Path planning method for wheel-legged robot

Abstract

The invention relates to a path planning method for a wheel-legged robot, and the method comprises the steps: obtaining three-dimensional environment data in a working space of the wheel-legged robot, setting a four-state two-dimensional grid map through the three-dimensional environment data, and carrying out the obstacle expansion of a forbidding grid in the four-state two-dimensional grid map; setting a unified path cost function, and performing obstacle avoidance and obstacle crossing constrained global path search on the four-state two-dimensional grid map by adopting an improved A algorithm to generate a continuous global reference path; the improved A algorithm is obtained by introducing a neighborhood expansion rule supporting obstacle crossing on the basis of the traditional 8 neighborhoods; wherein the neighborhood expansion rule supporting obstacle crossing is used for representing the obstacle crossing ability of the wheel-legged robot in the global path search process. According to the method, complex terrain trafficability and path comprehensive cost can be considered, and wheel-legged robot path planning is generated.

Description

Technical Field

[0001] This invention relates to the field of robot path planning technology, and in particular to a path planning method for wheeled and legged robots. Background Technology

[0002] As the application of robots in scenarios such as field inspection and building security continues to expand, the shortcomings of single wheeled robots and single legged robots in terms of terrain mobility and obstacle crossing are becoming increasingly apparent. Wheeled-legged robots, by integrating retractable leg structures and wheels into their chassis, use wheeled movement on flat surfaces and switch to obstacle crossing and stair climbing modes when encountering obstacles such as steps, giving them a significant advantage in complex environments.

[0003] In terms of path planning, existing solutions discretize the robot's workspace into a regular grid, dividing the grid into passable and impassable areas, and combining Dijkstra's algorithm and A... Classical global planning methods, such as algorithms and fast random trees, use heuristic functions to search for feasible paths from the starting point to the target point. However, this approach only distinguishes between passable and impassable areas, failing to express the difference between directly passable areas and those requiring terrain traversal. Consequently, the obstacle-crossing capability of wheeled robots is not fully reflected in the global planning stage. Furthermore, the cost function of path planning often focuses on path length or simple distance, without adequately considering the energy consumption and difficulty of different obstacle-crossing actions. Simultaneously, the lack of explicit constraints on obstacle-crossing height and safe distances between obstacles during neighborhood expansion makes it difficult to weigh and decide on path options in complex terrain. Moreover, many works on wheeled robots still employ binary grid modeling and neighborhood expansion methods for wheeled platforms, primarily ensuring safety by avoiding obstacles, while leaving the specific obstacle-crossing behavior to the underlying control and gait planning processes.

[0004] In summary, although existing path planning methods for wheeled and legged robots have achieved some results, they still have shortcomings in complex terrain, such as rough environmental modeling and a lack of comprehensive balance between obstacle crossing energy consumption and path length.

[0005] Therefore, there is an urgent need for a path planning method that can subdivide terrain accessibility levels in the environment, comprehensively consider path length and obstacle crossing energy consumption in terms of cost and search process, and impose clear constraints on obstacle crossing behavior. Summary of the Invention

[0006] To address the problems existing in the prior art, the purpose of this invention is to propose a path planning method for wheeled-legged robots. This method fully utilizes the obstacle-crossing capabilities of wheeled-legged robots in complex environments, achieving a reasonable trade-off between path length and obstacle-crossing energy consumption while ensuring safety. Specifically, it constructs an environment model with multiple grid states, discretizes complex maps and obstacle information into a two-dimensional grid map, designs path cost functions based on different terrain types and motion modes, and improves A... Within the framework of the algorithm, a global path search is performed, and the obtained path is smoothed to generate the final executable robot motion trajectory.

