Coal mine underground unmanned vehicle global planning method based on map fast dimension reduction

Abstract

The application discloses a kind of coal mine underground unmanned vehicle global planning methods based on map fast dimension reduction, comprising: using the RRT algorithm based on hybrid sampling records coal mine basic static map, and records roadway width and turnout size;After obtaining task, according to the minimum roadway width and turnout size recorded, design simplified core size and simplified mode, using map fast dimension reduction method for map dimension reduction;In the dimension reduction map, using the alternative impassable self-selected key point in the dimension reduction map completes global route planning;Using the Bezier curve based on slope interpolation to the global route obtained is smoothed, and is output to the decision control system of vehicle.The application carries out fast dimension reduction processing to map on the basis of not destroying high-precision map, can effectively reduce the workload of robot, reduce path planning time cost and save industrial computer power, and has important significance to coal mine underground vehicle unmanned technology landing.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of environmental map processing and unmanned driving technology, and particularly relates to a coal mine underground unmanned vehicle global planning method based on map fast dimension reduction. BACKGROUND

[0002] Coal mining and production is a typical high-risk and arduous industry, and using underground unmanned vehicles to replace existing manually driven vehicles is one of the inevitable ways to realize digital and intelligent production of coal mines, and is also a new development direction in the field of energy exploitation.

[0003] Ground unmanned vehicle technology is currently a popular research content. An unmanned vehicle generally includes a perception module, a decision module, a control module and other components, among which global planning plays a key role, that is, using perception information to obtain the shortest and safest route from the current location of the vehicle to the destination through static decision.

[0004] Patent publication No. CN110609557A discloses a hybrid path planning method for unmanned vehicles. The method combines ant colony algorithm and A* algorithm to realize global path planning by introducing the local optimal solution of the ant colony algorithm into the open table of the A* algorithm. Patent publication No. CN112306067A discloses a global path planning method and system. The global path planning method arranges multiple sampling points in the path planning area, and obtains a passable global path by repeatedly sampling and updating the sampling area in the planning area. Patent publication No. CN110333714A discloses a path planning method and device for unmanned vehicles. The improved artificial potential field-ant colony algorithm is used to plan a global path, and the optimal length cost of the global route is realized.

[0005] The aforementioned patents describe improved path planning methods for unmanned vehicles in ground environments, but these methods have the following drawbacks: 1. Existing methods all address the cost issues such as long path planning calculation time by improving the path planning algorithm, while ignoring the impact of environmental map construction on computational costs. In complex environments, such as underground coal mines, the environmental maps constructed by vehicle perception modules often contain a wealth of information. Furthermore, with the development of machine vision, these maps now include not only basic information for global path planning but also dynamic and complex information for intelligent decision-making, leading to information redundancy in global planning. Secondly, the perception information used for global path planning in surface environments is easily modified, allowing for the acquisition of environmental maps with different information focuses through multiple scans and sampling. However, underground environments are often limited by factors such as network accessibility, driving area safety, and limited available space, making it difficult to generate new maps through multiple environmental samplings in a short period. Vehicles frequently need to use the same environmental map for extended periods. Thirdly, unlike surface environments, underground coal mine roadways are characterized by harsh conditions, high methane levels, numerous interferences, complex personnel, and frequent unforeseen circumstances. Therefore, unmanned vehicles in underground mines typically require explosion-proof electronic and electrical equipment, with limited battery capacity and industrial control computer storage space. This significantly restricts the perception, decision-making, and control performance of unmanned vehicles in coal mines, making it difficult to directly apply most surface systems to underground unmanned vehicles. Summary of the Invention

[0006] The purpose of this invention is to provide a global planning method for unmanned vehicles in coal mines based on rapid map dimensionality reduction. This method can address the complex environment of coal mines, such as harsh conditions, high methane levels, numerous interferences, complex personnel, and frequent emergencies. It eliminates the need for repeated sampling and achieves high-precision map dimensionality reduction through simple calculations on the industrial control computer. By utilizing rapid dimensionality reduction of the environmental map rather than improving the planning algorithm, this method alleviates the resource constraints of unmanned vehicles caused by environmental factors, reduces the pressure on vehicle sensors and batteries, decreases the computing power of the industrial control computer, and enables rapid global planning for unmanned vehicles.

[0007] To achieve the above objectives, the global planning method for unmanned vehicles in underground coal mines based on map-based rapid dimensionality reduction of the present invention includes the following steps:

[0008] S1 uses the RRT algorithm based on hybrid sampling to record a basic high-precision static raster map of the coal mine, and records the roadway width and turnout size;

[0009] S2, After obtaining the task, design a simplified kernel size based on the recorded minimum alleyway width and turnout size, and use the simplified kernel to quickly reduce the dimensionality of the map;

[0010] S3, in the dimension reduction map, judge whether the selected key points including the location of the vehicle, the location of the task and the location of the sub task of the task assignment are passable grids in the dimension reduction map, if they are passable grids, use the path planning method to plan the path to obtain the global route, otherwise, use the reverse automaton based on the greedy strategy to calculate the selected key points and then plan the path to obtain the global route;

[0011] S4, using the Bezier curve based on the slope interpolation to smooth the obtained global route, and outputting to the decision control system of the vehicle.

