Unmanned operation path planning method and system in irregular surface type operation scene

Abstract

The application belongs to the technical field of mobile robot path planning and intelligent operation control, and discloses an unmanned operation path planning method and system in an irregular surface operation scene, which comprises the following steps: acquiring operation area boundary, obstacles, start and end points and operation width information, performing inflation processing on the boundary and obstacles to obtain a safe feasible space; generating an initial sub-region based on triangulation, and obtaining a convex sub-region set through greedy convex merging; generating a coverage path candidate in the first two minimum width directions for each sub-region; adopting a method combining adaptive large neighborhood search, dynamic programming and simulated annealing to cooperatively optimize sub-region access order and the candidate path, and generating a continuous path through visual line connection to output a complete executable coverage path. The application can realize full coverage path planning with high coverage rate and low repetition rate in a complex irregular environment, and output a high-quality planning path.

Description

Technical Field

[0001] This invention belongs to the field of mobile robot path planning and intelligent operation control technology, and more specifically, relates to an unmanned operation path planning method and system in irregular surface operation scenarios. Background Technology

[0002] In service operations such as lawn mowing and ground cleaning in parks, traditional methods mainly rely on manual labor, which suffers from high labor intensity, low efficiency, and high costs. With the continuous development of service robot technology, unmanned robots with autonomous navigation and operation capabilities are gradually being applied to these scenarios, and full-coverage path planning has become one of the key technologies for achieving unmanned operations.

[0003] The goal of full-coverage path planning is to enable the robot to achieve complete coverage of the target area while minimizing redundant paths and operation time, all within the constraints of the environment and the robot's operational capabilities. However, in practical applications, the work area typically has irregular boundaries and contains obstacles of various shapes, exhibiting complex spatial structures. This irregular environment makes traditional coverage path planning methods designed for regular areas difficult to apply directly, easily leading to problems such as incomplete coverage, excessive path redundancy, and low operational efficiency. The resulting path is also unlikely to achieve global optimization, directly limiting the efficiency and output quality of path planning. Summary of the Invention

[0004] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides an unmanned operation path planning method and system for irregular surface operation scenarios, the purpose of which is to improve the efficiency and quality of path planning.

[0005] To achieve the above objectives, the present invention provides a method for unmanned operation path planning in irregular surface operation scenarios, comprising: The robot obtains operational map information, including the boundary of the operational area, obstacles within the operational area, operational width, planning start point S, and planning end point E. It then performs anti-collision expansion processing on the boundary of the operational area and obstacles to obtain the safe and feasible space F for robot operation. The safe and feasible space F is triangulated to generate an initial set of convex subregions T; a greedy convex merging algorithm is performed on the initial set of convex subregions T to obtain a set of convex subregions C; for any convex subregion C in the set of convex subregions C... i Each non-degenerate edge is taken as a candidate scan direction, and after normalizing the candidate scan directions, C is obtained. i The set of candidate scanning directions is determined; the normal projection width of each candidate scanning direction is calculated, and the two candidate scanning directions with the smallest normal projection width are selected as C. iTwo scanning directions; For each scanning direction of each convex sub-region, a reciprocating parallel scanning line corresponding to each convex sub-region is generated according to the working width. After the scanning line intersects with the corresponding convex sub-region, several working line segments are formed, and a set of candidate paths for the robot to perform covering operations along the working line segments in the corresponding convex sub-region is generated. With the constraint that the robot is visited once in each convex sub-region and only one candidate path is selected in each convex sub-region, and with the objective function being to minimize the total cost of the robot from the planning starting point S through all convex sub-regions to the planning endpoint E, a joint optimization model is constructed. Solving the joint optimization model yields the optimal planned path characterized by the access order of the convex sub-regions and the selected candidate paths within each convex sub-region, thus completing the robot's operation path planning.

[0006] Furthermore, the ALNS-DP-SA joint optimization algorithm is used to solve the joint optimization model, specifically including: Sep1. In the m-th iteration, a random destruction operator is selected based on the current weights of the destruction operators in the Adaptive Large Neighborhood Search (ALNS) to access the convex sub-regions in the current solution in the order π. m-1 Perform a destruction operation to remove a portion of the convex subregion and obtain a partial access order; the higher the weight of the destruction operator, the higher the probability of it being selected. Sep2. Based on the current weights of the repair operators in the Adaptive Large Neighborhood Search (ALNS), a repair operator is randomly selected to re-insert the removed convex sub-regions into the partial access order, resulting in the convex region access order π under the m-th iteration. m ; Sep3, access sequence π in the convex region m In this case, dynamic programming (DP) is used to select the candidate path combination with the minimum total cost for the robot to move from the planning starting point S, through all convex sub-regions and perform coverage operations, to the planning ending point E. ; Sep4, Find the complete solution under the m-th iteration. The corresponding total cost Total cost of the current optimal solution If a comparison is made, Then use the complete solution. Update the current solution and the current optimal solution; if Greater than or equal to However, if the cost is less than the total cost of the current solution, then the complete solution is used. Update the current solution while keeping the current optimal solution unchanged; otherwise, use the simulated annealing algorithm (SA) to determine whether to use the complete solution. Update the current solution: If the acceptance criteria for simulated annealing are met, then use the complete solution. Update the current solution and keep the current optimal solution unchanged. If the acceptance condition of simulated annealing is not met, keep the current solution and the current optimal solution unchanged. Update the current weights of the destruction operator and repair operator selected in the m-th iteration. Sep5: Determine whether the maximum number of iterations M has been reached, or whether the number of consecutive unimproved iterations of the current optimal solution has reached the threshold M0. If not, let m = m + 1 and return to Sep1 for the next round of iteration. If yes, output the current optimal solution, which is the optimal planning path characterized by the convex sub-region access order and the selected candidate path in each convex sub-region.

[0007] Furthermore, it also includes converting the optimal planned path into structured operation path data that the robot can execute. The structured operation path data Path consists of multiple path points arranged in the order of path execution, wherein the data structure P of the nth path point... n for: P n ={"index": n,"position": (x n , y n ), "behavior_type": type}; Where, "index": n represents the path point index, and "position": (x n , y n ) represents the two-dimensional coordinates of the path point, and "behavior_type": type represents the code that characterizes the robot behavior type when the current path point is reached. The robot behavior type includes straight-line operation, turning in place, moving from the current operation line to the next operation line within a convex sub-region, moving from one convex sub-region to another, and stopping.

