Protruding area coverage path planning method and system for autonomous sweeper

Abstract

The invention discloses a convex area coverage path planning method and system for an autonomous sweeper, which are applied to the technical field of path planning of autonomous mobile robots, and the method comprises the steps: generating an area coverage parallel track through a self-adaptive direction track generation algorithm based on a rotary caliper; the method comprises the following steps: constructing a GTSP model based on a region coverage parallel track, and standardizing the GTSP model into an ATSP model by adopting Noon-Bean conversion; and solving the ATSP model by adopting an LKH-VNS hybrid search algorithm to obtain a path planning result. According to the method, efficient and intelligent path planning is realized through adaptive direction orbit generation, standardized modeling from the GTSP to the ATSP and collaborative optimization of the LKH-VNS hybrid search algorithm.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of autonomous mobile robot path planning, and more particularly to a convex region coverage path planning method and system for an autonomous cleaning vehicle. BACKGROUND

[0002] Coverage path planning (CPP) is a key technology widely applied in intelligent equipment such as autonomous cleaning vehicles and agricultural robots, and its goal is to achieve efficient and non-missing coverage of the target region under the premise of meeting the kinematic constraints of the vehicle. Existing coverage path planning methods mainly include two categories: fixed direction decomposition-based and graph theory modeling-based.

[0003] The first category of methods is based on the path planning strategy of fixed direction decomposition. Typically, the target region is first decomposed according to a preset direction (such as the horizontal or vertical axis), and parallel tracks with a spacing of the vehicle's working width w are generated through Boustrophedon decomposition or trapezoidal decomposition. The vehicle travels through the tracks in spatial order, and performs right-angle turns or 180° turn-around turns between adjacent tracks. This method ensures coverage integrity through fixed-direction track generation, but does not consider the influence of regional geometric features on the optimal scanning direction.

[0004] The second category of methods directly models the coverage path planning problem as a generalized traveling salesman problem (GTSP) or an asymmetric traveling salesman problem (ATSP). In this category of methods, the region is first decomposed into several tracks, and the start and end points of each track are taken as a node pair. All node pairs are divided into different clusters. By constructing a distance or cost matrix between nodes, meta-heuristic methods such as genetic algorithms and ant colony algorithms are used to optimize the order of track visits to reduce non-working travel distance and turn frequency. Some methods also use Noon-Bean conversion to standardize GTSP into ATSP, and introduce Dubins path model to consider kinematic constraints such as minimum turning radius, and use LKH solver to optimize the visit sequence.

[0005] Existing coverage path planning methods have significant limitations in engineering applications, mainly in the following aspects: First, the track generation method based on fixed direction decomposition does not fully consider the geometric features of the target region. The track direction is usually preset manually (such as the horizontal or vertical axis), and lacks adaptability. In irregular or complex regions, this fixed direction strategy leads to a significant increase in the number of tracks and turns, and a substantial increase in path length, affecting the efficiency of the operation. This method assumes that the vehicle can instantaneously change its heading, and ignores kinematic constraints such as minimum turning radius, resulting in significant trajectory deviations in actual execution.

[0006] Secondly, the method of directly modeling GTSP or ATSP can utilize the optimization tools of graph theory, but it also relies on fixed direction in the track generation stage, and the Euclidean distance is mostly used in the calculation of the distance between nodes, which fails to accurately reflect the actual kinematic constraints of the vehicle. Specifically, the modeling process passively accepts the input of fixed direction tracks and cannot dynamically adjust the scanning direction according to the path optimization requirements; in the calculation of the kinematic cost between nodes, the heading angle parameter is simplified as a fixed value, which fails to accurately reflect the continuous change characteristics of the heading during the turning process of the vehicle, resulting in a large deviation between the theoretically optimal solution and the actually feasible path.

[0007] In addition, the existing methods mostly use a single meta-heuristic algorithm (such as genetic algorithm, ant colony algorithm, etc.), which is prone to fall into local optimum, slow convergence, complex parameter tuning, and difficult to balance the quality of the solution and the computational efficiency when dealing with large-scale or high-complexity problems. The solution quality of the meta-heuristic algorithm fluctuates ± 7.2% due to parameter sensitivity, and the solution cost of the special solver is still 3.9% higher than the global optimum, although it responds quickly (such as LKH solver can complete 38 track solving in 0.89 seconds).

[0008] The deeper contradiction lies in the fragmentation of technical modules. The adaptive track generation technology (such as the rotating caliper algorithm) only focuses on the optimization of geometric parameters and does not form a closed loop with the path sequence decision; and the path optimization algorithm is limited by the preset track direction, making it difficult to achieve global coordination.