[0007] To achieve the above objectives, the present invention provides the following solution: A path planning method for wheeled-legged robots includes: Acquire three-dimensional environmental data in the workspace of the wheeled robot, use the three-dimensional environmental data to set up a four-state two-dimensional grid map, and expand the no-entry grid in the four-state two-dimensional grid map as an obstacle; Set a unified path cost function and adopt the improved A The algorithm performs a global path search with obstacle avoidance and obstacle crossing constraints on the four-state two-dimensional grid map, generating a continuous global reference path; the improved A The algorithm is derived by introducing a neighborhood expansion rule that supports obstacle crossing based on the traditional 8-neighborhood; The neighborhood expansion rule that supports obstacle crossing is used to represent the obstacle crossing capability of the wheel-legged robot during the global path search process.

[0008] Optionally, setting a four-state two-dimensional raster map using the three-dimensional environmental data includes: A two-dimensional grid map is set using the three-dimensional environmental data. The grids in the two-dimensional grid map are divided into directly passable grids, low-obstacle grids that can be crossed, high-obstacle grids that can be crossed, and no-entry grids, thus generating the four-state two-dimensional grid map. The directly passable grid is used to indicate areas that can be traversed; The traversable low-barrier grid and the traversable high-barrier grid are used to represent traversable obstacles of different height levels; The no-entry grid is used to indicate areas that cannot be passed through.

[0009] Optionally, obstacle inflation of the forbidden grid in the four-state two-dimensional grid map includes: Based on the body size and safety distance of the wheeled robot, the no-entry grid is expanded by means of grid points within a target number of grid radii around the obstacle as the center, thus obtaining the expanded obstacle area, which is then used as the final no-entry grid.

[0010] Optionally, setting the unified path cost function includes: If both adjacent grids are directly passable grids, and both the starting grid and the ending grid are directly passable grids, then the conventional horizontal or vertical movement cost between adjacent directly passable grids is set to a first fixed value, and the conventional diagonal movement cost is set to a second fixed value. If there is a single obstacle grid between the directly passable grids, and neither the directly passable grid nor the obstacle grid is within the restricted expansion zone, then the obstacle crossing cost for moving from the directly passable grid across a single traversable low obstacle grid to another directly passable grid is set to a normal movement cost that is a first target multiple, and the obstacle crossing cost for moving from the directly passable grid across a single traversable high obstacle grid to another directly passable grid is set to a normal movement cost that is a second target multiple.

[0011] Optionally, the improved A is adopted. The algorithm performs a global path search with obstacle avoidance and obstacle crossing constraints on the four-state two-dimensional grid map, including: Set a starting point and a target point. For any candidate node between the starting point and the target point, define the Euclidean distance between the candidate node and the target point. This distance is used to set a dynamic weighted heuristic function. The evaluation function is then determined using the dynamic weighted heuristic function. The global path search process for obstacle avoidance and obstacle crossing constraints on the four-state two-dimensional grid map follows the A... The basic framework of the algorithm is to maintain an open list and a closed list, add the starting point to the open list, set the cost of the current starting point to 0, and use the evaluation function to calculate the corresponding evaluation result. Each time, the node with the smallest evaluation result is selected from the open list as the current expansion node, the current expansion node is added to the closed list, and the current neighborhood is selected according to the neighborhood expansion rule. The corresponding cost is then selected, and the candidate cumulative cost is calculated. If the current domain is not visited or the cumulative cost of the candidates is less than the cost of the current domain, then the cost of the current domain is updated, the corresponding evaluation result is recalculated, the current extended node is taken as the parent node, and the current neighborhood is added to the open list. When each extended node taken from the open list is the same node, the discrete path sequence from the starting point to the target can be obtained by backtracking through the parent node chain, and the discrete path sequence is taken as the planning result.

[0012] Optionally, determining the evaluation function includes: ; ; in, For the evaluation function, To start from the beginning To any node The sum of the movement costs, For heuristic functions, For dynamic weighted heuristic functions, is the Euclidean distance between the candidate node and the target node.

[0013] Optionally, calculating the candidate cumulative cost includes: ; in, Cumulative cost for candidates The sum of the costs of all moves. The corresponding cost for the current neighborhood.

