UAV path planning method based on batch processing prior knowledge tree

Abstract

The present invention discloses a method for UAV path planning based on batch prior knowledge tree, by adding sampling point range control parameters to the combined force function of the artificial potential field method, dynamically adjusting the generation range of sampling points according to the density of obstacles, and defining a threshold value, when the combined force function of the sampling point is less than the threshold value, the point is retained, otherwise it is eliminated. This process greatly reduces the redundant sampling and waste of computing resources caused by the random generation sampling point strategy; in the process of path optimization, the present invention converts an ellipse sampling set into two local subset ellipse sampling sets, reduces the sampling area, and improves the sampling efficiency; and by updating the path cost from the starting point to the beacon point, increases the area of ​​the second ellipse, thereby increasing the probability that the passable area between obstacles is inside the ellipse; at the same time, introduces the obstacle density factor, increases the probability of beacon points being generated next to obstacles and the possibility of paths passing through complex obstacles, and optimizes the path.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle path planning, and in particular to a method for unmanned aerial vehicle path planning based on batch processing priori knowledge trees. Background Art

[0002] In the field of robot motion planning, the performance of the path planning algorithm plays a decisive role in whether the robot can complete the task efficiently and safely. Traditional path planning algorithms mainly include graph search-based algorithms and sampling-based algorithms. Graph search-based algorithms can find better paths in discrete space, but in high-dimensional space, the discrete state space grows exponentially and the amount of calculation increases dramatically; sampling-based algorithms avoid discretization problems by randomly sampling continuous planning domains, but their search is disordered, convergence speed is slow, and the quality of the solution is limited. The batch prior knowledge tree algorithm BIT* balances the two and performs an ordered search in the continuous domain through batch samples, which can accelerate convergence. However, with the continuous expansion of application scenarios, drones need to perform tasks in narrow and complex regional environments. The BIT* algorithm still has obvious limitations when facing path planning in narrow and complex areas. Although the ordered search characteristics of the algorithm help to exclude some invalid areas to a certain extent, due to its random sampling method, a large number of unnecessary sampling points will be generated, and key areas may not be fully sampled in narrow and complex areas, making it difficult to find a feasible path. Summary of the invention

[0003] Purpose of the invention: The present invention aims to provide a UAV path planning method based on batch processing prior knowledge tree to cope with complex and changing environments.

[0004] Technical solution: The UAV path planning method based on batch processing prior knowledge tree of the present invention comprises the following steps:

[0005] (1) Initialize the set, which includes the tree edge set, tree node set, sampling point set, and tree node priority queue and the priority queue of edges to be expanded ; Place the starting point into the tree node set and the tree node priority queue, and the target point into the sampling point set;

[0006] (2) Introducing the sampling point range control function and the threshold of the absolute value of the resultant force, combining the artificial potential field method to generate sampling points, and updating the sampling point set;

[0007] (3) According to the current and The minimum estimated cost is used to determine the optimal edge to be expanded, obtain the initial path, and enter the target point into the tree node set;

[0008] (4) Taking the starting point and the target point of the initial path as the focus, determine the major axis and the minor axis according to the optimal path length and the straight path length between the starting point and the target point, construct an ellipse, perform pruning optimization on the initial path obtained in step (3), perform batch sampling in the ellipse, and update the sampling point set;

[0009] (5) Introducing the obstacle density factor, the multi-arm gambling algorithm EXP3 is used to determine the beacon point from the set of sampling points updated in step (4);

[0010] (6) Take the target point out of the tree node set and put it back into the sampling point set, leaving it empty. and , put the starting point into , according to the corresponding and The minimum estimated cost is used to determine the optimal edge to be extended. After the optimal edge to be extended cannot reduce the path cost from the starting point to the beacon point, the first sub-ellipse is constructed with the starting point and the beacon point as the focus and the path cost from the starting point to the beacon point as the major axis; the second sub-ellipse is constructed with the target point and the beacon point as the focus and the current optimal path cost minus the path cost from the starting point to the beacon point as the major axis;