[0009] In summary, the existing regional coverage path planning method has the technical problems of fixed track generation direction, inaccurate handling of kinematic constraints, and insufficient global optimization capability, which makes it difficult to meet the efficient and intelligent coverage requirements of the autonomous cleaning vehicle in complex or large-scale regions.

[0010] Therefore, how to provide an autonomous cleaning vehicle convex region coverage path planning method and system that can effectively overcome the core defects of the existing regional coverage path planning method, such as the fragmentation of technical modules, the fixed direction of track generation, the inaccurate handling of kinematic constraints, and the insufficient global optimization capability, achieve efficient and non-missing coverage of any convex region, significantly reduce the number of tracks and turning times, optimize the path length, and improve the working efficiency of the autonomous cleaning vehicle is a problem that needs to be solved by those skilled in the art. SUMMARY

[0011] Therefore, the present application provides an autonomous cleaning vehicle convex region coverage path planning method and system. Through the cooperative optimization of adaptive direction track generation, GTSP to ATSP standardized modeling, and LKH-VNS hybrid search algorithm, efficient and intelligent path planning is achieved.

[0012] In order to achieve the above purpose, the present application adopts the following technical solutions: A convex region coverage path planning method for autonomous cleaning vehicles, comprising: Step 1: generating region coverage parallel tracks by a self-adaptive direction track generation algorithm based on a rotary caliper; Step 2: constructing a GTSP model based on the region coverage parallel tracks, and standardizing the GTSP model into an ATSP model by Noon-Bean conversion; Step 3: solving the ATSP model by a LKH-VNS hybrid search algorithm to obtain a path planning result; wherein the LKH-VNS hybrid search algorithm firstly calls a LKH solver to solve the ATSP model for a high-quality initial solution by a Lin-Kernighan heuristic mechanism and self-adaptive parameter tuning capability, then performs systematic global optimization and search based on the high-quality initial solution provided by the LKH, and finally compares the high-quality initial solution of the LKH and the optimized solution of the VNS, and selects the solution with lower cost as the final output.

[0013] Optionally, in step 1, the region coverage parallel tracks are generated by a self-adaptive direction track generation algorithm based on a rotary caliper, specifically as follows: calculating all pairs of antipodal points of the convex polygon by a rotary caliper algorithm ; wherein, is the number of pairs of antipodal points; for each pair of antipodal points, calculating two candidate scanning directions as follows:

[0014] wherein, , are respectively , are respectively

[0015] for each candidate direction , calculating the projection width of the polygon in the direction as follows:

[0016] wherein, denotes the projection of the polygon vertex in the direction ; selecting the direction with the minimum projection width as the optimal scanning direction from the two candidate scanning directions of each pair of antipodal points as follows:

[0017] following the optimal scanning direction Parallel tracks are generated with a distance w; wherein the distance w is the vehicle coverage width; rotating the polygon in the opposite direction by an angle to align the track direction with the coordinate axis; Generating vertical scan lines along the x-axis direction in the rotated coordinate system with a distance w; Calculating the intersection of the scan lines and the polygon boundary to form track segments; Restoring the track coordinates to the original coordinate system through inverse rotation transformation.

[0018] Optionally, in step 2, based on the area coverage parallel track, a GTSP model is constructed, specifically: The starting point and the ending point of each parallel track form a cluster ; wherein the starting point represents a node driving along the track in the positive direction in the cluster; the ending point represents a node driving along the track in the opposite direction in the cluster; GTSP is defined on a directed graph ; wherein, is a set of nodes, divided into m clusters , , …, , each cluster corresponds to a track; is a set of arcs; For each arc , the cost is defined as the Dubins distance from node to node , as follows:

[0019] wherein, represents the cost from node to node ; represents the Dubins distance from node to node ; represents the cost between nodes in the same cluster, set to the maximum positive number to avoid vehicles driving in both directions on the same track; is a cluster identification function, used to output the cluster to which the node belongs.

[0020] Optionally, the cost is defined as the Dubins distance from node to node , specifically: Modeling the autonomous cleaning vehicle as a Dubins vehicle, the Dubins vehicle is configured with Description, its dynamics described as follows:

[0021] wherein, is the vehicle speed; is the minimum turning radius; is the control input; is the heading angle of the Dubins vehicle in the plane coordinate system; The degree of maneuverability of the vehicle is defined as the number of tracks crossed by the vehicle during a turn ; wherein i is the track number at the start of the turn, and j is the track number at the end of the turn; the circular turn and the lemon turn corresponding to and in the Dubins path respectively are selected as the turning mode of the vehicle in the area coverage mode; wherein, and are the circular turn and the lemon turn respectively; The minimum length of maneuverability is as follows:

[0022] wherein, is the minimum length of maneuverability of the lemon turn; is the minimum length of maneuverability of the circular turn; is the number of tracks crossed by the vehicle; is the coverage width of the vehicle; The minimum length of vehicle maneuverability is as follows:

[0023] wherein, is the minimum length of vehicle maneuverability, i.e. the Dubins distance between nodes.