[0014] Optionally, generating the continuous global reference path includes: Based on the changes in the direction of the discrete path in the planning results, key nodes are extracted, and any three connected points among the key nodes are obtained to construct two adjacent direction vectors; the two adjacent direction vectors include: a first direction vector and a second direction vector. If the angle between the first direction vector and the second direction vector is less than a preset threshold, it proves that the direction is consistent. The middle node of the three-point connection has little impact on the overall trajectory shape. The middle node is merged as a redundant point, so that the node is only retained when the direction changes significantly or when the corresponding obstacle crossing start and end position is selected. After screening, the key node sequence is obtained. A cubic Bézier curve is constructed between any two adjacent key nodes in the key node sequence; the intermediate control point of the cubic Bézier curve is selected near the key node according to the entry and exit directions. By connecting the cubic Bézier curves in sequence, the smooth global path used by the wheel-legged robot during actual execution is obtained, which is the continuous global reference path.

[0015] Optionally, generating the continuous global reference path further includes: The cubic Bézier curve is sampled at preset intervals, and the sampled points are mapped to the four-state two-dimensional grid map. If the sampled point is in the prohibited grid, the intermediate control point is adjusted and the corresponding curve segment is regenerated until the obstacle avoidance requirements are met.

[0016] The beneficial effects of this invention are as follows: This invention constructs a four-state environment model that distinguishes between directly passable grids and grids with different levels of traversable obstacles. This refines the terrain passability levels during the environment modeling stage, enabling the obstacle-crossing capability of wheeled robots to be reflected in global planning. By simultaneously introducing path cost and obstacle-crossing cost into the cost function, a unified path evaluation index is established, allowing planning to make reasonable decisions between travel distance and obstacle-crossing energy consumption. Furthermore, by introducing neighborhood expansion rules supporting obstacle crossing, dynamic weighted heuristic functions, and Bezier smoothing, the accessibility, smoothness, and search efficiency of planned paths in complex terrain are improved. Therefore, this invention provides solid theoretical support and practical guidance for path planning and task execution of wheeled robots in complex environments. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a path planning method for a wheeled robot according to an embodiment of the present invention; Figure 2 This is an obstacle grid map according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the conventional 8-neighborhood and obstacle-crossing neighborhood expansion in an embodiment of the present invention; Figure 4 This is a schematic diagram of cubic Bézier curve smoothing according to an embodiment of the present invention. Detailed Implementation

[0019] 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.

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] like Figure 1 As shown, this embodiment discloses a path planning method for a wheeled-legged robot, including: acquiring three-dimensional environmental data in the workspace of the wheeled-legged robot; setting a four-state two-dimensional grid map using the three-dimensional environmental data; and performing obstacle inflation on the no-entry grids in the four-state two-dimensional grid map; setting a unified path cost function; and using an improved A... The algorithm performs a global path search with obstacle avoidance and obstacle crossing constraints on a four-state two-dimensional grid map, generating a continuous global reference path; the improved A The algorithm is obtained by introducing a neighborhood expansion rule that supports obstacle crossing on the basis of the traditional 8-neighborhood; the neighborhood expansion rule that supports obstacle crossing is used to represent the obstacle crossing ability of the wheeled robot during the global path search process.

