existing speed planner mostly considers the information of the surrounding environment, such as the density of obstacles within the perception range and the shortest distance from the current point to the obstacle, to adaptively adjust the flight speed, however, these strategies are incomplete, without considering the influence of speed direction change on flight safety, and cannot meet the demand of speed adapting to the trajectory curvature, that is, collision problem may still occur in places with large trajectory curvature. SUMMARY

[0004] In order to solve the above problems, the present application provides a UAV path planning method and system based on gradient descent method, the present application changes the temporary target point planning strategy and dynamically adjusts the speed, so that the planned path neither falls into local minimum value nor can realize the dynamic adjustment of speed adapting to the trajectory curvature.

[0005] In order to achieve the above purpose, the present application is realized by the following technical scheme:

[0006] In the first aspect, the present application provides a UAV path planning method based on gradient descent method, comprising:

[0007] An initial path search is performed in a preset map using a hybrid A* algorithm to obtain a global initial path;

[0008] Based on the global initial path, B-spline curve fitting is performed on the initial path points to obtain control points of the B-spline function; a speed adjustment factor is introduced, the control points are taken as decision variables, and the gradient descent method is used to obtain an optimal path;

[0009] Collision detection is performed in the optimal path, if the number of collisions is greater than a preset number of times, a temporary intermediate target point is selected near the obstacle, the temporary intermediate target point is replaced as the target point, path planning is performed again, and a final global optimal path is obtained.

[0010] Further, the UAV related parameter information and the environment information are acquired, the obstacle is dilated, and a three-dimensional map is obtained as a preset map.

[0011] Further, the UAV related parameter information includes a map resolution, a maximum speed and acceleration limit of the UAV, and the environment information includes a starting point position and a terminal point position of the UAV.

[0012] Further, the node expansion is performed according to a UAV state space model, the cost function is set as a function about control input, and the distance from the current expansion node to the target point is used as a heuristic cost function, global path search is performed, and an initial path point meeting the UAV kinematics and optimal control cost is obtained.

[0013] Further, the initial path point is fitted by a uniform cubic B-spline curve using a matrix expression of a B-spline function, initial control points of the B-spline function are obtained, and a smoothness cost, a collision cost and a dynamic feasibility cost are optimized; when the dynamic feasibility cost is optimized, a speed adjustment factor is introduced, the greater the speed adjustment factor, the tighter the constraint on the speed, and the smaller the speed adjustment factor, the looser the constraint on the speed.

[0014] Further, the flight speed of the UAV is controlled according to the curvature of the trajectory curve, the flight speed is reduced at a position with a large curvature of the trajectory curve, and the flight speed is increased at a position with a smooth trajectory curve.

[0015] Further, if the number of collisions is greater than a preset number of times, a rectangular sampling region is established around the collision point with the collision point as the center;

[0016] The sampling points are uniformly sampled in the rectangular sampling region, and the sampling points are put into a container;

[0017] The sampling points opposite to the current motion direction, the sampling points with a shortest distance to the obstacle less than a preset distance value, and the sampling points colliding with a line connecting the starting point and the terminal point are deleted;

[0018] The heuristic cost of the remaining sampling points is calculated, and the optimal point is selected as a temporary intermediate target point;

[0019] The target point is replaced by the temporary intermediate target point, and path planning is performed again.

[0020] In a second aspect, the application further provides a UAV path planning system based on a gradient descent method, comprising:

[0021] An initial path planning module is configured to perform initial path search in a preset map using a hybrid A* algorithm to obtain a global initial path;

[0022] An optimal path planning module is configured to perform B-spline curve fitting on the initial path points based on the global initial path to obtain control points of the B-spline function, introduce a speed adjustment factor, take the control points as decision variables, and obtain an optimal path by using a gradient descent method.

[0023] Collision detection is performed in the optimal path, if the number of collisions is greater than a preset number, a temporary intermediate target point is selected near the obstacle, the temporary intermediate target point is replaced by the target point, path planning is performed again, and a final global optimal path is obtained.