[0008] Furthermore, the total cost for the robot to move from the planned starting point S, through all convex sub-regions and perform covering operations, to the planned ending point E is... for: ; In the formula, Indicates the access order of the convex subregion. , This represents the number of convex subregions in the set C of convex subregions; Indicates the access order in the convex subregion The selected candidate path sequence is as follows. , Indicates the first Candidate paths selected for each visited convex subregion. Indicates the first The accessed convex subregion number, Indicates the first The candidate path number selected for each visited convex subregion. ; This indicates that the robot starts from the planning stage. To candidate paths The cost at the entry point; Indicates the robot in the Along the candidate path within the visited convex subregion The cost of performing a coverage operation; This indicates that the robot selects candidate paths. The exit point is moved to the candidate path. The cost at the entry point; This indicates that the robot selects candidate paths. The exit point is moved to the planned endpoint. The cost; cost Indicates a path along a broken line The execution time is calculated as follows: ; In the formula, Indicates the robot on the polyline path The time it takes to move along the straight line segment. Indicates the length of a straight line segment. For polyline paths The number of points on it; Indicates the time it takes for the robot to perform a turning maneuver. For the robot on the polyline path Two consecutive straight line segments and Path points The absolute value of the steering angle at that location; and They are respectively: ; ; In the formula, For the robot's linear acceleration, This is the robot's maximum linear velocity. For the robot's angular acceleration, This represents the maximum angular velocity of the robot. Among them, the cost for , , or The corresponding polyline path Each robot starts from the planning point. To candidate paths The polygonal path at the entry point, the robot in the... Along the candidate path within the visited convex subregion Performing coverage operations using polyline paths, and the robot's path selection from candidate paths. The exit point is moved to the candidate path. The polyline path at the entry point, and the robot's path selection from candidate paths. The exit point is moved to the planned endpoint. The broken line path.

[0009] Further, a greedy convex merging algorithm is performed on the initial set of convex subregions T to obtain a set of convex subregions C, including: (1) Establish adjacency relationships for each convex sub-region in the current set of regions to be processed, and record two convex sub-regions with a common edge as adjacent regions, so as to obtain the current set of regions to be merged composed of adjacent regions in the current set of regions to be processed; wherein, the initial set of regions to be processed is the initial set of convex sub-regions T. (2) Traverse the adjacent regions in the current set of regions to be processed in sequence according to the principle of priority of shared edge length; if the common edge length of the adjacent regions is less than or equal to the preset geometric judgment threshold ε, then the merging of the adjacent regions is not performed; otherwise, the adjacent regions are taken as candidate merging objects, and the union region of the candidate merging objects is calculated. (3) Determine whether the boundary vertices of the union region can form a convex polygon. If yes, perform a merge on the candidate merge objects to generate a merged convex sub-region and replace the corresponding adjacent region with the merged convex sub-region. If no, do not perform a merge on the candidate merge objects. After traversal, obtain the updated set of current regions to be processed. (4) Determine whether a merged convex sub-region has been generated during this round of traversal. If yes, return to step (1) for the next round of iteration. If no, use the current set of regions to be processed as the set of convex sub-regions C.

[0010] Furthermore, a set of candidate paths is generated for the robot to perform covering operations along the work line segment within the corresponding convex sub-region, including: Using the end of the scan line closest to the anchor edge of the convex region as the entry point, perform reciprocating coverage along the current scan direction to generate a candidate path; Using the other end of the scan line closest to the anchor edge of the convex region as the entry point, perform reciprocating coverage along the current scan direction to generate a candidate path; Using the end of the scan line furthest from the anchor edge of the convex region as the entry point, perform reciprocating coverage along the current scan direction to generate a candidate path; Using the other end of the scan line furthest from the anchor edge of the convex region as the entry point, perform reciprocating coverage along the current scan direction to generate a candidate path; The above candidate paths constitute the candidate path set corresponding to the current scanning direction; The anchor edge is the edge in the convex sub-region in the current scanning direction.

[0011] Further, the normal projection width of each candidate scanning direction is calculated, including: For each candidate scanning direction, take the centroid of the corresponding convex sub-region as the rotation center, and rotate the vertex coordinates of the convex sub-region clockwise by θ, so that the corresponding edge of the candidate scanning direction is parallel to the x-axis of the map coordinate system. At this time, the y-direction of the map coordinate system is the normal direction of the candidate scanning direction. Calculate the ordinates of all vertices in the rotated convex sub-region, and take the difference between the maximum and minimum ordinates as the normal projection width corresponding to the candidate scan direction.

[0012] The present invention also provides an unmanned operation path planning system for irregular surface operation scenarios, including a computer-readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is used to read executable instructions stored in the computer-readable storage medium and execute the unmanned operation path planning method for irregular surface operation scenarios described above.

[0013] The present invention also provides a computer program product, including a computer program that, when the computer program is run on a computer, causes the computer to execute the unmanned operation path planning method for irregular surface operation scenarios described above.

[0014] The present invention also provides a robot for irregular surface operation scenarios, wherein the robot performs unmanned operation along an optimal planned path, and the optimal planned path is obtained by any of the above-described unmanned operation path planning methods for irregular surface operation scenarios.

[0015] In summary, the above-described technical solutions conceived in this invention can achieve the following beneficial effects: 1. The method of this invention first divides a complex working area with irregular boundaries and internal obstacles into initial convex sub-regions through triangulation, and then uses greedy convex merging to form a more regular set of convex sub-regions suitable for reciprocating parallel scanning. This method avoids the problems of directional instability and connection difficulties when generating coverage paths in non-convex regions, while reducing the number of sub-regions and frequent switching, reducing the probability of redundant path generation, and improving the adaptability of region decomposition and coverage path planning, thereby improving the overall path planning efficiency and execution continuity.

[0016] Furthermore, when determining the scanning direction, this invention considers that selecting only the scanning direction with the smallest normal projection width can only obtain a locally optimal result for the coverage cost within the convex sub-region. The total cost of a complete coverage path also includes the transfer cost between adjacent sub-regions (the cost of the robot moving from the exit point of one convex sub-region to the entry point of its adjacent convex sub-region). Different scanning directions will form different scanning line layouts and further generate coverage path candidates with different entry and exit points, thus affecting the transfer cost between regions. Therefore, this invention further retains the second smallest width direction based on the smallest width direction and generates corresponding coverage path candidates for each scanning direction. This provides multiple entry and exit combinations for the global optimization of the subsequent joint optimization model, enabling it to comprehensively balance the cost of coverage within the convex sub-region and the cost of transfer between regions, thereby obtaining a better overall path. Simultaneously, retaining too many scanning directions increases the number of coverage path candidates, significantly expanding the scale of subsequent joint optimization. Therefore, this invention selects the first two smallest width directions for each convex sub-region and generates corresponding coverage path candidates for each direction. This ensures candidate diversity while controlling the search scale, achieving a balance between path quality and computational efficiency in irregular surface operation scenarios.