[0022] Specifically, this embodiment discloses a path planning method for wheeled and legged robots, including: firstly, acquiring terrain and obstacle information based on the robot's environmental perception, constructing a two-dimensional grid map containing multiple grid states, and dividing the grid into four states: white directly passable grids, gray traversable low-obstacle grids, blue traversable high-obstacle grids, and black prohibited grids. The black prohibited grids represent areas that cannot be passed, the white directly passable grids represent passable areas, and the gray traversable low-obstacle grids and blue traversable high-obstacle grids represent traversable obstacles of different height levels. The gray and blue grids are not used as landing points for path nodes, and their corresponding obstacle-crossing costs are 1.5 times and 2 times the cost of directly passable areas, respectively, reflecting the difference in obstacle-crossing energy consumption for obstacles of different heights. Based on this, for regular movement, only the case where both the starting and target grids are white, adjacent, and directly passable grids is considered, and the base cost is set according to the three movement directions: horizontal, vertical, and diagonal. For obstacle-crossing movement, the robot starts from a white, directly passable grid, crosses one obstacle grid, and reaches an adjacent white, directly passable grid. The obstacle-crossing cost is determined based on the obstacle level of the crossed grid. The sum of the regular movement cost and the obstacle-crossing movement cost is used as A. The actual cost of the algorithm, g(n), A The algorithm automatically compares the total cost of different candidate paths during the search process. Furthermore, based on the traditional 8-neighborhood expansion, this invention adds an obstacle-crossing neighborhood expansion rule. In addition to considering eight adjacent white grids as neighbors, it also detects white grids separated from it by one obstacle grid in the horizontal, vertical, and diagonal directions. If the obstacle grid is a traversable grid, the corresponding white grid is included as an obstacle-crossing candidate node in the search, thus reflecting the obstacle-crossing capability of the wheeled robot in the global path planning stage. Simultaneously, a dynamically weighted heuristic function based on Euclidean distance is used to adjust the heuristic weights according to the relative distance between the current node and the target node, improving search efficiency while also considering path quality.

[0023] Furthermore, to improve the trajectory realizability and motion smoothness of the path, this invention is based on improved A... After obtaining the discrete global path, the algorithm selects key nodes in the path as control points and smooths the polyline path using cubic Bézier curves to generate a geometrically continuous trajectory. To ensure that the smooth trajectory still meets obstacle avoidance requirements, the Bézier curve is sampled at preset intervals, and the sampled points are mapped onto a grid map. If a sampled point falls into a black grid or an obstacle expansion area, a collision risk is identified, and the corresponding trajectory segment is adjusted or regenerated, thereby avoiding crossing obstacle areas while maintaining smoothness.

[0024] In one embodiment, the overall process includes establishing a grid environment and cost model, and based on improved A... Global path search, path smoothing, and collision testing. The specific implementation process is as follows: The first step involves the wheeled robot acquiring terrain height information and obstacle distribution information in the workspace using methods such as LiDAR or depth cameras, thus obtaining a 3D environment dataset. ; in, For a 3D environment dataset, These are ground plane coordinates. This refers to the elevation of the terrain or the height of the obstacle at that location. To facilitate grid modeling, the robot's working plane is confined to a rectangular area. and given the raster resolution ,Pick and discretize the workspace into There are raster cells, where the number of raster rows and columns are respectively: ;; The mapping relationship between continuous space and raster index is as follows: ; No. line, number The world coordinates of the column grid center can be represented as: ; ; in, , .

[0025] The second step, as Figure 2 As shown, in order to balance the wheeled locomotion and obstacle-crossing capabilities of the wheeled robot, each grid is divided into four states based on its passability: (1) White grid that can be directly passed is designated as type 0: the terrain has small elevation changes and wheeled vehicles can pass directly; (2) Gray grids that can be crossed with low obstacles are designated as type 1: the obstacle height is within the low threshold range and can be crossed by low-difficulty obstacle-crossing actions; (3) Blue grids that can be crossed over high obstacles are designated as type 2: the obstacles are higher and can be crossed by high-difficulty obstacle-crossing maneuvers; (4) Black no-entry grid, designated as type 3: if the obstacle height or danger level exceeds the obstacle crossing ability of the wheeled robot, it is considered completely impassable.

[0026] Set three height thresholds This is used to distinguish different categories, and type functions are used. Indicates the first The state of a grid cell can be defined as follows: ; Furthermore, considering the robot's size and safety distance, the black no-entry grid is expanded: centered on the obstacle, the grid is expanded around it. Grid points within a grid radius are also marked as prohibited areas, thus obtaining the expansion barrier region.

[0027] Thirdly, in the aforementioned grid environment, wheeled robots exhibit two basic movement modes: one is conventional movement along the white grids, and the other is obstacle-crossing movement across gray or blue grids. The following establishes a unified cost model for these two movement modes, enabling a trade-off between path length and obstacle-crossing energy consumption during path planning.

