CN-121977572-A - Path planning method for wheeled leg type robot

CN121977572ACN 121977572 ACN121977572 ACN 121977572ACN-121977572-A

Abstract

The invention relates to a path planning method for a wheel-legged robot, which comprises the steps of obtaining three-dimensional environment data in a wheel-legged robot working space, setting a four-state two-dimensional grid map by utilizing the three-dimensional environment data, expanding obstacles on a forbidden grid in the four-state two-dimensional grid map, setting a unified path cost function, and adopting an improved A The algorithm searches the global path of obstacle avoidance and obstacle surmounting constraint on the four-state two-dimensional grid map to generate a continuous global reference path, and the improved A The algorithm is obtained by introducing a neighbor expansion rule supporting obstacle crossing based on the traditional 8 neighbor, wherein the neighbor expansion rule supporting obstacle crossing is used for representing obstacle crossing capability of the legged robot in the global path searching process. The invention can give consideration to the trafficability of complex terrain and the comprehensive cost of the path, and generate the path planning of the wheel-leg robot.

Inventors

ZHANG YAWEN
GUO HUI
WANG YANSONG
SUN PEI
MA XIPEI
HUANG SHUANG
LIU NINGNING

Assignees

上海工程技术大学

Dates

Publication Date: 20260505
Application Date: 20260212

Claims (9)

1. A method for path planning for a wheeled legged robot, comprising: Acquiring three-dimensional environment data in a wheel leg robot working space, setting a four-state two-dimensional grid map by utilizing the three-dimensional environment data, and expanding a forbidden grid in the four-state two-dimensional grid map; setting a unified path cost function and adopting an improved A The algorithm searches the global path of obstacle avoidance and obstacle surmounting constraint on the four-state two-dimensional grid map to generate a continuous global reference path, wherein the improved A The algorithm is obtained by introducing a neighbor expansion rule supporting obstacle crossing based on the traditional 8 neighbor; the neighbor expansion rule supporting obstacle crossing is used for representing obstacle crossing capability of the wheel leg robot in the global path searching process.
2. The path planning method for a legged robot according to claim 1, wherein setting a four-state two-dimensional grid map using the three-dimensional environment data comprises: setting a two-dimensional grid map by utilizing the three-dimensional environment data, dividing grids in the two-dimensional grid map into a directly passable grid, a low-obstacle-crossing grid, a high-obstacle-crossing grid and a forbidden grid, and generating the four-state two-dimensional grid map; The directly passable grid is used for representing a passable area; The spanable low obstacle grid and the spanable high obstacle grid representing spanable obstacles of different height levels; The forbidden grid is used for representing an area which cannot pass through.
3. The path planning method for a legged robot according to claim 1, wherein performing obstacle expansion on a forbidden grid in the four-state two-dimensional grid map comprises: And according to the body size and the safety distance of the wheel leg robot, performing obstacle expansion on the forbidden grid, namely taking the obstacle as a center, marking grid points around the obstacle in a grid radius range of a target number as forbidden areas, obtaining an expanded obstacle area, and taking the expanded obstacle area as a final forbidden grid.
4. The path planning method for a legged robot according to claim 2, wherein setting the unified path cost function includes: if the adjacent grids are the directly passable grids and the starting point grids and the end point grids are the directly passable grids, setting the horizontal or vertical conventional movement cost between the adjacent directly passable grids as a first fixed value and setting the diagonal conventional movement cost as a second fixed value; If a single barrier grid exists between the directly-passable grids and neither the directly-passable grids nor the barrier grids are in a forbidden expansion interval, setting the obstacle crossing cost of obstacle crossing movement from the directly-passable grids to the other directly-passable grids across a single low barrier grid as a conventional movement cost of a first target multiple, and setting the obstacle crossing cost of obstacle crossing movement from the directly-passable grids to the other directly-passable grids across a single high barrier grid as a conventional movement cost of a second target multiple.
5. The path planning method for a legged robot according to claim 1, wherein an improved a is adopted The algorithm for carrying out global path search of obstacle avoidance and obstacle surmounting constraint on the four-state two-dimensional grid map comprises the following steps: setting a starting point and a target point, defining Euclidean distance between the candidate node and the target point aiming at any candidate node between the starting point and the target point, and setting a dynamic weighting heuristic function, and determining an evaluation function by utilizing the dynamic weighting heuristic function; global path searching process of obstacle avoidance and obstacle surmounting constraint on four-state two-dimensional grid map is carried out by using A The basic framework of the algorithm is to maintain an open list and a closed list, add the starting point into the open list, enable the cost of the current starting point to be 0, calculate a corresponding evaluation result by utilizing the evaluation function, select a node with the smallest evaluation result from the open list as a current expansion node each time, add the current expansion node into the closed list, select a current neighborhood according to the neighborhood expansion rule, further select a corresponding cost, and calculate candidate accumulated cost; If the current field is not accessed or the candidate accumulated cost is smaller than the cost of the current field, updating the cost of the current field, recalculating a corresponding evaluation result, taking the current expansion node as a father node, adding the current neighborhood into the open list, and when expansion nodes taken out from the open list each time are the same nodes, backtracking to obtain a discrete path sequence from a starting point to a target through a father node chain, wherein the discrete path sequence is taken as a planning result.
6. The path planning method for a legged robot according to claim 5, wherein determining the evaluation function comprises: ; ; Wherein, the In order to evaluate the function of the device, To be from the starting point Any node Is used to determine the sum of the movement costs of (a), In order to be a heuristic function, For a dynamically weighted heuristic function, Is the Euclidean distance between the candidate node and the target point.
7. The path planning method for a legged robot according to claim 5, wherein calculating the candidate accumulated cost comprises: ; Wherein, the For the candidate accumulated cost to be a candidate, For the sum of all the movement costs, And corresponding cost is given to the current neighborhood.
8. The method for wheeled legged robot path planning according to claim 5, wherein generating the continuous global reference path comprises: Extracting key nodes according to trend changes of discrete paths in the planning result, and acquiring any three connecting points in the key nodes for constructing two adjacent sections of direction vectors, wherein the two adjacent sections of direction vectors comprise a first direction vector and a second direction vector; If the included angle between the first direction vector and the second direction vector is smaller than a preset threshold value, the trend is consistent, the influence of the middle node of the three connecting points on the overall track shape is small, the middle node is used as a redundant point for merging, so that the node is reserved only when the direction is obviously changed or the corresponding obstacle crossing start and stop positions are generated, and a key node sequence is obtained after screening; constructing a cubic Bezier curve between any adjacent key nodes in the key node sequence, wherein the middle control point of the cubic Bezier curve is selected near the key nodes according to the entering and leaving directions; And sequentially connecting the three Bezier curves to obtain a smooth global path used when the wheel-leg robot is actually executed, namely the continuous global reference path.
9. The method for wheel legged robot path planning according to claim 8, wherein generating the continuous global reference path further comprises: and sampling the cubic Bezier curve according to a preset interval, mapping sampling points to the four-state two-dimensional grid map, and if the sampling points are on the forbidden grid, adjusting the middle control points and regenerating the corresponding curve segments until the obstacle avoidance requirement is met.

