A lawn mowing robot path planning method based on an improved RRT algorithm

Abstract

A kind of lawn trimming robot path planning method based on improved RRT algorithm, based on spiral traversal method, map is covered to establish real-time grid map and carry out global path planning according to spiral direction, when encountering obstacle, improved RRT algorithm is used to carry out local path planning, the map constructed has good global consistency, through the navigation scheme that global path planning and local path planning dynamic obstacle avoidance are combined, improve the overall path optimization efficiency by combining potential field method, dynamic step search, adjust the extension direction of random tree by potential field method to avoid obstacle, by changing expansion step, improve the efficiency and precision of path expansion, compared with other algorithms, the present application has great advantage in the efficiency of path planning and trajectory optimization.

Description

Technical Field

[0001] This invention relates to the field of path planning for lawn mowing robots, and more particularly to a path planning method for lawn mowing robots based on an improved RRT algorithm. Background Technology

[0002] Currently, most urban lawn mowing relies on manual labor, which faces problems such as high labor costs, low work efficiency, and the inability to mow lawns cleanly and neatly. For large lawns, intelligent mobile lawn mowing robots can complete the task accurately, efficiently, and cost-effectively. Existing technologies generate robot paths through random tree expansion using RRT algorithms, such as hybrid RRT, Informed RRT*-Connect, intelligent algorithms based on Weighted Jacobi Fast Exploration Random Tree (WJRRT), motion planning algorithms combining kinematic control and fast exploration random sampling methods, real-time algorithms based on RRT*, goal-oriented (GO) sampling methods for robot motion planning, improved RRT algorithms, and robot arm optimization methods based on genetic algorithms, as well as a robot arm path planning method proposed in existing patent CN113084811A. However, these existing methods have shortcomings in dynamic obstacle avoidance, path planning efficiency, and trajectory optimization. Summary of the Invention

[0003] The purpose of this invention is to provide a path planning method for lawn mowing robots based on an improved RRT algorithm, which improves the efficiency of path expansion and performs trajectory optimization.

[0004] The technical solution adopted by this invention to solve the above-mentioned technical problems is: a path planning method for a lawn mowing robot based on an improved RRT algorithm, comprising the following steps: Step 1: Scan the working environment using a single-line LiDAR to obtain a grid map of the robot's working environment; Step 2: Match multiple LiDARs with local map points to obtain the robot's current position and set the robot's target position; Step 3: Initialize the algorithm by transforming the robot's current position and target position from the global coordinate system to the map coordinate system of the grid map to obtain the robot's starting node Xinit and target node Xgoal, and perform dilation processing on the obstacles to obtain the obstacle grid. Step 4: Set the starting node Xinit as the root node of the random tree, expand the random tree and sample to obtain the random sampling point Xrand, and ensure that the random sampling point Xrand is not inside the obstacle grid. Step 5: Traverse the random tree and calculate the random tree node Xnear that is closest to Xrand; Step 6: Expand Xrand along the Xnear direction to generate a new random tree node Xnew. Then, perform collision detection on the line connecting Xnear and Xnew and the obstacle grid. When the line connecting Xnear and Xnew passes through the obstacle grid, return to step 4 to reselect Xrand. When the line connecting Xnear and Xnew does not pass through the obstacle grid, retain Xnew. Step 7: Use the distance cost function to determine the distance cost between Xnear and Xnew retained in Step 6. If the distance cost function is not the minimum, return to Step 6 to regenerate Xnew. If the distance cost function is the minimum, retain Xnew. Step 8: Using Xnew, which was retained in Step 7, as the parent node, perform random tree expansion towards the target node Xgoal. Each time a new node is searched from the current node, the current node is used as the starting point, and the search is performed within a radius of a single grid length L. When there are no obstacles within the radius of L, the search is performed from the current node along the Xgoal direction with a step size of L to the new node. When there are obstacles within a radius of L, the repulsive force of the obstacle on the current node and the attractive force of the target node Xgoal on the current node are calculated by the potential field method. The resultant force of the repulsive force and the attractive force on the current node is obtained. Then, the node is extended to the new node in the direction of the resultant force with a step size of L / 2. When the target node Xgoal is found within a radius of L, proceed to step 9; Step 9: Starting from the current node when the target node Xgoal is found within a radius of L, search within a radius of L / 3. When there are no obstacles within a radius of L / 3, expand from the current node along the Xgoal direction to a new node with a step size of L / 3, until the target node Xgoal is reached. When there is an obstacle within a radius of L / 3, the repulsive force of the obstacle on the current node and the attractive force of the target node Xgoal on the current node are calculated by the potential field method. The resultant force of the repulsive force and the attractive force on the current node is obtained. Then, the node is extended to the new node with a step size of L / 6 along the direction of the resultant force until the target node Xgoal is reached. Step 10: Smooth the node connection lines obtained in Step 9 using third-order B-spline interpolation to obtain the robot path.