[0024] Optionally, in step 2, the GTSP model is standardized into the ATSP model by using Noon-Bean conversion, specifically: For each cluster in the directed graph , a sub-node chain is generated, the number of nodes in the chain is consistent with the number of nodes in the cluster; wherein, corresponds to the node for driving in the positive direction of the track; corresponds to the node for driving in the reverse direction of the track; The arc weight of the auxiliary graph is defined, all forward arcs in the sub-node chain in the cluster are set to 0, forcing the ATSP path to pass through all nodes in the chain; the inter-cluster arcs are mapped to , weight inheritance original GTSP between corresponding nodes cost ; Intercept each sub-chain first node mapping back to the solution of the original GTSP.

[0025] Optionally, in step 3, the VNS algorithm provides a high-quality initial solution based on LKH, and performs systematic global optimization and search, specifically: Swing phase: randomly select a solution S in the k-th neighborhood , generate a new solution S shake by a random disturbance mechanism to jump out of the local neighborhood of the current solution; the swing operation adopts a random 2-opt strategy, randomly selects two edges to reconnect, and the disturbance strength is controlled by the neighborhood parameter k; Local search phase: local optimization of the solution S shake generated by swing, find the local optimal solution S by deterministic neighborhood search; the local search adopts the best improvement strategy, and systematically searches for improved solutions in the 2-opt neighborhood until no better neighbor is found; Neighborhood transformation phase: according to the local search result, decide whether to accept the new solution and adjust the neighborhood structure, if better than the current optimal solution, accept the new solution and reset the neighborhood parameter k=1; otherwise, increase the neighborhood size k=k+1, and reset to 1 when k exceeds the maximum neighborhood n.

[0026] The application also provides a convex region coverage path planning system for an autonomous cleaning vehicle using a convex region coverage path planning method for the autonomous cleaning vehicle, comprising: Track generation module: used for generating region coverage parallel tracks by an adaptive direction track generation algorithm based on a rotary caliper; GTSP model construction and conversion module: used for constructing a GTSP model based on the region coverage parallel tracks, and standardizing the GTSP model into an ATSP model by Noon-Bean conversion; Path planning module: used for solving the ATSP model by using an LKH-VNS hybrid search algorithm to obtain a path planning result.

[0027] Compared with the prior art, the application discloses a convex region coverage path planning method and system for an autonomous cleaning vehicle. Through the collaborative design of adaptive direction track generation and path sequence optimization, efficient and non-missing coverage of any convex region is realized. Specifically, the optimal scanning direction is automatically determined based on a rotary caliper algorithm, the number of tracks and the number of turns are significantly reduced, the path length is optimized, the working efficiency of the autonomous cleaning vehicle is improved, and the problems of low track layout efficiency and redundant turning paths in the prior art are solved. Through the standardized modeling of GTSP to ATSP and the collaborative optimization of the LKH-VNS hybrid search algorithm, the intelligent level and the calculation efficiency of the path planning are improved. The Noon-Bean conversion method is adopted to fully utilize the advantages of mature ATSP solving algorithms, the fast convergence characteristics of the LKH solver and the global search ability of the VNS algorithm are combined, the calculation efficiency is significantly improved while ensuring the solution quality, and the contradiction between efficiency and quality in the prior art is solved. Further, through the accurate modeling of the Dubins vehicle model, it is ensured that the generated coverage path can accurately reflect the actual kinematic constraints of the vehicle, especially the minimum turning radius and the continuous change characteristics of the heading, and the deviation between the theoretically optimal solution and the actual feasible path in the prior art is solved, and the feasibility and safety of the path are improved. BRIEF DESCRIPTION OF DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only constitute the embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of the provided drawings.

[0029] Figure 1 The method flowchart provided by the present application.

[0030] Fig. 2(a) is a schematic diagram of a pair of heel point pair (j, m) in the present application.

[0031] Fig. 2(b) is a schematic diagram of a pair of heel point pair (j, m) in the present application. j, ).

[0032] Figure 3 The track coverage direction diagram of the present application.

[0033] Fig. 4(a) is a schematic diagram of a given set of parallel field tracks (dashed lines) in the GTSP node representation of the present application.

[0034] Figure 4(b) is a schematic diagram of each trajectory having two directed path options (dashed line, SP: start point, EP: end point) in the GTSP node representation of the present invention.