[0024] In a third aspect, the present application further provides a computer readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the steps of the UAV path planning method based on the gradient descent method of the first aspect.

[0025] In a fourth aspect, the present application further provides an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the UAV path planning method based on the gradient descent method of the first aspect when executing the program.

[0026] Compared with the prior art, the present application has the following advantages:

[0027] 1. The present application is based on the global initial path obtained by the hybrid A* algorithm, the initial path points are fitted by the B-spline curve to obtain the control points of the B-spline function, the speed adjustment factor is introduced, the control points are taken as the decision variables, the optimal path is obtained by using the gradient descent method, the properties of the B-spline curve are fully utilized, the geometric position relationship of the control points is used to approximate the radian of the curve, the speed is easy to control, a large amount of calculation burden is not increased, the speed dynamic adjustment suitable for the radian of the trajectory is realized through the speed adjustment factor, the phenomenon of sharp turning is avoided, and the planning safety is improved; at the same time, collision detection is performed in the optimal path, if the number of collisions is greater than a preset number, a temporary intermediate target point is selected near the obstacle, the temporary intermediate target point is replaced by the target point, path planning is performed again, and a final global optimal path is obtained, the guiding direction of the UAV is changed, and the planned path will not fall into a local minimum value;

[0028] 2、The application controls the flight speed of the unmanned aerial vehicle according to the curve radian, reduces the flight speed in the place where the track radian is large, increases the flight speed in the place where the track is relatively smooth, thereby avoiding the phenomenon of sharp turns, improving the planning safety, and simultaneously, the speed adjustment mechanism suitable to the track avoids the too conservative flight, so that the unmanned aerial vehicle has higher flexibility; the properties of the B-spline curve are fully utilized, the radian of the curve is approximately represented by using the geometric position relationship of the control points, so that the speed is convenient to control, and a large amount of calculation burden is not increased; in addition, for the local minimum value problem caused by the gradient descent solving method, a re-planning mechanism of an intermediate temporary target point is set, under the general condition, the unmanned aerial vehicle will be trapped in the local minimum value in a place very close to the obstacle, an optimal temporary target point is found around the local minimum value point by the sampling method, the guiding direction of the unmanned aerial vehicle is changed, thereby perfectly avoiding the problem, and the planning success rate is improved. BRIEF DESCRIPTION OF DRAWINGS

[0029] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification. The illustrations are shown to explain the present application and are not intended to limit the present application in any way.

[0030] Figure 1 It is a flowchart of embodiment 1 of the present application;

[0031] Figure 2 It is an algorithm flowchart of the intermediate temporary target point strategy of embodiment 1 of the present application;

[0032] Figure 3 It is a schematic diagram of the optimal intermediate temporary target point of embodiment 1 of the present application;

[0033] Figure 4 It is a two-dimensional schematic diagram of the simulation result speed-path of embodiment 1 of the present application;

[0034] Figure 5 It is a two-dimensional schematic diagram of the simulation result speed-path of the comparative algorithm ego-planner of embodiment 1 of the present application. DETAILED DESCRIPTION

[0035] The present application will be further described below in conjunction with the drawings and embodiments.

[0036] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0037] Embodiment 1:

[0038] The unmanned aerial vehicle has wide application in civil and military fields due to its small size, low cost and high sensitivity, such as rescue and disaster relief, aerial photography, logistics transportation and detective surveillance. Path planning is a key link for unmanned aerial vehicle to achieve the task. At present, there are a large number of research results that can generate a safe, smooth and kinematic characteristic trajectory between the starting point and the target point, which improves the autonomy and efficiency of path planning of the unmanned aerial vehicle. However, in the actual application scene, due to the limited field of view of the sensor of the unmanned aerial vehicle, the perception and control exist errors and other factors, the unmanned aerial vehicle still has a high probability of collision in the flight process, especially in the complex environment with a large number of large obstacles, which limits the autonomy and environmental adaptability of the unmanned aerial vehicle.