[0028] Let the current node be The corresponding raster index is Its adjacent white grid is The corresponding raster index is The geometric distance between the two is: ; For normal movement, this embodiment normalizes the base cost of the white grid to 1, and gives: ; The regular movement cost is only allowed when both the starting and ending grids are white, passable grids.

[0029] For obstacle-crossing movement, the robot moves from the white grid. Cross a gray or blue grid to reach the white grid on the other side. Let the current white raster index be... The direction vector is Then the intermediate obstacle grid is: ; The white grid where you land is: ; when Gray or blue grid If the grid is white and not within the restricted expansion zone, the obstacle-crossing maneuver is considered feasible. The approximate geometric length of the obstacle-crossing motion is: ; To reflect the difference in energy consumption when traversing obstacles of different heights, obstacle clearance coefficients for high and low obstacles are introduced: ; Based on this, the obstacle crossing cost is defined as: ; If the height of the obstacle exceeds the maximum obstacle-crossing height of the wheeled robot ,Right now If this occurs, the corresponding obstacle-crossing action is prohibited. Throughout the entire path, from the starting point... To any node The actual cost is the sum of the costs of all the moves taken: ; in, The cost can be either the conventional cost or the obstacle-crossing cost. Using this unified cost model, if a path frequently needs to cross tall blue obstacles, even if the geometric distance is short, its total cost will increase significantly, and therefore it will not be adopted in the final path planning.

[0030] Step four, as Figure 3 As shown, in A In the search, neighborhood generation directly determines the search directions the algorithm can attempt. Based on the traditional 8-neighborhood, a neighborhood expansion rule supporting obstacle crossing is added, allowing the wheeled robot's obstacle-crossing capability to be reflected globally during the search process. For the current node... Its regular neighborhood is defined as: ; This refers to all white, passable grid cells that are horizontally, vertically, or diagonally adjacent to the current grid cell. Based on this, an obstacle-crossing neighborhood is constructed according to the obstacle-crossing conditions described in step three. For each direction vector... If the middle grid Gray or blue, target obstacle If it is white and has not expanded, it is prohibited. Join the group as a neighbor who crosses the barrier: ; Ultimately, the node The neighborhood of is: ; from The candidate nodes obtained from the process use the conventional cost, from The candidate nodes obtained are evaluated using obstacle crossing costs. This way, the search tree is only expanded on white grids, ensuring that all path nodes are located in directly passable areas, while gray and blue grids only participate in cost calculations as intermediate obstacle crossing grids, reflecting obstacle crossing behavior at the search level but not directly becoming path nodes.

[0031] Fifth, after determining the grid environment and cost model, the improved A... The algorithm performs a global path search. Let the starting point be... The target point is Both correspond to white, directly passable grid cells. The Euclidean distance from the starting point to the target point is: ; For any candidate node The Euclidean distance from the target point is defined as: ; remember To balance search efficiency and path quality, a dynamically weighted heuristic function is introduced: ; in, It is a non-negative adjustment coefficient. For the current node Distance to the target node This represents the distance from the starting point to the target node. When a node moves further away from the target node, Larger When the value is relatively large, the heuristic function evaluates the function. The weight of nodes is higher, and the search process relies more on heuristic information to reduce ineffective expansion; as nodes gradually approach the target... Get smaller It gradually decreases and approaches 1, its behavior approaching that of traditional A. The algorithm ensures that the final path is optimal in terms of total cost.

[0032] Based on this, the heuristic function and the evaluation function are respectively: ; The search process follows A The basic framework consists of maintaining an open list (OPEN) and a closed list (CLOSED). The starting point is... Add to OPEN, making Calculate the corresponding and Each time, select from OPEN. The node with the smallest value As the current expansion node, move it into CLOSED, and then generate neighborhoods according to the neighborhood generation rules in step four. For each neighborhood The appropriate cost is selected based on whether it is conventional movement or obstacle-crossing movement. Calculate the candidate cumulative cost: ; If the neighborhood is not visited, or Then update at a lower cost. and recalculate and , will the current node Record as the parent node, Add or update to OPEN.