[0035] Figure 5 This is a schematic diagram of the GTSP node representation and two feasible GTSP solutions (red and black) in the GTSP node representation of the present invention.

[0036] Figure 6(a) is a schematic diagram of the bow-shaped turn of the present invention.

[0037] Figure 6(b) is a schematic diagram of the pear-shaped turn of the present invention.

[0038] Figure 7 This is a schematic diagram of the ATSP node of the present invention.

[0039] Figure 8 This is a schematic diagram of the LKH-VNS hybrid search algorithm of the present invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] Example 1: Embodiment 1 of this invention discloses a method for planning the coverage path of an autonomous sweeper in a convex area, such as... Figure 1 As shown, it includes: The coverage problem of the autonomous sweeper of this invention can be described as follows: The autonomous sweeper starts from a point on the boundary of a region and performs coverage path planning for a given task area, and then reaches the end point after completion. The task objective is to fully cover the task area and minimize the cumulative length of non-working travel, under the constraint that the autonomous sweeper needs to complete turning within a limited turning radius due to the limitations of its chassis structure and drive mode.

[0042] Since turning is often costly, reducing turning costs typically improves work efficiency. This study divides the region into several tracks, considers the minimum turning radius of vehicles to skip adjacent tracks as much as possible, and expresses track traversal as a weighted graph traversal. The problem of finding the optimal traversal order is equivalent to finding the shortest path in the graph, also known as the Traveling Salesman Problem (TSP), where each region track is considered a node in the TSP. The travel cost between tracks is related to the degree of maneuverability, and after constructing the cost matrix, the LKH-VNS hybrid search algorithm is used to solve it.

[0043] The task area in the actual application scenario is complex and varied. In order to simplify the model and focus on the core problem, the following reasonable assumptions are made: Assumption 1: The actual size and geometry of the autonomous cleaning vehicle are ignored, and it is regarded as a point. It is assumed that the speed is constant during driving, and the height and pitch angle remain stable; Assumption 2: The cleaning area is simplified as a static convex polygon, and the interference of small undulations and temporary obstacles (such as pedestrians and vehicles) on the ground is ignored; Assumption 3: The trajectory deviation of the cleaning vehicle during turning and straight driving is within the allowable range, and the influence of dynamic disturbances such as tire slip and mechanical vibration on the path tracking accuracy is temporarily ignored; Assumption 4: The battery capacity or fuel reserve of the cleaning vehicle is sufficient to support the completion of the entire coverage task, and it does not need to be charged or refueled during the task execution. The influence of battery capacity change on vehicle performance is ignored; Assumption 5: The weather conditions during the operation of the cleaning vehicle are good, and there is no influence of adverse weather such as rain, snow and strong wind. The ground is dry and flat, the friction coefficient is stable, and there is no road condition that affects the normal driving of the vehicle such as water accumulation and oil stains.

[0044] Step 1: Generate regional coverage parallel tracks through the adaptive direction track generation algorithm based on the rotary caliper.

[0045] A given cleaning area is covered by a set of parallel tracks, which start from one boundary of the area and end at the opposite side. It is assumed that the area will be covered by a vehicle with an effective working width w using the BF mode. Let S = {S1, S2, S3,..., Sm} be the region track number in any order (for example, from left to right), and the way the tracks completely cover the region is defined by the driving direction of the vehicle and the working width of the vehicle. The total number of tracks covering the entire working area is represented by the cardinality of the set N, which satisfies the inequality:

[0046] where the symbol represents the floor function; is the minimum length of vehicle maneuvering; is the vehicle coverage width.

[0047] Since the driving distance and energy consumption of the vehicle increase with the increase of the number of turns, the driving direction with the minimum number of turns in the region must be found. A support line is defined as a line L that intersects the boundary of a polygon and is completely located on one side of the polygon. A heel point pair is a pair of parallel support lines that are tangent to the vertices of a convex polygon, as shown in FIG. 2(a) and FIG. 2(b). FIG. 2(a) shows a heel point pair (j, m), and FIG. 2(b) shows a heel point pair (j, m1 j, m2 L, M are a pair of parallel support lines. Draw two parallel support lines in the same polygon, the distance between them is called the span or width of the polygon, and the optimal coverage direction is perpendicular to the minimum width of the polygon.