[0039] In view of the problems of local path planning falling into local minimum value and dynamic adjustment of speed, as shown in Figure 1 The embodiment provides a path planning method of unmanned aerial vehicle based on gradient descent method. By replacing the temporary target point planning strategy, a temporary intermediate target point is selected near the obstacle to guide the flight direction of the aircraft, so that the aircraft is successfully prevented from falling into local minimum value. According to the dynamic constraint of the trajectory radian on the flight speed, the adaptive change of the size of the unmanned aerial vehicle speed is realized, the flight safety is ensured and the planning success rate is improved, and the average flight speed is improved. The specific implementation steps of the method include:

[0040] S1, acquiring parameter information and environment information required for planning, performing inflation processing on the obstacles, and obtaining a three-dimensional map;

[0041] S2, using a hybrid A* algorithm for initial path search to obtain a global initial path with optimal control cost;

[0042] S3, establishing a B-spline optimization model, introducing a speed adjustment factor, taking the control points as decision variables, and using the gradient descent method to obtain an optimal path;

[0043] S4, on this basis, collision is detected at regular time intervals, if the current planning path is detected to collide for multiple times, it indicates that the planning path falls into local minimum value, a temporary target point replacement strategy is started, a temporary intermediate target point is found, and the original planning path is guided out of the local minimum point;

[0044] S5, continuously executing steps S2, S3 and S4 until the final target point is reached; the successfully planned path is a global optimal path.

[0045] In step S1, the information includes the map range and the depth camera sensing range; the map resolution, the maximum speed and the acceleration limit of the unmanned aerial vehicle are set; after the inflation processing of the obstacles, the Euclidean signed distance field is established; the starting point and the terminal point positions of the unmanned aerial vehicle are obtained.

[0046] In step S2, first, node expansion is performed according to a UAV state space model, a cost function is set as a function of control input, and a global path search is performed with a distance from a current expansion node to a target node as a heuristic cost function, and finally an initial path point satisfying UAV kinematics and optimal in control cost is obtained.

[0047] In step S3, uniform cubic B-spline curve fitting is performed on the initial path point by using a matrix expression of a B-spline function, initial control points of the B-spline function are obtained, and a smoothness cost, a collision cost and a kinematic feasibility cost are optimized. Specifically as follows:

[0048] S3.1 gives n+1 control points P0, P1,..., P n , defines the m-th (0≤m≤n-3) cubic uniform B-spline curve on the time node interval [t m , t m+1 ] determined by the four control points [P m , P m+3 ], that is:

[0049]

[0050] wherein, (t∈[t m , t m+1 ], Δt=t m+1 -t m ), N i，k (u) is a base function, which can be obtained by a recursive formula:

[0051]

[0052] wherein, i, n, k and l are constants; u i , u i+1 ,..., u i+k are node vectors with an interval of 1. The control points fitted by formula (1) are taken as decision variables.

[0053] S3.2 Let K be the number of optimized control points.

[0054] Optimization target 1: smoothness cost J s , wherein Jerk is a third derivative of position, which can be obtained by a recursive formula:

[0055]

[0056]

[0057]

[0058] where V i is the ith velocity control point; A i is the ith acceleration control point.

[0059] S3.3 Optimization Objective 2: Collision Cost J c The shortest distance d(P i ) from the current point to nearby obstacles is obtained from the ESDF, to avoid planning a path that is too close to obstacles, the collision cost is set as an exponential term, i.e.:

[0060]

[0061]

[0062] where d thr is a safety threshold, which is a constant.

[0063] S3.4 Optimization Objective 3: Dynamic Feasibility Cost J f The maximum speed and acceleration values of the UAV are V max and A max , respectively, which are set in step S1. A speed adjustment factor η i is introduced, and η i is an expression related to the curvature of the curve, which can be approximately obtained according to the position relationship of control points P0, P1, …, P n .

[0064]

[0065] where β is a known constant; the greater the value of η i , the tighter the constraint on the speed, and the smaller the value of η i , the looser the constraint on the speed.