[0033] When the extended node retrieved from OPEN each time is exactly a node When the target is not found, a discrete path sequence from the starting point to the target can be obtained by backtracking through the parent node chain. If OPEN is empty and the target is still not found, it means that there is no feasible path under the current environment and constraints. In practical use, by adjusting the obstacle crossing height threshold and cost weight, a reasonable decision can be made between path length, obstacle crossing frequency and energy consumption.

[0034] Step 6, as follows Figure 4 As shown, the planning result obtained from step five is a polygonal path connecting the center points of the white grid sequentially. Although it satisfies obstacle avoidance and obstacle crossing constraints, it has many inflection points and abrupt changes in direction, which is not conducive to the smooth tracking of the wheeled robot. To improve the smoothness of the trajectory, a cubic Bézier curve is used for smoothing based on the discrete path.

[0035] First, extract the key nodes based on the changes in the discrete path. Let the discrete path be represented sequentially as follows: Three points along a single survey route Construct two adjacent direction vector segments: ; like and If the included angle is less than a preset threshold, it indicates that the directions at that point are basically consistent, and the intermediate node is considered to be... Nodes with minimal impact on the overall trajectory shape can be merged as redundant points, and are only retained when there is a significant change in direction or at the corresponding obstacle crossing start and end points. After screening, the key node sequence is obtained. ,in and These represent the start and end points of the path, respectively. Then, a cubic Bézier curve is constructed between any pair of adjacent key nodes. Let the four control points of this curve be _____. intermediate control point Selected near key nodes based on the entry and exit directions, this segment controls the specific shape of the curve. The smooth trajectory can be represented as: ; By appropriately setting the control points of each segment of the Bézier curve, the original corners are transformed into smooth transitions while maintaining the original path and obstacle-crossing positions, facilitating tracking by the wheeled robot. Connecting the cubic Bézier curves sequentially yields the smooth global path used by the wheeled robot during actual execution.

[0036] The cubic Bézier curve is sampled at preset intervals, and the sampled points are mapped to the grid map. If a sampled point enters a restricted grid, the intermediate control point is adjusted and the corresponding curve segment is regenerated until the obstacle avoidance requirements are met.

[0037] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A path planning method for wheeled-legged robots, characterized in that, include: Acquire three-dimensional environmental data in the workspace of the wheeled robot, use the three-dimensional environmental data to set up a four-state two-dimensional grid map, and expand the no-entry grid in the four-state two-dimensional grid map as an obstacle; Set a unified path cost function and adopt the improved A The algorithm performs a global path search with obstacle avoidance and obstacle crossing constraints on the four-state two-dimensional grid map, generating a continuous global reference path; the improved A The algorithm is derived by introducing a neighborhood expansion rule that supports obstacle crossing based on the traditional 8-neighborhood; The neighborhood expansion rule that supports obstacle crossing is used to represent the obstacle crossing capability of the wheel-legged robot during the global path search process.

2. The path planning method for a wheeled-legged robot according to claim 1, characterized in that, Setting up a four-state two-dimensional raster map using the aforementioned three-dimensional environmental data includes: A two-dimensional grid map is set using the three-dimensional environmental data. The grids in the two-dimensional grid map are divided into directly passable grids, low-obstacle grids that can be crossed, high-obstacle grids that can be crossed, and no-entry grids, thus generating the four-state two-dimensional grid map. The directly passable grid is used to indicate areas that can be traversed; The traversable low-barrier grid and the traversable high-barrier grid are used to represent traversable obstacles of different height levels; The no-entry grid is used to indicate areas that cannot be passed through.

3. The path planning method for a wheeled-legged robot according to claim 1, characterized in that, Obstacle inflation of the restricted grid cells in the four-state two-dimensional grid map includes: Based on the body size and safety distance of the wheeled robot, the no-entry grid is expanded by means of grid points within a target number of grid radii around the obstacle as the center, thus obtaining the expanded obstacle area, which is then used as the final no-entry grid.