[0011] (7) Leave it blank again and , and place the starting point in , and then according to the corresponding and The minimum estimated cost is used to determine the optimal edge to be expanded until the scheduled time is reached;

[0012] (8) Determine whether the target point enters the tree node set. If so, proceed to the next step. Otherwise, return to the ellipse batch sampling in step (4);

[0013] (9) Perform pruning optimization on the first sub-ellipse and the second sub-ellipse, and batch sample in the two sub-ellipses; leave blank and Place the sample points on the path into ,according to and The minimum estimated cost is used to determine the optimal edge to be extended, and the optimized path is updated again until the optimal edge to be extended cannot reduce the current path cost;

[0014] (10) Determine whether the iteration is completed. If completed, output the path. If not completed, return to the batch sampling of step (9);

[0015] (11) Perform third-order B-spline smoothing on the output path of step (10) to obtain the final path.

[0016] Furthermore, in step (2), the absolute value of the combined force for

[0017] ;

[0018] ;

[0019] ;

[0020] ;

[0021] in, is the sampling point range control function, ; is the gravitational potential field function, is the repulsive potential field function; Indicates the area of ​​the map. Indicates The area of ​​the obstacle, is the total amount of obstacles; , and is the threshold constant, For sampling point To the destination The Euclidean distance of and is the gravitational gain coefficient; is the repulsion coefficient; is the distance factor adjustment coefficient, and its value is a constant greater than zero.

[0022] Calculate the average combined force of all current valid sampling points, threshold for

[0023] ;

[0024] in, is the threshold adaptive coefficient, ; is the number of valid sampling points, is the combined force of the effective sampling points; Less than If , the sampling point is retained, otherwise, the sampling point is removed.

[0025] Furthermore, step (3) is as follows:

[0026] (31) If the current tree node priority queue The minimum estimated cost is less than or equal to the priority queue of the edge to be expanded The minimum estimated cost is then expanded until Greater than ; Otherwise, the edge to be expanded with the lowest estimated cost is selected from the priority queue of edges to be expanded as the optimal edge to be expanded;

[0027] (32) If the estimated cost of the optimal edge to be expanded is less than the current path cost, go to step (33); otherwise, clear the queue and end this iteration;

[0028] (33) If the optimal edge to be expanded has no collision and can reduce the cost, go to step (34); otherwise, clear the queue and end this iteration;

[0029] (34) If the terminal node of the optimal edge to be expanded is in the current tree node list, the node is rewired and the process goes to step (35); otherwise, the terminal node is moved from the sampling node list to the tree node set and the process goes to step (35);

[0030] (35) Add the optimal edge to be expanded to the tree edge set and clean up useless edges, and return to step (31) until the optimal edge to be expanded cannot reduce the current path cost and the initial path is determined.

[0031] Furthermore, in step (4), the major axis of the ellipse is the current optimal path length , the short axis is , is the straight line path length between the start point and the destination point.

[0032] Furthermore, in step (5), the obstacle density factor is introduced, and the beacon point is determined from the updated sampling point set obtained in step (4) using the EXP3 algorithm, as follows:

[0033] (51) The updated sampling point set obtained in step (4) , initialize the weight of each node in the node set to 1;

[0034] (52) Calculation Moment The probability of selecting a node as a beacon ;

[0035] (53) Based on the selection probability of the beacon selected by the EXP3 algorithm, a node is selected as the beacon according to the selection probability of the beacon , and introduce the obstacle density factor to calculate the beacon Rewards , and then normalize it to get the normalized reward ;

[0036] (54) Update the beacon weight and obtain the updated beacon weight ;

[0037] (55) Repeat steps (52) to (54) until the predetermined time is reached and generate the final beacon. .

[0038] Furthermore, Moment The probability of selecting a node as a beacon for

[0039] ;

[0040] in, is the current beacon weight, is the number of sampling points.

[0041] Further, beacon Rewards for

[0042] ;

[0043] ;

[0044] ;

[0045] in, and is the weight parameter, From beacon To the destination The potential path cost of is the density of obstacles; A is the beacon and target point The center is the midpoint of The area of ​​the circular region; is the indicator function;

[0046] Normalized Reward for

[0047] ;

[0048] in, is the absolute value of the reward.