Description

Path planning method for wheeled leg type robot Technical Field The invention relates to the technical field of robot path planning, in particular to a path planning method for a wheel-leg robot. Background With the continuous expansion of the application of robots in the scenes of outdoor inspection, building security and the like, the defects of single wheeled robots and single legged robots in the aspects of terrain trafficability and obstacle surmounting are increasingly highlighted. The wheel leg type robot has obvious advantages in a complex environment by integrating telescopic leg structures and wheels on a chassis, adopting wheel type movement on a flat road surface and switching into a movement mode of obstacle crossing, step climbing and the like when encountering obstacles such as steps and the like. In the aspect of path planning, the prior art scheme discretizes the robot work space into a regular grid, divides the grid into a passable area and a non-passable area, and combines the Dijkstra algorithm and the AAnd (3) carrying out searching by using heuristic functions to obtain a feasible path from the starting point to the target point by using classical global planning methods such as an algorithm, a rapid random tree and the like. However, the technical scheme only distinguishes passable areas from non-passable areas, and is difficult to express the distinction between passable areas and terrain to be spanned, so that the obstacle crossing capability of the wheel-legged robot is not fully embodied in the global planning stage. On the other hand, the cost function of path planning is mainly path length or simple distance, energy consumption cost and passing difficulty of different obstacle surmounting actions are not fully considered, meanwhile, clear constraint is not carried out on obstacle surmounting height and safety distance between obstacles in neighborhood expansion, and therefore the path scheme is difficult to weigh and decide under complex terrain. In addition, aiming at the wheel leg type robot, a plurality of works still keep on using a binary grid modeling and neighborhood expansion mode oriented to a wheel platform, safety is ensured mainly by bypassing obstacles, and specific obstacle surmounting behaviors are reserved for bottom layer control and gait planning processing. In summary, although the existing path planning method of the legged robot achieves a certain result, the method still has the defects of rough environment modeling, lack of comprehensive trade-off between obstacle surmounting energy consumption and path length and the like under complex terrain. Therefore, there is a need for a path planning method that can subdivide the terrain passable level in the environment, comprehensively consider path length and obstacle surmounting energy consumption in the cost and search process, and impose explicit constraints on obstacle surmounting behaviors. Disclosure of Invention In order to solve the problems in the prior art, the invention aims to provide a path planning method for a wheel-legged robot, which fully utilizes the obstacle surmounting capability of the wheel-legged robot in a complex environment and realizes reasonable balance between path length and obstacle surmounting energy consumption on the premise of ensuring safety. Specifically, an environment model with multiple grid states is constructed, complex map and obstacle information are discretized into a two-dimensional grid map, a path cost function is designed based on different terrain types and motion modes, and A is improvedAnd carrying out global path search under the framework of the algorithm, and carrying out smoothing treatment on the obtained path to generate a final executable robot motion track. In order to achieve the above object, the present invention provides the following solutions: A method for path planning for a wheeled legged robot, comprising: Acquiring three-dimensional environment data in a wheel leg robot working space, setting a four-state two-dimensional grid map by utilizing the three-dimensional environment data, and expanding a forbidden grid in the four-state two-dimensional grid map; setting a unified path cost function and adopting an improved A The algorithm searches the global path of obstacle avoidance and obstacle surmounting constraint on the four-state two-dimensional grid map to generate a continuous global reference path, wherein the improved AThe algorithm is obtained by introducing a neighbor expansion rule supporting obstacle crossing based on the traditional 8 neighbor; the neighbor expansion rule supporting obstacle crossing is used for representing obstacle crossing capability of the wheel leg robot in the global path searching process. Optionally, setting the four-state two-dimensional grid map using the three-dimensional environment data includes: setting a two-dimensional grid map by utilizing the three-dimensional environment data, dividing grids in th