[0005] Preferably, steps 4-10 are repeated multiple times to generate multiple different robot paths based on different random sampling points Xrand each time. Then, the multiple robot paths are evaluated using path length, smoothness, trajectory curvature, and average distance between the path and obstacles as evaluation indicators, and the robot path with the highest evaluation indicator is retained.

[0006] According to the above technical solution, the beneficial effects of the present invention are: This invention addresses the mapping and path planning problems of robots in unknown lawn environments. It proposes a method based on spiral traversal to overlay the map and create a real-time grid map, performing global path planning in a spiral direction. When encountering obstacles, an improved RRT algorithm is used for local path planning. The constructed map exhibits good global consistency. A navigation scheme combining global and local path planning with dynamic obstacle avoidance, along with the potential field method and dynamic step size search, improves overall path optimization efficiency. Obstacle avoidance is achieved by adjusting the expansion direction of the random tree using the potential field method, and the efficiency and accuracy of path expansion are improved by changing the expansion step size. Compared to other algorithms, this invention has significant advantages in path planning efficiency and trajectory optimization. Detailed Implementation

[0007] A path planning method for a lawn mowing robot based on an improved RRT algorithm includes the following steps: Step 1: Scan the working environment using a single-line LiDAR to obtain a grid map of the robot's working environment.

[0008] Step 2: Match the robot's current position with local map points using multiple lidar sensors, and set the robot's target position.

[0009] Step 3: Initialize the algorithm by transforming the robot's current position and target position from the global coordinate system to the map coordinate system of the grid map to obtain the robot's starting node Xinit and target node Xgoal. Then, perform dilation processing on the obstacles to obtain the obstacle grid.

[0010] Step 4: Set the starting node Xinit as the root node of the random tree, expand the random tree and sample to obtain the random sampling point Xrand, and ensure that the random sampling point Xrand is not within the obstacle grid.

[0011] Step 5: Traverse the random tree and calculate the random tree node Xnear that is closest to Xrand.

[0012] Step 6: Expand Xrand along the Xnear direction to generate a new random tree node Xnew. Then, perform collision detection on the line connecting Xnear and Xnew and the obstacle grid. When the line connecting Xnear and Xnew passes through the obstacle grid, return to step 4 to reselect Xrand. When the line connecting Xnear and Xnew does not pass through the obstacle grid, retain Xnew.

[0013] Step 7: Use the distance cost function to determine the distance cost between Xnear and Xnew retained in Step 6. If the distance cost function is not the minimum, return to Step 6 to regenerate Xnew. If the distance cost function is the minimum, retain Xnew.

[0014] Step 8: Using Xnew, which was retained in Step 7, as the parent node, perform random tree expansion towards the target node Xgoal. Each time a new node is searched from the current node, the current node is used as the starting point, and the search is performed within a radius of a single grid length L. When there are no obstacles within the radius of L, the search is performed from the current node along the Xgoal direction with a step size of L to the new node.

[0015] When there are obstacles within a radius of L, the repulsive force of the obstacle on the current node and the attractive force of the target node Xgoal on the current node are calculated using the potential field method. The resultant force of the repulsive and attractive forces on the current node is obtained. Then, the node is extended to a new node along the direction of the resultant force with a step size of L / 2.

[0016] When the target node Xgoal is found within a radius of L, proceed to step 9.