[0066] The speed feasibility cost adapted to the curvature of the trajectory is:

[0067]

[0068] The acceleration feasibility cost is:

[0069]

[0070] Therefore, the dynamic feasibility cost is:

[0071]

[0072] Therefore, the objective function of the optimization is:

[0073] J = λ s J s + λc J c +λ f J f (8)

[0074] where λ s , λ c and λ f are weight coefficients, which can be manually adjusted according to requirements. Thus, solving the optimal path problem is converted into an unconstrained nonlinear optimization problem, which is solved by using the gradient descent method. Finally, the B-spline curve fitted according to the optimized control points is the optimal path.

[0075] In step S4, in the real-time planning process of the UAV, the surrounding obstacle environment information is unknown, and only real-time environment information can be obtained through sensors. Since the sensing range of the sensor is limited, the area outside the sensing range is considered to be an unobstructed feasible area until the UAV senses new obstacles. Therefore, the entire process is a process of planning while walking. If the same path is detected to collide multiple times, it means that the path planning fails and is likely to fall into a local minimum value. At this time, the target point replacement algorithm is started, and the specific method is as follows:

[0076] S4.1 After multiple collisions are detected, record the collision point P.

[0077] S4.2 Uniformly sample in the rectangular area with P as the center, l as the length, and d as the width, to obtain the sampling point T and put it into the container.

[0078] S4.3 To prevent the trajectory from walking back, delete the sampling points opposite to the current motion direction; record the position P' of the point before the collision point, and if delete the sampling point T.

[0079] S4.4 Detect whether the remaining sampling points are occupied by obstacles, and consider that the point is occupied if the shortest distance to the obstacle is less than a sufficiently small value; delete the occupied sampling points in the container.

[0080] S4.5 Detect whether the remaining sampling points in the container collide with the line connecting the start point and the end point, and delete the sampling point if a collision occurs.

[0081] S4.6 Calculate the heuristic cost of the remaining sampling points, and select the optimal point as the intermediate target point.

[0082] S4.7 Replace the target point with the selected intermediate target point and re-plan.

[0083] S4.8 To avoid the UAV stopping flying when reaching the intermediate target point, replace the target point with the final target point when the UAV reaches a certain distance from the intermediate target point, and continue planning.

[0084] In step S5, if the UAV enters a new flight area, the hybrid A* search, B-spline optimization and collision detection are re-performed until the target point is reached.

[0085] Embodiment 2

[0086] To further explain embodiment 1, this embodiment provides a UAV path planning method based on gradient descent method on the basis of embodiment 1, including the following steps:

[0087] S1, prepare a 40m*20m*10m map, set the map resolution to 0.1, set the point cloud obstacle in the map to occupy the environment and inflate the obstacle, establish ESDF, and each obstacle has a unique index identifier. Initialize the binocular camera T265 and depth camera parameters, set the initial point coordinates to (-16, 0), the maximum speed of the UAV to 3.0m / s, and the maximum acceleration to 2.0m / s 2 At this time, the state of the UAV is stationary and waiting for the target point.

[0088] S2, after obtaining the target point position information, change the state of the UAV to waiting for path generation, obtain the obstacle point cloud information through the sensor, start the hybrid A* algorithm, and obtain an initial path that satisfies the kinematics of the UAV without obstacles. By uniformly sampling the initial path, an initial control point sequence is obtained according to the B-spline fitting matrix expression, a B-spline optimization model with speed adjustment suitable for the radius of the trajectory is established, and the smoothness, safety and dynamics feasibility of the trajectory are optimized respectively, and the problem is converted into an unconstrained nonlinear optimization problem:

[0089] min λ s J s +λ c J c +λ f J f (9)

[0090] First, assign values to the parameters by experience, and finally select the values of parameters λ s , λ c and λ f as 5, 10 and 2 through multiple parameter adjustments and tests. Use the gradient descent method to solve the problem to obtain the optimal control point sequence, and use the Cox-de Boor formula to obtain the trajectory points. Finally, an executable optimal path is obtained, and in this process, cubic uniform B-spline curve fitting is used.