[0017] 2. Furthermore, the method of this invention jointly optimizes the sub-region access order and candidate paths for each sub-region, fully considering the coupling between sub-problems. By adaptively perturbing the access order through large neighborhood search and using dynamic programming to decode the optimal path candidates under a fixed order, and then combining simulated annealing criteria to accept candidate solutions, the overall path is collaboratively optimized. This method ensures the effective connection between the paths within each sub-region and the global order, further improving the global optimality and execution continuity of path planning.

[0018] 3. Furthermore, after path planning is completed, this invention outputs structured path data including path point numbers, two-dimensional coordinates of the path points, and robot behavior type codes. This allows the path planning results to be directly integrated with the robot control system, enabling segmented execution and state switching control of the path. Simultaneously, this structured path data facilitates trajectory tracking, speed planning, and anomaly handling, thereby improving the controllability of path execution and system integration capabilities, and enhancing the engineering application effectiveness of the unmanned operation system.

[0019] 4. Furthermore, existing coverage path planning typically uses total path length, number of turns, or repetition coverage rate as optimization metrics. When optimizing multiple metrics simultaneously, a weighted summation method is often required, with each metric's weight manually set. Since these metrics have different dimensions, and the weight values ​​directly affect the optimization results, they are easily influenced by empirical judgment. This invention converts straight-line travel, turning, and inter-region transfers into execution time based on the robot's acceleration and deceleration motion model, unifying different motion costs into a time dimension. Each cost is directly calculated from path length, turning angle, and the robot's speed and acceleration parameters, eliminating the need to manually set weights between path length and number of turns. This reduces the subjectivity of weight selection and makes the optimization results more consistent with the robot's actual operating conditions.

[0020] In summary, this invention provides a highly efficient, fully-coverage path planning method that can balance region decomposition, local path generation, and global collaborative optimization for operation scenarios with irregular surface boundaries and complex obstacle distribution. This method improves the adaptability and operational efficiency of unmanned operation systems in complex environments and is applicable to autonomous lawn mowing robots and other service robot systems. Attached Figure Description

[0021] Figure 1 Flowchart of the unmanned operation path planning method for irregular surface operation scenarios provided in this embodiment of the invention; Figure 2 This is a schematic diagram of irregular surface path planning provided in an embodiment of the present invention; Figure 3 A schematic diagram of the constrained Delaunay triangulation results provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the convex sub-region after greedy convex merging provided in an embodiment of the present invention; Figure 5 A schematic diagram illustrating four entry methods of a convex polygon under the same scanning direction provided in an embodiment of the present invention; Figure 6 A flowchart of the ALNS-DP-SA collaborative optimization algorithm provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the path planning visualization results provided in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0023] In this invention, the terms "first," "second," etc., used in the invention and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0024] Example 1 This embodiment provides a method for unmanned operation path planning in irregular surface operation scenarios, and its overall process is as follows: Figure 1 As shown, it includes the following steps: S1. Obtain the robot's work map information in the map coordinate system, including the work area boundary, obstacle information within the work area, planning start point S, planning end point E, and work width parameters, and initialize the path planning parameters.

[0025] Taking the path planning of a lawnmower robot in a park as an example, the first step in the operation of the lawnmower robot is to construct a map of the park's operating environment. For example... Figure 2 As shown, the work area has irregular boundaries and contains multiple obstacles of different shapes. The robot performs a full-coverage path planning task in this environment. This map can be obtained through UAV mapping, LiDAR scanning, or a pre-established electronic map. The main information in the work map includes: work area boundaries, obstacle areas (representing obstacle information within the work area), the robot's initial work starting point, the robot's planned end point, and the robot's work width parameters.

[0026] The boundary of the work area is modeled using a closed polygon, denoted as: P={p1, p2, …, p z}, where each vertex p u =(x u , y u ), u∈{1,2,…,z}, where z is the total number of vertices on the boundary of the work area. These vertices are arranged in clockwise or counterclockwise order, forming a non-self-intersecting polygonal region. Obstacles are represented by a polygon set: O={O1, O2, …, O x}, where each obstacle O v It consists of several vertices, v∈{1,2,…,x}, all of which are located within the working area; the planning starting point is denoted as S=(x s , y s The planned endpoint is denoted as E=(x e , y e All the coordinates above are represented in the same relative coordinate system.

[0027] After completing the operational map modeling, it is necessary to further initialize the algorithm parameters required for region decomposition, cover candidate generation, and ALNS-DP-SA collaborative optimization in this embodiment of the invention. Specifically, this includes: setting a geometric judgment threshold ε, used for sub-region shared edge length judgment, geometric cleanup, and convexity judgment; and setting the number of candidate scan directions N. θ In this embodiment, N is taken as N θ =2, used to select the two smallest width directions as the scanning direction of the corresponding convex sub-region; set the maximum number of iterations M and the threshold for consecutive unimproved iterations M0 to control the termination condition of the ALNS-DP-SA algorithm; set the maximum number of sub-regions D that can be removed in a single destructive operation. max Used to limit the perturbation scale of adaptive large neighborhood search; sets the initial weight ω of the disruption operator. d and repair operator initial weight ω r This is used to select different destruction and repair operators according to weights during the iteration process; the weight update reaction coefficient ρ is set to control the degree of influence of the operator's historical performance on the weight update; the simulated annealing initial temperature T0, cooling coefficient α, and minimum temperature T are set. min It is used to control the probability of accepting inferior solutions and the temperature decay process.

[0028] S2. Based on the robot's external dimensions and preset safety distance, anti-collision expansion treatment is applied to the boundaries of the work area and obstacles to obtain a safe and feasible space for robot operation.

[0029] After obtaining the work area boundary P and the obstacle set O, to ensure that the robot does not collide during the execution of coverage tasks and transfers between sub-regions, the original work map needs to undergo anti-collision inflation processing. The main goal of inflation processing is to map the robot's own size and safety redundancy onto the map, allowing the robot to be approximated as a point mass when planning its path. The specific operations are as follows: (1) Calculate the expansion distance: based on the robot's equivalent radius r (characterizing the robot's external dimensions) and the preset safety distance d s Calculate the expansion distance:

[0030] (2) Boundary processing: The boundary P of the work area is offset inward to ensure that the robot's planned path maintains a safety distance of at least d from the original boundary. The offset boundary of the work area is denoted as P. -d .

[0031] (3) Obstacle handling: For each obstacle O v Perform an outward offset to ensure the robot's planned path maintains a safe distance of at least d from obstacles. The set of obstacles after the offset is denoted as O. +d .

[0032] (4) Generating a safe and feasible space for robot operation: Combining the boundaries of the operation area after collision-resistant expansion treatment with obstacles to obtain a safe and feasible space F:

[0033] Where P -d O represents the safe working boundary formed after the working area boundary is offset inward by a distance d. v +d This represents the safe obstacle avoidance zone formed after the v-th obstacle expands outward by a distance d. The symbol "\" indicates the set difference operation, which removes the expanded obstacle zone. This space is the area where the robot can safely pass and perform covering operations.