[0049] Furthermore, the updated beacon weight for

[0050] ;

[0051] in, is the learning rate, .

[0052] Furthermore, batch sampling is performed within the ellipse, and the sampling points The calculation equation is as follows:

[0053] ;

[0054] in, is the scaling factor, , for A random number uniformly distributed on

[0055] The sampling point Perform inverse rotation and inverse translation operations in sequence to obtain the coordinates of the sampling point in the original coordinate system.

[0056] Furthermore, in step (11), the third-order B-spline curve is

[0057] ;

[0058] in, A collection of time, is a fourth-order matrix, is a set of control points.

[0059] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: 1. The present invention adds a sampling point range control parameter to the resultant force function of the artificial potential field method, which can dynamically adjust the generation range of the sampling points according to the density of obstacles and define the threshold value. , when the combined force function of the sampling point is less than the threshold, the point is retained, otherwise it is removed. This process greatly reduces the redundant sampling and waste of computing resources caused by the random generation of sampling point strategy; 2. The present invention converts an ellipse sampling set into two local subset ellipse sampling sets, reduces the sampling area, and improves the sampling efficiency; and by updating the path cost from the starting point to the beacon point, increases the area of ​​the second ellipse, thereby increasing the probability that the passable area between obstacles is inside the ellipse; at the same time, an obstacle density factor is introduced to make the probability of beacon points being generated next to obstacles greater, increasing the possibility of the path passing through complex obstacles, thereby optimizing the path; 3. When the present invention performs batch sampling inside the ellipse, the present invention adopts a sampling method based on the ellipse parameter equation, introduces a scaling factor, and can randomly obtain the coordinates of the sampling points inside the ellipse through simple trigonometric functions. Compared with the method of random sampling and then judging whether the point is inside the ellipse, the computational overhead is smaller and the efficiency is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a flow chart of the present invention;

[0061] Figure 2 It is a spatial schematic diagram of an ellipse of the present invention;

[0062] Figure 3is a spatial schematic diagram of the first and second sub-ellipses of the present invention;

[0063] Figures 4 to 7 is a schematic diagram of steps (4) to (9) of the present invention;

[0064] Figure 8 is a schematic diagram of pruning in step (9) of the present invention;

[0065] Fig. 9 It is a schematic diagram of tree node expansion of the present invention. DETAILED DESCRIPTION

[0066] The present invention will be further described below in conjunction with the accompanying drawings.

[0067] The UAV path planning method based on batch processing prior knowledge tree of the present invention comprises the following steps:

[0068] (1) Initialize the set, which includes the tree edge set, tree node set, sampling point set, and tree node priority queue and the priority queue of edges to be expanded ; Place the starting point into the tree node set and the tree node priority queue, and the target point into the sampling point set.

[0069] (2) Introduce the sampling point range control function and the threshold of the absolute value of the resultant force, combine the artificial potential field method to generate sampling points, and update the sampling point set. The specific steps are as follows:

[0070] (21) Random sampling is performed within the region using a batch operation.

[0071] (22) The improved artificial potential field method is used to guide the generation of sampling points. The generation range of sampling points is dynamically adjusted according to the density of obstacles. When the number of obstacles is large, the generation range of sampling points will be larger, and when the number of obstacles is small, the generation range of sampling points will be smaller. Suppose a point is randomly sampled in the map. The improved artificial potential field method will be based on the absolute value of the resultant force Determine whether the sampling point meets the generation conditions, if not, remove it. The formula for the absolute value of the combined force is as follows:

[0072] ;

[0073] In the formula, is the sampling point range control parameter, and its value is determined by the obstacle ratio in the environment. . The smaller it is, the weaker the ability of the artificial potential field method to influence the generation of sampling points is, and the sampling points tend to be generated in the global map; The larger it is, the stronger the ability of the artificial potential field method to influence the generation of sampling points is, and the sampling points tend to be generated within the range between the starting point and the target point. The specific formula is as follows:

[0074] ;