[0048] Generate regional coverage parallel tracks by adaptive direction track generation algorithm based on rotary caliper, specifically: Calculate all pairs of heel points of convex polygons using the rotary caliper algorithm ; wherein, is the number of pairs of heel points; For each pair of heel points, calculate two candidate scanning directions as follows:

[0049] wherein, , are the corresponding horizontal and vertical coordinates, respectively; , are the corresponding horizontal and vertical coordinates, respectively; , are the corresponding horizontal and vertical coordinates, respectively; For each candidate direction , calculate the projection width of the polygon in that direction as follows:

[0050] wherein, denotes the projection of polygon vertex in direction ; From the two candidate scanning directions of each pair of heel points, select the direction with the smallest projection width as the optimal scanning direction as follows:

[0051] Generate parallel tracks in the optimal scanning direction with a spacing w; wherein, the spacing w is the vehicle coverage width; Rotate the polygon in the opposite direction by an angle so that the track direction is aligned with the coordinate axis; Generate vertical scanning lines in the x-axis direction in the rotated coordinate system with a spacing w; Calculate the intersection points of the scanning lines and the polygon boundary to form track segments; Restore the track coordinates to the original coordinate system by inverse rotation transformation.

[0052] As Figure 3The optimal driving direction automatically generated for the random polygon region, i.e., the covering track of the region, is shown.

[0053] Step 2: Based on the region covering parallel tracks, a GTSP model is constructed, and Noon-Bean transformation is used to standardize the GTSP model into an ATSP model.

[0054] Based on the region covering parallel tracks, a GTSP model is constructed, specifically: As can be seen from the above, the region is decomposed into parallel tracks , the start point and the end point of each parallel track form a cluster ; wherein the start point in the cluster represents the node along which the vehicle drives in the positive direction; the end point in the cluster represents the node along which the vehicle drives in the negative direction; GTSP is defined on a directed graph ; wherein is a set of nodes, which is divided into m clusters , , …, , each cluster corresponds to a track; is a set of arcs; the goal of GTSP is to find a minimum cost cycle containing exactly one node from each cluster; As shown in FIG. 4(a) and FIG. 4(b), the GTSP node representation, wherein, as shown in FIG. 4(a), a given set of parallel field tracks (dashed lines) is represented, and FIG. 4(b) represents each track with two directed path options (dashed lines, SP: start point, EP: end point). According to the above definition of GTSP, S = {S1, S2, S3, …, Sm} is the arbitrary ordered region track number (for example, from left to right), i.e., an arbitrary ordered set of field tracks. Each field track is composed of two end points and a line segment. The vehicle can cover the field track from any one end point to the other end point, so each track has two directional path options. This algorithm regards each track as a cluster of two nodes. Each node represents a directed path of the track on which the vehicle can drive. The two directed paths in the same cluster are parallel to the corresponding actual track, but in opposite directions. The node set of the corresponding GTSP can be written as N = {N1, N2, N3, …, Nm} in the range. The node set is divided into m clusters, and each cluster contains exactly two nodes. An example of node representation is shown in Figure 5The Dubins path is generated by a set of configurations. For each node, the start point (SP) is the point where the vehicle enters the track and the end point (EP) is the point where the vehicle leaves the track. The Dubins car's heading at the start and end points is equal to the direction of the corresponding directed path, so each node contains two configurations: the SP configuration and the EP configuration, from the node notation.

[0055] Two nodes in the same cluster should never be connected together, because the vehicle should not cross the same track in two different directions. Therefore, the cost between nodes in the same cluster is set to where is a large positive number. Nodes belonging to different clusters are connected by Dubins paths, which are always computed from the EP configuration of one node to the SP configuration of the other node. For each arc the cost is defined as the Dubins distance from node to node , as follows:

[0056] where denotes the cost from node to node ; denotes the Dubins distance from node to node ; denotes the inter-cluster node cost, which is set to a large positive number to avoid the vehicle driving in the same track in two different directions; is the cluster identification function, which outputs the cluster to which a node belongs.

[0057] The Dubins distances and may be different, meaning that "the distance of the vehicle moving from track i to track j" and "the distance of the vehicle moving from track j to track i" are different, which is a kind of asymmetry. After computing the cost of each pair of nodes, the cost matrix of the GTSP is constructed. The cost matrix is actually described by the non-working distances, which means that the formation of the GTSP is to minimize the total non-working driving distance during the turns. With the above cost expression, there is no constraint that the consecutive end and start points must be located on the same boundary or the same side of the track, which has no impact on the feasibility of the solution but gives the algorithm more freedom to find the optimal solution.