[0034] Through the above anti-collision expansion process, the original working area is transformed into a safe and feasible space F in which the robot can safely pass and operate, providing basic data for subsequent region decomposition, convex merging, cover path candidate generation and joint optimization.

[0035] S3. Based on the safe and feasible space, the initial set of convex sub-regions is generated by constrained Delaunay triangulation.

[0036] After completing the boundary expansion of the work area and obstacle anti-collision expansion, a safe and feasible space F is obtained. To transform the irregular safe and feasible space into basic units that facilitate subsequent convex merging and cover path generation, in this embodiment of the invention, a constrained Delaunay triangulation method is used to initially partition F, generating an initial set of convex sub-regions T, such as... Figure 3 As shown. The specific process of constrained Delaunay triangulation is as follows: (1) Extracting the vertex set: First, extract the outer boundary vertices from the safe and feasible space, and then collect each expanded obstacle boundary vertex in turn to form a vertex set for triangulation. All vertices are represented in the same relative coordinate system, providing a basic point set for subsequent triangulation.

[0037] (2) Constructing a set of constraint edges: Connect adjacent vertices of the outer boundary and adjacent vertices of the obstacle boundary to form a set of constraint edges. These constraint edges must be retained during the triangulation process to ensure that the generated triangles do not cross the boundaries of the work area or the obstacle.

[0038] (3) Generate the initial triangulation: Perform Delaunay triangulation on the vertex set to make each triangle as close to an equilateral shape as possible, avoid elongated triangles, and improve the stability of subsequent convex merging and path generation. Delaunay triangulation follows the empty circle property, that is, the circumcircle of any triangle does not contain other vertices.

[0039] (4) Inserting and restoring constraint edges: Since ordinary Delaunay triangulation may not be able to completely preserve the outer boundary and obstacle boundary, it is necessary to insert constraint edges one by one into the initial triangulation. If a constraint edge intersects with the triangle edge in the initial triangulation, the intersection area is subjected to edge flipping, local reconstruction or splitting until the constraint edge becomes a triangulation edge.

[0040] (5) Local Delaunay optimization: After the constrained edges are restored, local optimization is performed on the unconstrained edges in the initial triangular mesh. For any quadrilateral region formed by two adjacent triangles, if the common edge is not a constrained edge and the shape of the local triangle can be improved by flipping, then the edge flipping operation is performed; constrained edges are not allowed to be flipped.

[0041] (6) Trimming and filtering invalid triangles: For each triangle in the triangulated network after inserting and restoring constraint edges and local Delaunay optimization, determine its validity within the safe and feasible space. Triangles that are completely within the safe and feasible space are retained; those that are partially within the space are trimmed, and it is determined whether the area after trimming exceeds a preset threshold. If it exceeds the preset threshold, the trimmed polygon is retained; otherwise, the trimmed polygon is discarded; triangles located outside obstacles or the space are discarded.

[0042] (7) Generate the initial triangular sub-region set: After the above processing, the retained triangles or the effective polygons after clipping constitute the initial convex sub-region set T. Each sub-region is located inside the safe and feasible space, does not intersect with the obstacle expansion region, and meets the area threshold requirement. This set serves as the input for the subsequent greedy convex merging, providing the basic unit for generating the convex sub-regions of the final coverage path.

[0043] The initial set of convex sub-regions T will be used as input for the subsequent greedy convex merging step to further merge adjacent sub-regions that still satisfy convexity after merging, thereby obtaining a set of convex sub-regions C suitable for generating covering paths.

[0044] S4. Perform a greedy convex merging operation on the initial triangular subregion set to merge adjacent subregions that share edges and still satisfy convexity after merging, thus obtaining a convex subregion set.

[0045] After completing the constrained Delaunay triangulation, an initial set of convex subregions T is obtained. Since the number of initial convex subregions is large, directly using them for cover path generation would lead to frequent subregion switching, an increase in transfer paths, and higher overall planning complexity. Therefore, this step performs greedy convex merging on adjacent subregions that still satisfy convexity after merging, to form a smaller set of convex subregions C with a shape more suitable for reciprocating scans, such as... Figure 4 As shown. The specific steps are as follows: (1) Constructing the set of regions to be merged: Establish adjacency relationships for each sub-region in the current set of regions to be processed. If two sub-regions have a common edge, then the two sub-regions are recorded as adjacent regions, thus obtaining the current set of regions to be merged consisting of adjacent regions in the current set of regions to be processed. Among them, the initial set of regions to be processed is the initial set of convex sub-regions T obtained by constrained Delaunay triangulation.

[0046] (2) Filtering candidate merge regions: Iterate through the adjacent regions in the current set of regions to be processed in turn; the traversal is performed according to the preset traversal order or according to the principle of priority based on the length of the shared edge. If the length of the common edge of the adjacent regions is less than or equal to the preset geometric judgment threshold ε, then no merging is performed; otherwise, the adjacent regions are taken as candidate merge objects. Calculate the union region of the merged regions for the candidate merge objects.

[0047] (3) Perform convexity judgment and greedy merging: Perform convexity detection on the merged union region to determine whether its boundary vertices can form a convex polygon. If so, merge the candidate merge objects to generate a merged convex sub-region. Replace the corresponding adjacent regions with the merged convex sub-region. Convex sub-regions that have already participated in merging in this round will not participate in subsequent merging in this round. If there are concave corners, self-intersecting boundaries, or internal holes after merging, abandon the merging. After the traversal is completed, the updated set of regions to be processed is obtained.

[0048] (4) Iterative update, terminate merging and output results: Determine whether convex sub-region merging was performed during this round of traversal (i.e. whether merged convex sub-regions were generated). If yes, return to step (1) for the next round of iteration. If no, use the current set of regions to be processed as the set of convex sub-regions C. Each convex sub-region is located in the safe and feasible space F and meets the convexity requirement. It can be used as the basic unit for subsequent scanning direction selection and coverage path candidate generation.

[0049] Through the above processing, the initial convex sub-regions are merged into larger convex sub-regions, effectively reducing the number of sub-regions and the number of inter-region switching, and improving the efficiency of subsequent coverage path generation and global path optimization.

[0050] S5. For each convex sub-region, traverse its edge directions and calculate the normal projection width of the corresponding edge. Select the first two unique scanning directions of each convex sub-region in ascending order of normal projection width. Specifically, since each convex sub-region in the set C is a convex polygon, it is suitable to generate the coverage path using a parallel reciprocating scanning method. In order to reduce the number of scan rows and unnecessary turns, this step selects the scanning direction for parallel reciprocating scanning of each convex sub-region based on its own geometry.