[0075] In the formula, Indicates the area of ​​the map. Indicates The area of ​​obstacles; n is the total number of obstacles;

[0076] is the gravitational potential field function. When the agent is too far away from the target point when calculating the gravitational potential field in the traditional artificial potential field method, it will cause local optimality. Therefore, an improved gravitational potential field function is added to divide the gravitational potential field function according to different distances. The formula is as follows:

[0077] ;

[0078] In the formula, and is the defined threshold constant, For sampling point To the destination The Euclidean distance of and is the gravitational gain coefficient. Less than the defined threshold constant When , the gravitational potential field function is consistent with the gravitational potential field function of the traditional artificial potential field method; when exist and In between, add As the new gravitational gain coefficient, ;when Greater than a defined threshold constant When is a fixed value. By definition and , which can realize dynamic adjustment of gravitational potential field function .

[0079] It is a repulsive potential field function. In the repulsive potential field function of the traditional artificial potential field method, it is first necessary to define the repulsive potential field range of an obstacle. When the intelligent agent moves within the range, the repulsive force of the obstacle on the intelligent agent is calculated. However, in some actual scenarios, when the obstacle is too close to the target point, the repulsive force of the obstacle on the intelligent agent will be greater than the gravitational effect of the target point on the intelligent agent. The improved repulsive potential field function used in the present invention adds the distance factor between the target point and the intelligent agent on the basis of the traditional repulsive potential field function. , when the agent approaches the target point, it can reduce the influence of the repulsive potential field function on the agent. The formula is as follows:

[0080] ;

[0081] In the formula, is the repulsion coefficient; is the distance factor adjustment coefficient, and its value is an arbitrary constant greater than zero.

[0082] The present invention sets a threshold , According to the absolute value distribution of the combined force of the currently generated sampling points, the setting is performed, the average combined force of all current valid sampling points is calculated, and the threshold is set to a certain proportion of the average value. The formula is as follows:

[0083] ;

[0084] in, is the threshold adaptive coefficient, , which can ensure that the sampling points are effectively screened within a certain range; is the number of valid sampling points, is the combined force of the effective sampling points; Less than If , the sampling point is retained, otherwise, the sampling point is removed.

[0085] (3) According to the current and The minimum estimated cost is used to determine the optimal edge to be expanded, obtain the initial path, and enter the target point into the tree node set.

[0086] (31) If the current tree node priority queue The minimum estimated cost is less than or equal to the priority queue of the edge to be expanded The minimum estimated cost is then expanded until Greater than ; Otherwise, the edge to be expanded with the lowest estimated cost is selected from the priority queue of edges to be expanded as the optimal edge to be expanded;

[0087] (32) If the estimated cost of the optimal edge to be expanded is less than the current path cost, go to step (33); otherwise, clear the queue and end this iteration;

[0088] (33) If the optimal edge to be expanded has no collision and can reduce the cost, go to step (34); otherwise, clear the queue and end this iteration;

[0089] (34) If the terminal node of the optimal edge to be expanded is in the current tree node list, the node is rewired and the process goes to step (35); otherwise, the terminal node is moved from the sampling node list to the tree node set and the process goes to step (35);

[0090] (35) Add the optimal edge to be expanded to the tree edge set and clean up useless edges, and return to step (31) until the optimal edge to be expanded cannot reduce the current path cost and the initial path is determined.

[0091] (4) Taking the starting point and the target point of the initial path as the focus, determine the major axis and the minor axis according to the optimal path length and the straight path length between the starting point and the target point, construct an ellipse, perform pruning optimization on the initial path obtained in step (3), perform batch sampling in the ellipse, and update the sampling point set.

[0092] Pruning operation. The path cost from the constructed starting point to the target point is used as the major axis of the ellipse, and the starting point and the target point are used as the foci to construct the ellipse. According to the properties of the ellipse: the sum of the distances from any point on the edge of the ellipse to the two foci of the ellipse is the length of the major axis, so if the sampling point is outside the ellipse, then its distance to the two foci must be greater than the current initial path, and it cannot improve the current path, so it is removed. And any point on any edge that is outside the ellipse cannot improve the current path cost, and it also needs to be removed. The specific steps are as follows:

[0093] (41) Calculate the major axis of the ellipse, which is the current optimal path length .