4. The path planning method for a wheeled-legged robot according to claim 2, characterized in that, Setting the unified path cost function includes: If both adjacent grids are directly passable grids, and both the starting grid and the ending grid are directly passable grids, then the conventional horizontal or vertical movement cost between adjacent directly passable grids is set to a first fixed value, and the conventional diagonal movement cost is set to a second fixed value. If there is a single obstacle grid between the directly passable grids, and neither the directly passable grid nor the obstacle grid is within the restricted expansion zone, then the obstacle crossing cost for moving from the directly passable grid across a single traversable low obstacle grid to another directly passable grid is set to a normal movement cost that is a first target multiple, and the obstacle crossing cost for moving from the directly passable grid across a single traversable high obstacle grid to another directly passable grid is set to a normal movement cost that is a second target multiple.

5. The path planning method for a wheeled-legged robot according to claim 1, characterized in that, Adopting the improved A The algorithm performs a global path search with obstacle avoidance and obstacle crossing constraints on the four-state two-dimensional grid map, including: Set a starting point and a target point. For any candidate node between the starting point and the target point, define the Euclidean distance between the candidate node and the target point. This distance is used to set a dynamic weighted heuristic function. The evaluation function is then determined using the dynamic weighted heuristic function. The global path search process for obstacle avoidance and obstacle crossing constraints on the four-state two-dimensional grid map follows the A... The basic framework of the algorithm is to maintain an open list and a closed list, add the starting point to the open list, set the cost of the current starting point to 0, and use the evaluation function to calculate the corresponding evaluation result. Each time, the node with the smallest evaluation result is selected from the open list as the current expansion node, the current expansion node is added to the closed list, and the current neighborhood is selected according to the neighborhood expansion rule. The corresponding cost is then selected, and the candidate cumulative cost is calculated. If the current domain is not visited or the cumulative cost of the candidates is less than the cost of the current domain, then the cost of the current domain is updated, the corresponding evaluation result is recalculated, the current extended node is taken as the parent node, and the current neighborhood is added to the open list. When each extended node taken from the open list is the same node, the discrete path sequence from the starting point to the target can be obtained by backtracking through the parent node chain, and the discrete path sequence is taken as the planning result.

6. The path planning method for a wheeled-legged robot according to claim 5, characterized in that, Determining the evaluation function includes: ； ； in, For the evaluation function, To start from the beginning To any node The sum of the movement costs, For heuristic functions, For dynamic weighted heuristic functions, is the Euclidean distance between the candidate node and the target node.

7. The path planning method for a wheeled-legged robot according to claim 5, characterized in that, Calculating the candidate cumulative cost includes: ； in, Cumulative cost for candidates The sum of the costs of all moves. The corresponding cost for the current neighborhood.

8. The path planning method for a wheeled-legged robot according to claim 5, characterized in that, Generating the continuous global reference path includes: Based on the changes in the direction of the discrete path in the planning results, key nodes are extracted, and any three connected points among the key nodes are obtained to construct two adjacent direction vectors; the two adjacent direction vectors include: a first direction vector and a second direction vector; If the angle between the first direction vector and the second direction vector is less than a preset threshold, it proves that the direction is consistent. The middle node of the three-point connection has little impact on the overall trajectory shape. The middle node is merged as a redundant point, so that the node is only retained when the direction changes significantly or when the corresponding obstacle crossing start and end position is selected. After screening, the key node sequence is obtained. A cubic Bézier curve is constructed between any two adjacent key nodes in the key node sequence; the intermediate control point of the cubic Bézier curve is selected near the key node according to the entry and exit directions. By connecting the cubic Bézier curves in sequence, the smooth global path used by the wheel-legged robot during actual execution is obtained, which is the continuous global reference path.

9. The path planning method for a wheeled-legged robot according to claim 8, characterized in that, Generating the continuous global reference path further includes: The cubic Bézier curve is sampled at preset intervals, and the sampled points are mapped to the four-state two-dimensional grid map. If the sampled point is in the prohibited grid, the intermediate control point is adjusted and the corresponding curve segment is regenerated until the obstacle avoidance requirements are met.

Images