[0091] S3, if the path fails in step S2, the planning strategy is started again to re-plan, and after multiple failures, the state of the unmanned aerial vehicle is changed to waiting for a target, and the target point is set again; if the path is successfully generated in step S2, the state of the unmanned aerial vehicle waiting for path generation is changed to the state of executing the path, the unmanned aerial vehicle is unlocked, and the unmanned aerial vehicle flies according to the planned path. Since the sensing range of the sensor is limited, whenever the unmanned aerial vehicle detects new environmental information, the path is re-planned to meet the real-time requirement.

[0092] S4, before reaching the target point, steps S2 and S3 are continuously executed, and collision detection is performed every 0.01s through ROS message publishing. If it is detected that the current path collides, the state of the unmanned aerial vehicle executing the path is changed to the re-planning state, and the path is re-planned until the planned path does not collide with the obstacle; if the current path is detected to collide multiple times, it indicates that the planned path is trapped in a local minimum point, the current planned path fails, and the intermediate temporary target point strategy is started, as shown in Figure 3 、 Figure 4 and Figure 5 , the target point is replaced, the state of the unmanned aerial vehicle is changed to re-planning, the current planned path is guided out of the local minimum point, and the state of the unmanned aerial vehicle is changed to the state of executing the path.

[0093] S5, step S4 is repeated until the final target point position is reached, and the state of the unmanned aerial vehicle is changed to landing. If the target point cannot be reached, a point not occupied by the obstacle is found on a circle with the target point as the center and a radius of 0.3m around the target point, and the point is changed to the target point.

[0094] The feasibility and effect of the application are verified through 20 simulation experiments. The specific simulation results can be seen in Figure 4 and Figure 5 , and the comparison of the obtained data is shown in Table 1:

[0095] Table 1 Comparison of data of unmanned aerial vehicle path planning and speed planning algorithm based on gradient descent method

[0096]

[0097] In 3 groups of 20 simulation experiments, the same obstacle environment, the same starting point and target point, and the maximum speed limit of 3m / s and the maximum acceleration limit of 2m 2 / s 2, the sensor parameters are all the same. From the above experimental data, it can be seen that in the set obstacle environment, the algorithm in the embodiment improves the success rate of planning, which is increased by 75% and 5% respectively compared with the other two algorithms, and significantly reduces the length of the planned path. Although more planning time is consumed, the flight speed is not reduced during flight, and the average speed is slightly improved. Therefore, compared with other algorithms, the algorithm in the embodiment is more suitable for large obstacle environment, and has advancement.

[0098] From Figure 4 and Figure 5 The path-velocity diagram can be seen, Figure 4 For the algorithm in the embodiment, Figure 5 For the ego-planner algorithm. Figure 4 In the embodiment, the unmanned aerial vehicle significantly reduces the speed at the place where the planned path is curved to a large extent, and significantly increases the flight speed at the place where the trajectory is smooth, which meets the desired result. In addition Figure 4 The planned path in the embodiment mostly maintains straight flight in the obstacle-free place, reduces the path length while improving the flight efficiency, while Figure 5 There are no less than one unnecessary turn.

[0099] Embodiment 3:

[0100] The embodiment provides an unmanned aerial vehicle path planning system based on gradient descent method, comprising:

[0101] The initial path planning module is configured to use a hybrid A* algorithm to perform initial path search in a preset map to obtain a global initial path;

[0102] The optimal path planning module is configured to perform B-spline curve fitting on the initial path points based on the global initial path to obtain control points of the B-spline function; introduce a speed adjustment factor, take the control points as decision variables, and use the gradient descent method to obtain an optimal path;

[0103] Collision detection is performed in the optimal path. If the number of collisions is greater than a preset number, a temporary intermediate target point is selected near the obstacle, the temporary intermediate target point is replaced with the target point, and path planning is performed again to obtain a final global optimal path.

[0104] The working method of the system is the same as the unmanned aerial vehicle path planning method based on gradient descent method in embodiment 1, which will not be repeated here.

[0105] Embodiment 4:

[0106] The embodiment provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement steps of the UAV path planning method based on the gradient descent method in the embodiment 1.