[0094] (42) Calculate the minor axis of the ellipse, whose length is , is the length of the straight line path between the start and end points.

[0095] (43) Calculate the center point of the ellipse, which is the starting point and target point The midpoint between is given by the following formula:

[0096] ;

[0097] (44) Eliminate sampling points outside the ellipse. Since the center point of the ellipse is not at the origin and its major axis may be at an angle to the x-axis, the coordinate transformation method is used to establish an ellipse coordinate system. In the ellipse coordinate system, the center point of the ellipse is at the origin and its major axis is the x-axis. For the sampling points in the original coordinate system , first you need to translate it to a relative position with the center of the ellipse as the reference point, the formula is as follows:

[0098] ;

[0099] Next, let the angle of rotation of the ellipse be (Counterclockwise is the positive direction). The point after translation Coordinates rotated around the origin The calculation formula is

[0100] ;

[0101] For a sampling point after coordinate transformation ; if it satisfies the equation

[0102] ;

[0103] Then remove it.

[0104] In the formula, is the semi-major axis, and its formula is as follows:

[0105] ;

[0106] is the semi-minor axis, and its formula is as follows:

[0107] ;

[0108] (45) Eliminate edges in the tree. For an edge in a tree, if only one of its two endpoints can satisfy the equation after coordinate transformation:

[0109] ;

[0110] Then remove this edge.

[0111] (5) The obstacle density factor is introduced and the multi-arm gambling algorithm EXP3 is used to determine the beacon point from the updated sampling point set obtained in step (4).

[0112] (51) The updated sampling point set obtained in step (4) is , initialize the weight of each node in the node set to 1;

[0113] (52) Calculation Moment The probability of selecting a node as a beacon

[0114] ;

[0115] in, is the current beacon weight, is the number of sampling points.

[0116] (53) Based on the selection probability of the beacon selected by the EXP3 algorithm, a node is selected as the beacon according to the selection probability of the beacon , and introduce the obstacle density factor to calculate the beacon Rewards , and then normalize it to get the normalized reward ;

[0117] Beacon Rewards for

[0118] ;

[0119] ;

[0120] ;

[0121] in, and is the weight parameter, From beacon To the destination The potential path cost of is the density of obstacles; A is the beacon and target point The center is the midpoint of The area of ​​the circular region; is the indicator function;

[0122] Normalized Reward for

[0123] ;

[0124] in, is the absolute value of the reward.

[0125] (54) Update the beacon weight and obtain the updated beacon weight for

[0126] ;

[0127] in, is the learning rate, .

[0128] (55) Repeat steps (52) to (54) until the predetermined time is reached and generate the final beacon. .

[0129] (6) Take the target point out of the tree node set and put it back into the sampling point set, leaving it empty. and , put the starting point into According to the corresponding and The minimum estimated cost is used to determine the optimal edge to be extended. After the optimal edge to be extended cannot reduce the path cost from the starting point to the beacon point, the path cost from the starting point to the beacon point is calculated with the starting point and the beacon point as the focus. The first sub-ellipse is constructed for the major axis; the second sub-ellipse is constructed for the major axis with the target point and the beacon point as the foci and the current optimal path cost minus the path cost from the starting point to the beacon point.

[0130] (7) Leave it blank again and , and place the starting point in , and then according to the corresponding and The minimum estimated cost is used to determine the optimal edge to be expanded until the scheduled time is reached.

[0131] (8) Determine whether the target point enters the tree node set. If so, proceed to the next step. Otherwise, return to the ellipse batch sampling in step (4).

[0132] (9) Perform pruning optimization on the first sub-ellipse and the second sub-ellipse, and batch sample in the two sub-ellipses. Leave blank and , put the sampling points on the path into ,according to and The minimum estimated cost is used to determine the optimal edge to be extended, and the optimized path is updated again until the optimal edge to be extended cannot reduce the current path cost.