[0051] For any convex subregion Ci First, the boundary vertices are extracted, and the edges of the convex sub-region are formed based on adjacent vertices. Each non-degenerate edge corresponds to a candidate scan direction for the convex sub-region. If the length of an edge is too short, less than a preset geometric judgment threshold, it is considered a degenerate edge and is not included in the candidate scan direction calculation; the judgment threshold must be set to ensure that at least one non-degenerate edge exists in the convex sub-region. Since scan lines with opposite directions but parallel to each other are equivalent in the coverage operation, it is necessary to normalize the candidate scan directions, merging parallel and opposite non-degenerate edge directions into a single unique scan direction. After normalization, the set of candidate scan directions for the convex sub-region is obtained.

[0052] For each candidate scan direction, with the convex sub-region C i With the centroid as the center of rotation, for the convex subregion C i The vertex coordinates are rotated clockwise by θ so that the corresponding edge of the candidate scan direction is parallel to the x-axis of the map coordinate system. After rotation, the y-direction of the coordinate system is the normal direction of the candidate scan direction. The ordinates of all boundary vertices of the rotated convex sub-region are calculated, and the difference between the maximum and minimum values ​​is taken as the normal projection width corresponding to the candidate scan direction, i.e.:

[0053] Among them, C i Let represent the i-th convex sub-region, and θ represent the angle between the candidate scan direction of the sub-region before rotation and the x-axis of the map coordinate system. This represents the coordinates of the vertices of the boundary of the convex subregion after rotation. Let w(C) represent the ordinate of the vertex after rotation. i ,θ) represents the convex subregion C i Normal projection width in the candidate scanning direction.

[0054] After calculating the corresponding normal projection width for all candidate scan directions within the same convex sub-region, they are sorted in ascending order of width. The top two unique scan directions are selected as the minimum and second minimum width scan directions for that sub-region, respectively. Simultaneously, the edges of each scan direction within the convex sub-region are saved and used as anchor edges for the corresponding scan directions. These anchor edges are used to subsequently determine the starting side and initial scan line position of the scan path, and based on this, candidate coverage paths with different entry points, exit points, and traversal directions are generated.

[0055] S6. For each scanning direction of the convex sub-region, generate reciprocating parallel scan lines corresponding to each sub-region according to the working width, and generate multiple coverage path candidates for each scanning direction.

[0056] Specifically, using the boundary edge of the sub-region corresponding to the scanning direction as the anchor edge, and the working width W as the spacing between adjacent scan lines, a set of scan lines parallel to the scanning direction are generated within each convex sub-region. After the scan lines intersect with the convex sub-region, several working line segments are formed. The robot performs the covering operation along the working line segments, and a reciprocating scanning path is formed by connecting adjacent line segments.

[0057] like Figure 5 As shown, this step generates four candidate coverage paths for each scanning direction: (1) Edge-following scanning path: Take the end of the scan line closest to an anchor edge of the convex sub-region as the starting point and perform reciprocating coverage along the scanning direction; the anchor edge is the anchor edge corresponding to the scanning direction.

[0058] (2) Edge-following reverse path: Starting from the other end of the scan line that is closest to one of the anchor edges of the convex area, perform reciprocating coverage along the scanning direction.

[0059] (3) Far-side scanning path: Take the end of the scan line that is farthest from an anchor edge as the starting point (entry point) and perform reciprocating coverage along the scanning direction.

[0060] (4) Reverse path of far-side scanning: Starting from the other end of the scan line that is furthest from an anchor edge, perform reciprocating coverage along the scanning direction.

[0061] Therefore, for a convex sub-region with two scanning directions, a candidate path set consisting of eight candidate coverage paths is generated; for a convex sub-region with one scanning direction, a candidate path set consisting of four candidate coverage paths is generated. Each path includes an entry point (starting point), an exit point (the point after a set of scan lines has been completed, i.e., the end point of the scanning path inside the convex sub-region), and an internal reciprocating scan point sequence, providing candidate solutions for joint optimization of sub-regions.

[0062] S7. Construct a joint optimization model for sub-region access order and coverage path candidates. The model aims to minimize the total cost of robot execution.

[0063] Specifically, after obtaining the candidate path set for all convex sub-regions, a joint optimization model for the sub-region access order and candidate coverage paths is constructed. This model is constrained by the principle that each convex sub-region is visited once, and only one candidate coverage path is selected for each convex sub-region. Since different candidate paths have different entry points, exit points, and internal coverage costs, the sub-region access order and candidate path selection jointly affect the transfer cost (external cost) between adjacent sub-regions; therefore, both need to be jointly optimized under the same objective function.

[0064] Assume the sub-region access order is as follows:

[0065] in, This represents the number of convex subregions in the set C. Indicates the first The visited convex subregions are numbered. Let the selected candidate path sequence be:

[0066] in, Indicates the first The candidate path number selected for each access sub-region. Indicates the first The candidate paths selected for each visited sub-region.

[0067] The joint optimization model aims to minimize the total cost of robot execution, and the objective function is:

[0068] in, Indicates the starting point of the plan To candidate paths The connection cost of the entry point; Indicate candidate path The internal execution cost (coverage cost); Indicates adjacent candidate paths and The cost of transfer between them; Indicate candidate path From the exit point to the planned endpoint The connection cost.

[0069] In this embodiment of the invention, path execution time is used as the cost. To reflect the kinematic characteristics of the robot, this embodiment uses a uniform acceleration-uniform velocity-uniform deceleration model to calculate the path execution time as the cost. Let any polyline path be:

[0070] in, This represents the number of points along the polyline path. For the first line path The coordinates of the points.

[0071] For length of In the straight-line motion segment, if the robot can accelerate to its maximum linear velocity... If the distance is short and the robot begins to decelerate before reaching its maximum linear velocity, then the motion segment consists of an acceleration segment, a constant velocity segment, and a deceleration segment. Therefore, the motion time for a straight segment... It can be represented as:

[0072] in, This refers to the linear acceleration of the robot.

[0073] For two consecutive line segments and Path points The turning angle at that point is:

[0074] Similarly, the robot's turning process employs a uniform angular acceleration-uniform angular velocity-uniform angular deceleration model. For the turning angle... Its turning time It can be represented as:

[0075] in, For the robot's angular acceleration, This represents the maximum angular velocity of the robot.

[0076] Therefore, the broken line path The execution time can be expressed as:

[0077] S8. The ALNS-DP-SA algorithm is used to solve the joint optimization model. The adaptive large neighborhood search (ALNS) is used to perturb and repair the access order of sub-regions, the dynamic programming (DP) is used to decode the optimal coverage path candidates of each sub-region under a fixed access order, and the simulated annealing criterion (SA) is used to accept candidate solutions that meet the probability conditions.