[0058] The cost is defined as the Dubins distance from node to node , specifically: The autonomous cleaning vehicle is modeled as a Dubins vehicle, which is a vehicle that can only move forward at a constant speed and turn in a plane with an upper bound on the curvature (or lower bound on the turning radius). The Dubins vehicle configuration is denoted by The dynamics of the Dubins vehicle are described as follows:

[0059] where, is the vehicle velocity; is the minimum turning radius; is the control input; is the heading angle of the Dubins vehicle in a plane coordinate system; The study of Dubins provides a sufficient path family that makes it possible to compute the optimal path between any two configurations. This sufficient path family consists of six elements, which contain the optimal path. Since there is a smaller sufficient path family, the length of each element can be computed and the shortest path is selected. The main result shows that the optimal path is contained in a finite number of Dubins paths. This reduces the problem of determining the shortest path between any two configurations by only checking six path types, which are divided into two series: Series CCC: , ; Series CSC: , , , ; where C is a curve or arc segment with a radius of S is a straight line segment, L is a left turn with a minimum radius, and R is a right turn with a minimum radius; , , is the length of the corresponding path segment.

[0060] The maneuverability of the vehicle is defined as the number of tracks that the vehicle crosses during steering ; where i is the track number at the start of steering, and j is the track number at the end of steering; as shown in FIG. 6(a), FIG. 6(b), the turning mode commonly used by the vehicle in the area coverage mode is selected, which corresponds to the bow turning and the pear turning in the Dubins path and respectively; where and are the bow turning and the pear turning respectively; FIG. 6(a) represents the bow turning, and FIG. 6(b) represents the pear turning; r is the turning radius of the vehicle; W = zw; The minimum length of maneuverability is as follows:

[0061] in, The minimum maneuver length for a pear-shaped turn; The minimum maneuvering length for a bow-shaped turn; The number of tracks that the vehicle crosses; For vehicle coverage width; The minimum length for vehicle maneuvering is as follows:

[0062] in, The minimum distance for vehicle maneuvering is the Dubins distance between nodes.

[0063] The Noon-Bean transformation is used to standardize the GTSP model to the ATSP model, such as... Figure 7 As shown ( Figure 7 The blue and black colors represent the intra-cluster arcs and inter-cluster arcs of an ATSP node, respectively. For directed graphs Each cluster in Generate a chain of child nodes The number of nodes within the chain is the same as the number of nodes within the cluster; among them, Corresponding to the forward travel node on the track; Corresponding reverse travel node on the track; Define auxiliary graph Arc weights, all forward arcs in the cluster's child node chain The weight is set to 0, forcing the ATSP path to traverse all nodes in the chain; inter-cluster arcs Mapped to The weights inherit the corresponding inter-node costs from the original GTSP. ; Through this construction, the optimal path of ATSP is found in the auxiliary graph. The solution must exist in the form of a cluster chain traversal, and the first node of each sub-chain is extracted and mapped back to the original GTSP solution.

[0064] This method ensures that every cluster is visited by the ATSP path through a zero-weight chain structure within each cluster, and that the transformed solution is equivalent to the GTSP global optimal solution. Mathematically, let the total cost of the original GTSP optimal solution be... The total cost of the ATSP solution after conversion is Since there is no additional cost added to the arc within the cluster, we can obtain , thus ensuring the consistency of the solution. In addition, the asymmetric nature of ATSP can more accurately depict the impact of sweeper turning motion on path cost (such as the distance difference between clockwise and counterclockwise turns), further improving the adaptability of the model to actual constraints. Through the above conversion, the coverage path planning problem can be efficiently solved by using mature ATSP algorithms, laying a foundation for the design of the subsequent LKH-VNS hybrid search algorithm.

[0065] Step 3: Solve the ATSP model using the LKH-VNS hybrid search algorithm to obtain the path planning result.

[0066] Since the ATSP problem is considered an NP-hard problem, which refers to a class of computational problems whose given solutions can be verified in polynomial time, but finding the optimal solution is extremely difficult, and there is currently no known polynomial time algorithm. Therefore, a heuristic algorithm can be used. Based on the complementary characteristics of VNS and LKH algorithms, the present application proposes a LKH-VNS collaborative optimization hybrid algorithm, aiming to combine the fast convergence ability of LKH solver with the global search advantage of VNS algorithm. This hybrid strategy follows the design principle of "fast start-global optimization-quality guarantee", and realizes the dual improvement of calculation efficiency and solution quality through three core stages.

[0067] LKH algorithm, as one of the best TSP exact solution algorithms currently, has extremely fast calculation speed and strong local optimization ability, and can converge to high-quality solutions in a short time. However, LKH is based on local search mechanism and is prone to fall into local optimum in complex search space. VNS algorithm has strong global search ability through systematic neighborhood transformation strategy, but its solving efficiency is largely dependent on the quality of the initial solution - when starting from a poor random initial solution, the algorithm needs to perform a large amount of invalid exploration to converge to a high-quality region.