[0107] Embodiment 5:

[0108] The embodiment provides an electronic device, which comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, and the processor implements steps of the UAV path planning method based on the gradient descent method in the embodiment 1 when executing the program.

[0109] The above merely provides preferred embodiments of the embodiment, and is not intended to limit the embodiment. The embodiment can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the embodiment shall be included in the protection scope of the embodiment.

Claims

1. A UAV path planning method based on gradient descent, characterized in that, include: Using Mixed A The algorithm performs an initial path search in a pre-defined map to obtain a global initial path; Based on the global initial path, B-spline curves are fitted to the initial path points to obtain the control points of the B-spline function; a velocity adjustment factor is introduced, and the control points are used as decision variables to obtain the optimal path using the gradient descent method. Collision detection is performed on the optimal path. If the number of collisions exceeds a pre-determined number, a temporary intermediate target point is selected near the obstacle, and this temporary intermediate target point is replaced with the target point. Path planning is then performed again to obtain the final globally optimal path. Speed ​​adjustment factor. for: ； in, The coefficients are known constants; , and The first Individual points, the first Individual points and the first Individual points.

2. The UAV path planning method based on gradient descent as described in claim 1, characterized in that, The system acquires relevant parameters and environmental information about the drone, inflates obstacles, and obtains a 3D map as a preset map.

3. The UAV path planning method based on gradient descent as described in claim 2, characterized in that, The drone-related parameter information includes map resolution, maximum speed and acceleration limit; environmental information includes the drone's starting point and ending point.

4. The UAV path planning method based on gradient descent as described in claim 1, characterized in that, Node expansion is performed based on the UAV state space model. The cost function is set as a function of the control input, and the distance from the current expanded node to the target point is used as the heuristic cost function to perform a global path search, thereby obtaining the initial path point that satisfies the UAV kinematics and has the optimal control cost.

5. The UAV path planning method based on gradient descent as described in claim 4, characterized in that, The initial path points are fitted with uniform cubic B-spline curves using the matrix expression of the B-spline function to obtain the initial control points of the B-spline function, and the smoothness cost, collision cost, and dynamic feasibility cost are optimized. When optimizing the dynamic feasibility cost, a velocity adjustment factor is introduced. The larger the velocity adjustment factor, the tighter the constraint on the velocity, and the smaller the value of the velocity adjustment factor, the looser the constraint on the velocity.

6. The UAV path planning method based on gradient descent as described in claim 5, characterized in that, The flight speed of the drone is controlled based on the curvature of the trajectory curve. The flight speed is reduced when the trajectory curve is large and increased when the trajectory curve is smooth.

7. The UAV path planning method based on gradient descent as described in claim 1, characterized in that, If the number of collisions exceeds the pre-set number, a rectangular sampling area is established around the collision point, centered on the point of collision. Sample evenly within the rectangular sampling area to obtain sampling points, and place them in a container; Delete sampling points that are opposite to the current direction of movement; delete sampling points whose shortest distance to an obstacle is less than a preset distance value; and delete sampling points that collide with the line connecting the start and end points. Calculate the heuristic cost of the remaining sampling points and select the optimal point as a temporary intermediate target point; Change the target point to a temporary intermediate target point and replan the route.

8. A UAV path planning system based on gradient descent, characterized in that, include: The initial path planning module is configured to use hybrid A. The algorithm performs an initial path search in a pre-defined map to obtain a global initial path; The optimal path planning module is configured to: use the global initial path as a basis, fit the initial path points with a B-spline curve to obtain the control points of the B-spline function; introduce a speed adjustment factor, use the control points as decision variables, and use the gradient descent method to find the optimal path. Collision detection is performed on the optimal path. If the number of collisions exceeds a pre-determined number, a temporary intermediate target point is selected near the obstacle, and this temporary intermediate target point is replaced with the target point. Path planning is then performed again to obtain the final globally optimal path. Speed ​​adjustment factor. for: ； in, The coefficients are known constants; , and The first Individual points, the first Individual points and the first Individual points.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the UAV path planning method based on gradient descent as described in any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the UAV path planning method based on gradient descent as described in any one of claims 1-7.

Images