[0078] Specifically, after constructing the joint optimization model in step S7, this step uses the ALNS-DP-SA algorithm to solve for the sub-region access order and coverage path candidate selection. The algorithm flowchart is as follows: Figure 6 As shown. The specific execution steps of this algorithm are as follows: (1) Constructing a candidate path set: Based on the candidate paths covering each convex sub-region obtained in step S6, a candidate path set is constructed. For the i-th convex sub-region C i The set of candidate paths is denoted as K. i Each candidate path includes an entry point, an exit point, a list of internal scan points, and an internal operation cost.

[0079] (2) Generating an initial solution: First, a greedy strategy is used to generate an initial sub-region access order π0. The greedy strategy can select the next sub-region to be visited step by step based on the connection cost between the current region exit and the unvisited region entrance. After obtaining the initial access order, the dynamic programming method is called on this fixed order to select the candidate path combination k0 with the minimum total cost from the candidate path set of each sub-region, to obtain the initial complete solution (π0, k0), which is used as the current solution and the current optimal solution.

[0080] (3) Initialize ALNS-DP-SA algorithm parameters: Initialize the maximum number of iterations M, the threshold for consecutive unimproved iterations M0, and the maximum number of sub-regions allowed to be removed in a single destructive operation D. max Simulated annealing initial temperature T0, cooling coefficient α, minimum temperature T min , Disrupt operator weight ω d Repair operator weights ω r And update the operator weight response coefficient ρ. Set the current iteration number to 0, and the number of consecutive unimproved iterations to 0.

[0081] (4) Selecting the destruction operator: In the m-th iteration, a destruction operator is randomly selected based on the current weight of the destruction operators in ALNS, and the sub-regions in the current solution are accessed in order of π. m-1 Performing a destruction operation removes a portion of the sub-regions, resulting in a partial access order; the higher the weight of the destruction operator, the higher its probability of being selected. Destruction operators include: Random removal operator: Randomly removes several sub-regions from the current access order; Worst-case transition removal operator: Calculates the transition cost between adjacent sub-regions in the current path, and prioritizes removing sub-regions that cause higher transition costs; Related region removal operator: Select a sub-region as a seed and remove several sub-regions that are spatially close to it or have a strong connection with it; Continuous segment removal operator: Removes a continuous sub-region segment from the current access order.

[0082] (5) Selecting a repair operator: Randomly select a repair operator based on the current weight of the repair operators in ALNS, and re-insert the removed sub-regions into the partial access order to obtain a new candidate access order, i.e., the sub-region access order π under the current iteration. m The repair operators include: Optimal insertion operator: Try each possible insertion position one by one, and select the position that minimizes the increase in total cost for insertion; Regret-value insertion operator: compares the cost difference between the optimal and second-best insertion positions in a sub-region and prioritizes inserting into the sub-region with the larger regret value; Direction-aware insertion operator: Combining the entry, exit, and scanning direction information of candidate coverage paths, it prioritizes the insertion at locations that are more conducive to inter-region connectivity.

[0083] (6) Dynamic programming to decode candidate paths: For the access order π of the sub-regions obtained after repair m Dynamic programming is used to select the optimal coverage path candidates for each sub-region. Dynamic programming processes each sub-region sequentially according to the candidate access order, comparing the candidate paths within each sub-region and comprehensively considering the internal operation cost of the current candidate path as well as the transition connection cost between it and the candidate path in the previous sub-region, to obtain the candidate path combination with the minimum total cost under this fixed access order. .

[0084] Specifically, for the first sub-region in the access sequence, the connection cost from the planning starting point S to the entrance of each candidate path and its internal operation cost are calculated to obtain the cumulative cost when each candidate path is the current path. For subsequent sub-regions, for each candidate path in the current sub-region, the transfer cost from the exit of the candidate path in the previous sub-region to the entrance of the current candidate path is calculated sequentially. The minimum value of the cumulative cost from the previous sub-region and the internal operation cost of the current candidate path is selected as the minimum cumulative cost to reach the current candidate path, and the corresponding predecessor candidate path is recorded. For the last sub-region, the connection cost from its candidate path exit to the planning endpoint E is also calculated, and the candidate path with the minimum total cost is selected. The optimal candidate path combination for each sub-region is obtained by backtracking the recorded predecessor candidate paths. This yields the complete solution for the current iteration. and its total cost .

[0085] (7) Determine whether to accept a candidate solution: calculate the total cost of the complete solution under the current iteration. Compare the total cost of the current optimal solution (i.e., the optimal solution in the first m-1 iterations) with the total cost of the complete solution in the current iteration. If the total cost is less than the current optimal solution, then the complete solution for the current iteration is accepted directly. And update it with the current solution (i.e., the solution retained after the m-th iteration and used as the starting point for the (m+1)-th iteration search) and the current optimal solution, and reset the consecutive unimproved count to zero; if If the total cost of the solution is greater than or equal to the total cost of the current optimal solution, but less than the total cost of the current solution (the solution retained after the (m-1)th iteration and used as the starting point for the mth iteration search), then the complete solution of the current iteration is directly accepted and updated as the current solution, while keeping the current optimal solution unchanged, and the consecutive unimproved number is incremented by one; otherwise, the simulated annealing criterion is used to determine whether to use the complete solution of the current iteration. Update the current solution and increment the number of consecutive unimproved solutions by one. The smaller the difference between the current total cost and the current temperature, the higher the complete solution under the current iteration. The higher the probability of acceptance, the better. If the complete solution under the current iteration... If the acceptance criteria for simulated annealing (Metropolis criterion) are met, then the complete solution for the current iteration is accepted. The current solution is updated using this solution, while the current optimal solution remains unchanged; otherwise, the current solution and the current optimal solution remain unchanged.

[0086] (8) Update operator weights: Assign scores to the corresponding operators based on the performance of the destruction and repair operators selected in this round. If the operator produces a new current optimal solution, that is, the total cost of the complete solution in the current iteration. If the total cost of the current optimal solution before the update is less than the total cost of the previous solution, then the destruction operator and repair operator selected in the current iteration are assigned the first score; if the complete solution in the current iteration is... If no new optimal solution is generated, but it is accepted as the current solution, then the destruction operator and repair operator selected in the current iteration are assigned a second score; if the complete solution in the current iteration is... If no new optimal solution is generated and neither is accepted as the current solution, then the destruction and repair operators selected in the current iteration will not be scored. The first score is higher than the second score, and the higher the cumulative score, the higher the weight of the corresponding operator. After a preset iteration period, the weights of the operators are updated based on their cumulative scores and usage counts, so that operators with better performance have a higher selection probability in subsequent iterations.

[0087] (9) Update the temperature and determine the termination condition: Update the simulated annealing temperature according to the cooling coefficient to gradually reduce the temperature:

[0088] Then determine whether the maximum number of iterations M has been reached, or whether the number of consecutive unimproved iterations has reached the threshold M0. If the termination condition is not met, return to step (4) to continue iterating; if the termination condition is met, stop the search.