[0068] Therefore, the LKH-VNS hybrid search algorithm fully utilizes the advantages of the two algorithms: using the fast convergence characteristics of LKH to obtain a high-quality initial solution in a short time, providing an excellent search starting point for VNS; and then using the variable neighborhood search mechanism of VNS to perform global optimization based on the starting point, avoiding the long convergence process from poor solutions, and realizing effective global search.

[0069] Solve the ATSP model using the LKH-VNS hybrid search algorithm, as shown in Figure 8 , specifically: Call the LKH solver to solve the ATSP model to obtain a high-quality initial solution close to the optimal solution in a very short time (usually 0.3-0.5 seconds) through the Lin-Kernighan heuristic mechanism and adaptive parameter tuning ability, denoted as S lkh , (initialize the global optimal solution S best to Slkh ); the LKH solver can quickly converge to the high-quality area in the search space through k-opt neighborhood operation and candidate set strategy, providing a good starting point for subsequent VNS optimization. The key of this stage is to fully utilize the computational efficiency advantage of LKH and avoid the cold start problem of VNS algorithm starting from a random solution. The initial solution generated by LKH not only has a lower cost, but more importantly, its solution structure is usually located in the high-quality area of the search space, providing a favorable exploration basis for the variable neighborhood search of VNS; Based on the high-quality initial solution provided by LKH (as the current solution S curr of VNS algorithm), the VNS algorithm performs systematic global optimization and search as follows: Initialization: set the maximum number of iterations MAXGEN, the number of searches in each field M, the number of field structures n, and initialize the iteration counter gen=1 and the field index k=1; Swing phase: randomly select a solution in the kth neighborhood , jump out of the local neighborhood of the current solution through a random disturbance mechanism to generate a new solution S shake ; the swing operation adopts a random 2-opt strategy, randomly selects two edges to reconnect, and the disturbance strength is controlled by the neighborhood parameter k; Local search phase: locally optimize the solution S shake generated by swing, and find the local optimal solution through deterministic neighborhood search; the local search adopts the best improvement strategy to systematically find improved solutions in the 2-opt neighborhood until no better neighbor is found; Neighborhood transformation phase: decide whether to accept the new solution and adjust the neighborhood structure according to the local search results. If is better than the current optimal solution, accept the new solution and reset the neighborhood parameter k=1; otherwise, increase the neighborhood size k=k+1, and reset it to 1 when k exceeds the maximum neighborhood n; The final stage of the algorithm compares the LKH high-quality initial solution and the VNS optimized solution, and selects the solution with lower cost as the final output. This double protection mechanism ensures the robustness of the algorithm: even if VNS fails to achieve further improvement based on the LKH solution, the algorithm can still guarantee at least the solution quality level of LKH.

[0070] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between embodiments can be referred to each other. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the related parts can be referred to the method part.

[0071] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and that the appended claims are intended to cover all such modifications that do not depart from the true spirit and scope of the application. Therefore, the application is not limited to the embodiments shown but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for planning a path to cover a convex area using an autonomous sweeper, characterized in that, include: Step 1: Generate parallel tracks covering the region using an adaptive orientation track generation algorithm based on rotating calipers; Step 2: Based on the parallel orbits covered by the region, construct the GTSP model and use the Noon-Bean transformation to normalize the GTSP model into the ATSP model; Step 3: Solve the ATSP model using the LKH-VNS hybrid search algorithm to obtain the path planning results. The LKH-VNS hybrid search algorithm first calls the LKH solver to obtain a high-quality initial solution for the ATSP model through the Lin-Kernighan heuristic mechanism and adaptive parameter tuning capability. Then, the VNS algorithm performs systematic global optimization and search based on the high-quality initial solution provided by LKH. Finally, the high-quality initial solution of LKH and the optimized solution of VNS are compared, and the solution with lower cost is selected as the final output.

2. The method for planning a convex area coverage path for an autonomous sweeper according to claim 1, characterized in that, In step 1, an adaptive orientation track generation algorithm based on rotating calipers is used to generate parallel tracks covering the region, specifically: Calculate all pairs of heel points of a convex polygon using the rotating caliper algorithm. ;in, This refers to the number of heel points; For each pair of heel points, two candidate scanning directions are calculated as follows: in, , They are respectively The corresponding x and y coordinates; , They are respectively The corresponding x and y coordinates; For each candidate direction Calculate the projected width of the polygon in this direction. ,as follows: in, Represents the vertices of a polygon In direction Projection on; For each pair of heel points, select the direction with the smallest projected width from the two candidate scanning directions as the optimal scanning direction. ,as follows: Along the optimal scanning direction Parallel tracks are generated with a spacing w; where the spacing w is the vehicle coverage width. Rotate the polygon in the opposite direction Angle, to align the orbital direction with the coordinate axes; In a rotating coordinate system, vertical scan lines are generated along the x-axis with a spacing w. Calculate the intersection points of the scan line and the polygon boundary to form the track segment; The orbital coordinates are restored to the original coordinate system through inverse rotation transformation.