[0133] Batch sampling is performed within the ellipse, sampling points The calculation equation is as follows:

[0134] The coordinate axis is translated and rotated so that the center of the ellipse coincides with the origin and the major axis is on the x-axis. The parametric equation of the ellipse is used for uniform sampling in the coordinate system after the coordinate transformation. The formula is as follows:

[0135] ;

[0136] In the formula, is the defined scaling factor used for sampling inside the ellipse. To make the sampling points more evenly distributed inside the ellipse, , for A uniformly distributed random number.

[0137] Then in , uniform sampling within the range The value of and , get the sampling point .

[0138] The sampling point The coordinates of the coordinate system after the inverse rotation operation are converted to the coordinate system after the translation , the inverse rotation formula is as follows:

[0139] ;

[0140] Then Convert to the coordinates of the original coordinate system through the inverse translation operation , the inverse translation formula is as follows:

[0141] ;

[0142] Finally, the sampling points inside the sub-ellipse in the original coordinate system are obtained.

[0143] (10) Determine whether the iteration is completed. If completed, output the path. If not completed, return to the batch sampling of step (9);

[0144] (11) Perform third-order B-spline smoothing on the output path of step (10) to obtain the final path.

[0145] The equation of the third-order B-spline curve is:

[0146] ;

[0147] In the formula, is a set of time, whose value is

[0148] ;

[0149] To normalize the time, the formula is:

[0150] ;

[0151] ;

[0152] is a known fourth-order matrix whose value is

[0153] ;

[0154] is a set of control points, whose values ​​are

[0155] .

Claims

1. A UAV path planning method based on batch processing prior knowledge tree, characterized in that: The following steps are involved: (1) Initialize the set, which includes the tree edge set, tree node set, sampling point set, and tree node priority queue and the priority queue of edges to be expanded ; Place the starting point into the tree node set and the tree node priority queue, and the target point into the sampling point set; (2) Introducing the sampling point range control function and the threshold of the absolute value of the resultant force, combining the artificial potential field method to generate sampling points, and updating the sampling point set; (3) According to the current and The minimum estimated cost is used to determine the optimal edge to be expanded, obtain the initial path, and enter the target point into the tree node set; (4) Taking the starting point and target point of the initial path as the focus, determine the major axis and minor axis according to the optimal path length and the straight path length between the starting point and the target point, construct an ellipse, perform pruning optimization on the initial path obtained in step (3), perform batch sampling in the ellipse, and update the sampling point set; (5) Introducing the obstacle density factor, the multi-arm gambling algorithm EXP3 is used to determine the beacon point from the updated sampling point set in step (4); (6) Take the target point out of the tree node set and put it back into the sampling point set, leaving it empty. and , put the starting point into , according to the corresponding and The minimum estimated cost is used to determine the optimal edge to be extended. When the optimal edge to be extended cannot reduce the path cost from the starting point to the beacon point, the first sub-ellipse is constructed with the starting point and the beacon point as the foci and the path cost from the starting point to the beacon point as the major axis. The second sub-ellipse is constructed with the target point and the beacon point as the foci and the current optimal path cost minus the path cost from the starting point to the beacon point as the major axis. (7) Leave it blank again and , and place the starting point in , and then according to the corresponding and The minimum estimated cost is used to determine the optimal edge to be expanded until the scheduled time is reached; (8) Determine whether the target point enters the tree node set. If so, proceed to the next step. Otherwise, return to the ellipse batch sampling in step (4); (9) Perform pruning optimization on the first and second sub-ellipses, and perform batch sampling in the two sub-ellipses; Leave blank and Place the sample points on the path ,according to and The minimum estimated cost is used to determine the optimal edge to be extended, and the optimized path is updated again until the optimal edge to be extended cannot reduce the current path cost; (10) Determine whether the iteration is completed. If it is completed, output the path. If it is not completed, return to the batch sampling of step (9); (11) The output path of step (10) is smoothed by a third-order B-spline to obtain the final path.