[0089] (10) Output optimization results: After the algorithm ends, output the sub-region access order corresponding to the current optimal solution, as well as the candidate coverage path selected for each sub-region, as the final optimal solution. The results include the access order of each sub-region, candidate path entry points, exit points, and internal scan point lists, and serve as input for step S9 to generate a continuous executable path.

[0090] S9. Based on the sub-region access order and candidate path selection results obtained by the ALNS-DP-SA algorithm, connect adjacent covered path segments using the line-of-sight transfer within the safe and feasible space to generate a continuous path.

[0091] Specifically, the coverage paths within each sub-region are sequentially connected with the inter-region transfer paths to generate a continuous executable path that starts from the planning starting point, passes through all convex sub-regions, and finally reaches the planning endpoint, such as... Figure 7 As shown.

[0092] Furthermore, inter-region transfers utilize line-of-sight transfer connections. First, it is determined whether a direct line-of-sight segment exists between the two points to be connected (two adjacent regions, an exit, and an entrance) within the safe and feasible space. If it exists, this line segment is directly used as the transfer path, and the two points to be connected are the exit point of the previous sub-region and the entrance point of the next sub-region within two adjacent sub-regions. If it does not exist, a visual diagram is constructed using the path endpoints, the boundary points of the safe and feasible space, and the boundary points of obstacles. In the visual diagram, the line-of-sight segment is used as the edge, and the line segment length is used as the edge weight. Dijkstra's algorithm is then used to search for the shortest safe polyline transfer path between the two points.

[0093] After generating the transfer paths sequentially from the planning start point to the first sub-region entrance point, from the adjacent sub-region exit point to the next sub-region entrance point, and from the last sub-region exit point to the planning endpoint, the starting point connecting path, the internal coverage paths of each sub-region, the inter-region transfer paths, and the endpoint connecting path are spliced ​​together according to the access order to form a complete continuous path:

[0094] in, This represents the connecting path from the planning starting point to the first candidate path entry point in the sub-region. Indicates the first Internal coverage path for each access sub-region Indicates the first Access sub-region to the first Transfer paths between access sub-regions This represents the connection path from the exit point of the last candidate path in the sub-region to the planned endpoint.

[0095] S10. Output the complete coverage path after all sub-regions have been covered.

[0096] In this embodiment of the invention, a set of structured and executable task path data is output for autonomous robot line-following operations and human-robot collaborative takeover. The task path data Path consists of multiple path points arranged in the order of path execution, and the data structure P of the nth path point... n ={"index": n,"position": (x n , y n), "behavior_type": type}; where "index": n represents the path point index, "position": (x n , y n The coordinates of the path point are 2D coordinates, and "behavior_type": type represents the code of the robot's behavior type when reaching the current path point, as shown in Table 1: Table 1 Classification of Behavioral Types

[0097] Example 2 This invention provides an unmanned operation path planning system for irregular surface operation scenarios, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the unmanned operation path planning method for irregular surface operation scenarios in Embodiment 1 above.

[0098] The relevant technical solutions are the same as above, and will not be repeated here.

[0099] Example 3 This invention provides a computer program product, including a computer program that, when run on a computer, causes the computer to execute the steps of the unmanned operation path planning method for irregular surface operation scenarios described in Embodiment 1 above.

[0100] The relevant technical solutions are the same as above, and will not be repeated here.

[0101] Example 4 This invention provides a robot for irregular surface work scenarios. The robot performs unmanned work along an optimal planned path, which is obtained by the unmanned work path planning method for irregular surface work scenarios in Embodiment 1 above.

[0102] The relevant technical solutions are the same as above, and will not be repeated here.

[0103] Those skilled in the art will readily understand that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for unmanned operation path planning in irregular surface operation scenarios, characterized in that, include: The robot obtains operational map information, including the boundary of the operational area, obstacles within the operational area, operational width, planning start point S, and planning end point E. It then performs anti-collision expansion processing on the boundary of the operational area and obstacles to obtain the safe and feasible space F for robot operation. The safe and feasible space F is triangulated to generate an initial set of convex subregions T; a greedy convex merging algorithm is performed on the initial set of convex subregions T to obtain a set of convex subregions C; for any convex subregion C in the set of convex subregions C... i Each non-degenerate edge is taken as a candidate scan direction, and after normalizing the candidate scan directions, C is obtained. i The set of candidate scanning directions is determined; the normal projection width of each candidate scanning direction is calculated, and the two candidate scanning directions with the smallest normal projection width are selected as C. i Two scanning directions; For each scanning direction of each convex sub-region, a reciprocating parallel scanning line corresponding to each convex sub-region is generated according to the working width. After the scanning line intersects with the corresponding convex sub-region, several working line segments are formed, and a set of candidate paths for the robot to perform covering operations along the working line segments in the corresponding convex sub-region is generated. With the constraint that the robot is visited once in each convex sub-region and only one candidate path is selected in each convex sub-region, and with the objective function being to minimize the total cost of the robot from the planning starting point S through all convex sub-regions to the planning endpoint E, a joint optimization model is constructed. Solving the joint optimization model yields the optimal planned path characterized by the access order of the convex sub-regions and the selected candidate paths within each convex sub-region, thus completing the robot's operation path planning.

2. The unmanned operation path planning method according to claim 1, characterized in that, The joint optimization model is solved using the ALNS-DP-SA joint optimization algorithm, specifically including: Sep1. In the m-th iteration, a random destruction operator is selected based on the current weights of the destruction operators in the Adaptive Large Neighborhood Search (ALNS) to access the convex sub-regions in the current solution in the order π. m-1 Perform a destruction operation to remove a portion of the convex subregion and obtain a partial access order; the higher the weight of the destruction operator, the higher the probability of it being selected. Sep2. Based on the current weights of the repair operators in the Adaptive Large Neighborhood Search (ALNS), a repair operator is randomly selected to re-insert the removed convex sub-regions into the partial access order, resulting in the convex region access order π under the m-th iteration. m ; Sep3, access sequence π in the convex region m In this case, dynamic programming (DP) is used to select the candidate path combination with the minimum total cost for the robot to move from the planning starting point S, through all convex sub-regions and perform coverage operations, to the planning ending point E. ; Sep4, Find the complete solution under the m-th iteration. The corresponding total cost Total cost of the current optimal solution If a comparison is made, Then use the complete solution. Update the current solution and the current optimal solution; if Greater than or equal to However, if the cost is less than the total cost of the current solution, then the complete solution is used. Update the current solution while keeping the current optimal solution unchanged; otherwise, use the simulated annealing algorithm (SA) to determine whether to use the complete solution. Update the current solution: If the acceptance criteria for simulated annealing are met, then use the complete solution. Update the current solution and keep the current optimal solution unchanged. If the acceptance condition of simulated annealing is not met, keep the current solution and the current optimal solution unchanged. Update the current weights of the destruction operator and repair operator selected in the m-th iteration. Sep5: Determine whether the maximum number of iterations M has been reached, or whether the number of consecutive times the current optimal solution has not been updated has reached the threshold M0. If not, let m = m + 1 and return to Sep1 for the next round of iteration. If yes, output the current optimal solution, which is the optimal planning path characterized by the convex sub-region access order and the selected candidate path in each convex sub-region.