3. The method for planning a convex area coverage path for an autonomous sweeper according to claim 1, characterized in that, In step 2, based on the parallel orbits covered by the region, a GTSP model is constructed, specifically as follows: The starting point of each parallel track and the end point constituting a cluster ; Among them, the starting point In a cluster, it represents a node traveling in the positive direction along the track; the endpoint. In a cluster, it represents a node that travels in the opposite direction along the track; GTSP is defined in directed graphs Above; among them, It is a set of nodes, divided into m clusters. , , ..., Each cluster corresponds to one orbit; It is a set of arcs; For each arc ,cost Defined as a slave node To the node The Dubins distance is as follows: in, Represents a node To the node The cost; Represents a node To the node The distance to Dubins; This represents the cost between nodes in the same cluster and is set to the maximum positive number to prevent vehicles from traveling in both directions on the same track. This is a cluster identifier function used to output the cluster to which a node belongs.

4. The method for planning a convex area coverage path for an autonomous sweeper according to claim 3, characterized in that, cost Defined as a slave node To the node The Dubins distance is as follows: The autonomous sweeper is modeled as a Dubins vehicle, and the Dubins vehicle configuration is used... The description, its dynamic description, is as follows: in, For vehicle speed; Minimum turning radius; For control input; Let the heading angle of the Dubins vehicle in the planar coordinate system be denoted as ... denoted; The maneuverability of a vehicle is defined as the number of tracks it crosses during a turn. Where i is the track number at the start of the turn, and j is the track number at the end of the turn; select the corresponding Dubins path and Bow-shaped turns and pear-shaped turns are used as turning modes for vehicles in area coverage mode; among them, and These are respectively a bow-shaped turn and a pear-shaped turn; The minimum maneuver length is as follows: in, The minimum maneuver length for a pear-shaped turn; The minimum maneuvering length for a bow-shaped turn; The number of tracks that the vehicle crosses; For vehicle coverage width; The minimum length for vehicle maneuvering is as follows: in, The minimum distance for vehicle maneuvering is the Dubins distance between nodes.

5. The method for planning a convex area coverage path for an autonomous sweeper according to claim 1, characterized in that, In step 2, the GTSP model is standardized to the ATSP model using the Noon-Bean transformation, specifically as follows: For directed graphs Each cluster in Generate a chain of child nodes The number of nodes within the chain is the same as the number of nodes within the cluster; among them, Corresponding to the forward travel node on the track; Corresponding reverse travel node on the track; Define auxiliary graph Arc weights, all forward arcs in the cluster's child node chain The weight is set to 0, forcing the ATSP path to traverse all nodes in the chain; inter-cluster arcs Mapped to The weights inherit the corresponding inter-node costs from the original GTSP. ; Extract the solution from the first node of each subchain and map it back to the original GTSP.

6. The method for planning a convex area coverage path for an autonomous sweeper according to claim 1, characterized in that, In step 3, the VNS algorithm performs systematic global optimization and search based on the high-quality initial solution provided by LKH, specifically as follows: Swing phase: in the k-th neighborhood Randomly select solutions By using a random perturbation mechanism, a new solution S is generated by escaping the local neighborhood of the current solution. shake The swing operation employs a random 2-opt strategy, randomly selecting two edges for reconnection, with the perturbation intensity controlled by the neighborhood parameter k. Local search phase: Solving the solution S generated by the swing. shake Local optimization is performed, and a local optimum is found through deterministic neighborhood search. The local search employs an optimal improvement strategy, systematically searching for improved solutions in the 2-opt neighborhood until no better neighbors can be found. Neighborhood transformation stage: Based on the local search results, decide whether to accept the new solution and adjust the neighborhood structure. If the solution is better than the current optimal solution, accept the new solution and reset the neighborhood parameter k=1; otherwise, increase the neighborhood size k=k+1, and reset it to 1 when k exceeds the maximum neighborhood n.

7. An autonomous sweeper convex area coverage path planning system utilizing the autonomous sweeper convex area coverage path planning method according to any one of claims 1-6, characterized in that, include: Track generation module: Used to generate parallel tracks covering the region using an adaptive orientation track generation algorithm based on rotating calipers; GTSP model construction and conversion module: used to construct a GTSP model based on the parallel orbits covered in the region, and to standardize the GTSP model into an ATSP model using Noon-Bean conversion; Path planning module: Used to solve the ATSP model using the LKH-VNS hybrid search algorithm to obtain path planning results.

Images