[0017] Step 9: Starting from the current node where the target node Xgoal is found within a radius of L, search within a radius of L / 3. When there are no obstacles within a radius of L / 3, expand from the current node along the Xgoal direction to a new node with a step size of L / 3, until the target node Xgoal is reached.

[0018] When there is an obstacle within a radius of L / 3, the repulsive force of the obstacle on the current node and the attractive force of the target node Xgoal on the current node are calculated by the potential field method. The resultant force of the repulsive force and the attractive force on the current node is obtained. Then, the node is extended to a new node with a step size of L / 6 along the direction of the resultant force until the target node Xgoal is reached.

[0019] Step 10: Smooth the node connection lines obtained in Step 9 using third-order B-spline interpolation to obtain the robot path.

[0020] Then, steps 4-10 are repeated multiple times to generate multiple different robot paths based on different random sampling points Xrand each time. The multiple robot paths are then evaluated using path length, smoothness, trajectory curvature, and average distance between the path and obstacles as evaluation indicators, and the robot path with the highest evaluation indicator is retained.

Claims

1. A path planning method for a lawn mowing robot based on an improved RRT algorithm, characterized in that, Includes the following steps: Step 1: Scan the working environment using a single-line LiDAR to obtain a grid map of the robot's working environment; Step 2: Match multiple LiDARs with local map points to obtain the robot's current position and set the robot's target position; Step 3: Initialize the algorithm by transforming the robot's current position and target position from the global coordinate system to the map coordinate system of the grid map to obtain the robot's starting node Xinit and target node Xgoal, and perform dilation processing on the obstacles to obtain the obstacle grid. Step 4: Set the starting node Xinit as the root node of the random tree, expand the random tree and sample to obtain the random sampling point Xrand, and ensure that the random sampling point Xrand is not inside the obstacle grid. Step 5: Traverse the random tree and calculate the random tree node Xnear that is closest to Xrand; Step 6: Expand Xrand along the Xnear direction to generate a new random tree node Xnew. Then, perform collision detection on the line connecting Xnear and Xnew and the obstacle grid. When the line connecting Xnear and Xnew passes through the obstacle grid, return to step 4 to reselect Xrand. When the line connecting Xnear and Xnew does not pass through the obstacle grid, retain Xnew. Step 7: Use the distance cost function to determine the distance cost between Xnear and Xnew retained in Step 6. If the distance cost function is not the minimum, return to Step 6 to regenerate Xnew. If the distance cost function is the minimum, retain Xnew. Step 8: Using Xnew, which was retained in Step 7, as the parent node, perform random tree expansion towards the target node Xgoal. Each time a new node is searched from the current node, the current node is used as the starting point, and the search is performed within a radius of a single grid length L. When there are no obstacles within the radius of L, the search is performed from the current node along the Xgoal direction with a step size of L to the new node. When there are obstacles within a radius of L, the repulsive force of the obstacle on the current node and the attractive force of the target node Xgoal on the current node are calculated by the potential field method. The resultant force of the repulsive force and the attractive force on the current node is obtained. Then, the node is extended to the new node in the direction of the resultant force with a step size of L / 2. When the target node Xgoal is found within a radius of L, proceed to step 9; Step 9: Starting from the current node when the target node Xgoal is found within a radius of L, search within a radius of L / 3. When there are no obstacles within a radius of L / 3, expand from the current node along the Xgoal direction to a new node with a step size of L / 3, until the target node Xgoal is reached. When there is an obstacle within a radius of L / 3, the repulsive force of the obstacle on the current node and the attractive force of the target node Xgoal on the current node are calculated by the potential field method. The resultant force of the repulsive force and the attractive force on the current node is obtained. Then, the node is extended to the new node with a step size of L / 6 along the direction of the resultant force until the target node Xgoal is reached. Step 10: Smooth the node connection lines obtained in Step 9 using third-order B-spline interpolation to obtain the robot path.

2. The path planning method for a lawn mowing robot based on an improved RRT algorithm according to claim 1, characterized in that: Repeat steps 4-10 multiple times to generate multiple different robot paths based on different random sampling points Xrand each time. Then, evaluate the multiple robot paths using path length, smoothness, trajectory curvature, and average distance between the path and obstacles as evaluation indicators, and retain the robot path with the highest evaluation indicator.