2. The UAV path planning method based on batch processing prior knowledge tree according to claim 1 is characterized in that: In step (2), the absolute value of the combined force for ； ； ； ； in, is the sampling point range control function, ; is the gravitational potential field function, is the repulsive potential field function; Indicates the area of ​​the map, Indicates the The area of ​​the obstacle, is the total amount of obstacles; 、 and is the threshold constant, For sampling points To the destination The Euclidean distance, and is the gravitational gain coefficient; is the repulsion coefficient; is the distance factor adjustment coefficient, and its value is a constant greater than zero. Calculate the average combined force of all current valid sampling points, threshold for ； in, is the threshold adaptive coefficient, ; is the number of valid sampling points, is the combined force of the effective sampling points; when Less than If , the sampling point is retained, otherwise it is removed.

3. The UAV path planning method based on batch processing prior knowledge tree according to claim 1 is characterized in that: Step (3) is as follows: (31) If the current tree node priority queue The minimum estimated cost is less than or equal to the priority queue of the edge to be expanded The minimum estimated cost is then expanded until Greater than Otherwise, the edge to be expanded with the lowest estimated cost is selected from the priority queue of edges to be expanded as the optimal edge to be expanded; (32) If the estimated cost of the optimal edge to be expanded is less than the current path cost, proceed to step (33); otherwise, clear the queue and end this iteration; (33) If the optimal edge to be expanded has no collision and can reduce the cost, then go to step (34); otherwise, clear the queue and end this iteration; (34) If the terminal node of the optimal edge to be expanded is in the current tree node list, the node is rewired and the process goes to step (35); otherwise, the terminal node is moved from the sampling node list to the tree node set and the process goes to step (35); (35) Add the optimal edge to be expanded to the tree edge set and clean up the useless edges, and return to step (31) until the optimal edge to be expanded cannot reduce the current path cost, and the initial path is determined.

4. The UAV path planning method based on batch processing prior knowledge tree according to claim 1 is characterized in that: In step (4), the major axis of the ellipse is the current optimal path length , the short axis is , is the straight line length between the starting point and the destination point.

5. The UAV path planning method based on batch processing prior knowledge tree according to claim 1 is characterized in that: In step (5), the obstacle density factor is introduced, and the EXP3 algorithm is used to determine the beacon point from the updated sampling point set obtained in step (4), as follows: (51) The updated sampling point set obtained in step (4) , initialize the weight of each node in the node set to 1; (52) Calculation Moment The probability of selecting a node as a beacon ; (53) Based on the selection probability of the beacon selected by the EXP3 algorithm, a node is selected as the beacon according to the selection probability of the beacon , and introduce the obstacle density factor to calculate the beacon Rewards , then normalize it to get the normalized reward ; (54) Update the beacon weight and obtain the updated beacon weight ; (55) Repeat steps (52) to (54) until the predetermined time is reached and the final beacon is generated. .

6. The UAV path planning method based on batch processing prior knowledge tree according to claim 5 is characterized in that: Moment The probability of selecting a node as a beacon for ； in, is the current beacon weight, is the number of sampling points.

7. The UAV path planning method based on batch processing prior knowledge tree according to claim 6 is characterized in that: Beacon Rewards for ； ； ； in, and is the weight parameter, From beacon To the destination The potential path cost, is the density of obstacles; A is based on beacon and target point The midpoint of is the center and the radius r is The area of ​​the circular region; is the indicator function; Normalized Reward for ； in, is the absolute value of the reward.

8. The UAV path planning method based on batch processing prior knowledge tree according to claim 7 is characterized in that: Updated beacon weight for ； in, is the learning rate, .

9. The UAV path planning method based on batch processing prior knowledge tree according to claim 1 is characterized in that: Batch sampling is performed within the ellipse, and the sampling points The calculation equation is as follows: ； in, is the scaling factor, , for A random number uniformly distributed on The sampling point Perform inverse rotation and inverse translation operations in sequence to obtain the coordinates of the sampling point in the original coordinate system.

10. The UAV path planning method based on batch processing prior knowledge tree according to claim 1 is characterized in that: In step (11), the third-order B-spline curve is ； in, A collection of time, is a fourth-order matrix, is a set of control points.

Images