3. The unmanned operation path planning method according to claim 2, characterized in that, It also includes converting the optimal planned path into structured operation path data that the robot can execute. This structured operation path data consists of multiple path points arranged in the order of path execution, wherein the data structure P of the nth path point... n for: P n ={"index": n,"position": (x n , y n ),"behavior_type": type}; Where, "index": n represents the path point index, and "position": (x n , y n ) represents the two-dimensional coordinates of the path point, and "behavior_type": type represents the code that characterizes the robot behavior type when the current path point is reached. The robot behavior type includes straight-line operation, turning in place, moving from the current operation line to the next operation line within a convex sub-region, moving from one convex sub-region to another, and stopping.

4. The unmanned operation path planning method according to any one of claims 1-3, characterized in that, The total cost of the robot starting from the planned starting point S, performing coverage operations through all convex sub-regions, and reaching the planned ending point E. for: ； In the formula, Indicates the access order of the convex subregion. , This represents the number of convex subregions in the set C of convex subregions; Indicates the access order in the convex subregion The selected candidate path sequence is as follows. , Indicates the first Candidate paths selected for each visited convex subregion. Indicates the first The accessed convex subregion number, Indicates the first The candidate path number selected for each visited convex subregion. ; This indicates that the robot starts from the planning stage. To candidate paths The cost at the entry point; Indicates the robot in the Along the candidate path within the visited convex subregion The cost of performing a coverage operation; This indicates that the robot selects candidate paths. The exit point is moved to the candidate path. The cost at the entry point; This indicates that the robot selects candidate paths. The exit point is moved to the planned endpoint. The cost; cost Indicates a path along a broken line The execution time is calculated as follows: ； In the formula, Indicates the robot on the polyline path The time it takes to move along the straight line segment. Indicates the length of a straight line segment. For polyline paths The number of points on it; Indicates the time it takes for the robot to perform a turning maneuver. For the robot on the polyline path Two consecutive straight line segments and Path points The absolute value of the steering angle at that location; and They are respectively: ； ； In the formula, For the robot's linear acceleration, This is the robot's maximum linear velocity. For the robot's angular acceleration, This represents the maximum angular velocity of the robot. Among them, the cost for , , or The corresponding polyline path Each robot starts from the planning point. To candidate paths The polygonal path at the entry point, the robot in the... Along the candidate path within the visited convex subregion Performing coverage operations using polyline paths, and the robot's path selection from candidate paths. The exit point is moved to the candidate path. The polyline path at the entry point, and the robot's path selection from candidate paths. The exit point is moved to the planned endpoint. The broken line path.

5. The unmanned operation path planning method according to any one of claims 1-3, characterized in that, A greedy convex merging algorithm is performed on the initial set of convex subregions T to obtain a set of convex subregions C, including: (1) Establish adjacency relationships for each convex sub-region in the current set of regions to be processed, and record two convex sub-regions with a common edge as adjacent regions, so as to obtain the current set of regions to be merged composed of adjacent regions in the current set of regions to be processed; wherein, the initial set of regions to be processed is the initial set of convex sub-regions T. (2) Traverse the adjacent regions in the current set of regions to be processed in sequence according to the principle of priority of shared edge length; if the common edge length of the adjacent regions is less than or equal to the preset geometric judgment threshold ε, then the merging of the adjacent regions is not performed; otherwise, the adjacent regions are taken as candidate merging objects, and the union region of the candidate merging objects is calculated. (3) Determine whether the boundary vertices of the union region can form a convex polygon. If yes, perform a merge on the candidate merge objects to generate a merged convex sub-region and replace the corresponding adjacent region with the merged convex sub-region. If no, do not perform a merge on the candidate merge objects. After traversal, obtain the updated set of current regions to be processed. (4) Determine whether a merged convex sub-region has been generated during this round of traversal. If yes, return to step (1) for the next round of iteration. If no, use the current set of regions to be processed as the set of convex sub-regions C.

6. The unmanned operation path planning method according to claim 5, characterized in that, Generate a set of candidate paths for the robot to perform covering operations along the work line segment within the corresponding convex sub-region, including: Using the end of the scan line closest to the anchor edge of the convex region as the entry point, perform reciprocating coverage along the current scan direction to generate a candidate path; Using the other end of the scan line closest to the anchor edge of the convex region as the entry point, perform reciprocating coverage along the current scan direction to generate a candidate path; Using the end of the scan line furthest from the anchor edge of the convex region as the entry point, perform reciprocating coverage along the current scan direction to generate a candidate path; Using the other end of the scan line furthest from the anchor edge of the convex region as the entry point, perform reciprocating coverage along the current scan direction to generate a candidate path; The above candidate paths constitute the candidate path set corresponding to the current scanning direction; The anchor edge is the edge in the convex sub-region in the current scanning direction.

7. The unmanned operation path planning method according to any one of claims 1-3, characterized in that, Calculate the normal projection width for each candidate scan direction, including: For each candidate scanning direction, take the centroid of the corresponding convex sub-region as the rotation center, and rotate the vertex coordinates of the convex sub-region clockwise by θ, so that the corresponding edge of the candidate scanning direction is parallel to the x-axis of the map coordinate system. At this time, the y-direction of the map coordinate system is the normal direction of the candidate scanning direction. Calculate the ordinates of all vertices in the rotated convex sub-region, and take the difference between the maximum and minimum ordinates as the normal projection width corresponding to the candidate scan direction.

8. A path planning system for unmanned operations in irregular surface operation scenarios, characterized in that, Includes computer-readable storage media and processors; The computer-readable storage medium is used to store executable instructions; The processor is used to read executable instructions stored in the computer-readable storage medium and execute the unmanned operation path planning method for irregular surface operation scenarios as described in any one of claims 1-7.

9. A computer program product, characterized in that, The method includes a computer program that, when run on a computer, causes the computer to execute the unmanned operation path planning method for irregular surface operation scenarios as described in any one of claims 1-7.

10. A robot for working in irregular surface scenarios, characterized in that, The robot performs unmanned operations along the optimal planned path, which is obtained by the unmanned operation path planning method for irregular surface operation scenarios as described in any one of claims 1-7.

Images