[0012] Further, in step S2, the process of using the simplified kernel to quickly reduce the dimension of the map includes the following steps:

[0013] S21, determining the size of the simplified kernel according to the high-precision static grid map of S1; specifically including:

[0014] S211, calculating the greatest common divisor σ of the number of long and wide grids of the map;

[0015] S212, obtaining the recorded minimum lane width and the number of grid occupied by the turnout mouth ξ;

[0016] S213, calculating the least common multiple μ of σ and ξ, and taking μ as the length and width size of the simplified kernel;

[0017] S22, determining the simplified kernel parameters according to the type of high-precision static grid map; specifically including:

[0018] S221, determining the type of static grid map according to the grid map information, when the grid value only contains two values, defining the high-precision grid map as a binary map, otherwise defining it as a numerical map;

[0019] S222, standardizing the high-precision grid map. When the map is a binary map, the passable grid value is standardized to 0, and different passable grid values are standardized to 1, at this time the binary map is converted into a 0-1 grid map, the formula is:

[0020]

[0021] In the formula, G(i,j) represents the value of the i-th row and j-th column grid in the original high-precision grid map, and GN(i,j) represents the value of the i-th row and j-th column grid after standardization;

[0022] When the map is a numerical map, the passable grid value range is normalized to [0, 255] through standardization, at this time the numerical map is standardized to a gray scale grid map, the formula is:

[0023]

[0024] wherein G max is the maximum value in the numerical map, G min is the minimum value.

[0025] S23, after the map is standardized, the simplified kernel is used to reduce the dimension of the map. When the simplified kernel is used to reduce the dimension of the map, the simplified kernel traverses the map from the upper left corner of the map model to reduce the dimension. When the current map grid group is reduced in dimension, the nearest non-repeating grid group is selected for fast dimension reduction processing. The traversal ends to obtain the reduced dimension grid map.

[0026] Further, in step S23, the simplified kernel comprises the following definition:

[0027] The simplified kernel is a square matrix with a size of n x n. When the map is standardized to a 0-1 grid map, the formula used by the simplified kernel is:

[0028] f(i,j) = max(max(G(i:(i-s1+1),j:(j-s2+1)))

[0029] wherein G(i:(i-s1+1),j:(j-s2+1)) represents the grid neighborhood data of the i-th row and j-th column grid in the original grid, g i,j represents the data after simplification using maximum feature extraction, and s1 and s2 are the number of rows and columns of the simplified kernel, respectively. When the map is standardized to a gray scale grid map, the formula used by the simplified kernel is:

[0030]

[0031] Further, in step S3, the process of selecting key points by the reverse automaton computer based on the greedy strategy comprises the following steps:

[0032] S31, it is confirmed whether the self-selected key point is reduced in dimension to an impassable grid in the reduced dimension grid map, and the self-selected key point is the grid where the underground vehicle is located, the grid where the task is located, and the specified position that must be passed through by other vehicles. Otherwise, jump out of step S3 and enter step S4, and if yes, enter step S32;

[0033] S32, the evolution range of the reverse automaton is determined. When the dimension reduction range of the self-selected key point is in the middle of the whole map, the evolution range is composed of the reduced dimension grid and the eight neighborhood reduced dimension grids around the self-selected key point. When the dimension reduction range of the self-selected key point is in the middle of the upper, lower, left and right boundaries of the map, the evolution range is composed of the reduced dimension grid and the five neighborhood reduced dimension grids around the self-selected key point. When the dimension reduction range of the self-selected key point is in the middle of the upper, lower, left and right boundaries of the map, Figure 4If the vertex angle is 90°, the evolution range is composed of the selected key point and its three adjacent grids;

[0034] S33, designing the evolution rule of the reverse automaton according to the map type; specifically including:

[0035] S331, when the map is a binary map, setting the cell value of the obstacle grid in the evolution range in step S32 to -1, setting the selected key point as the evolution starting point, setting the cell value of the selected key point to 1, and setting the cell value of other grids to 0. When the map is a numerical map, setting the cell value of the obstacle grid in the evolution range to -1, setting the selected key point as the evolution starting point, setting the cell value of the selected key point to 1, and setting the numerical value of other grids in the evolution range using the following formula:

[0036]

[0037] In the formula, A(g, h) is the cell value of the gth column and hth row in the evolution range, and m and n are the length and width dimensions of the high-precision grid map in step S1, respectively;

[0038] S332, starting the evolution of the reverse automaton from the evolution starting point in step S331, the grid with a cell value of -1 does not participate in the evolution, and the neighbor cell value of the center evolution point is increased by 1 based on the value of the center evolution point until all the evolution grids in the evolution range are evolved;

[0039] S34, obtaining the evolution end point of the reverse automaton using a greedy strategy; specifically including:

[0040] S341, selecting the farthest edge from the evolution starting point in the evolution range, and recording the cell value of the evolved cell with the smallest numerical value except for the obstacle and the grid that cannot be evolved;

[0041] S342, starting from the minimum value cell to deduce the cells with decreasing numerical values in reverse, and using the greedy method and the Euclidean distance to retain the minimum distance cell as the evolution end point;

[0042] S343, if the recorded evolution end point is not unique, calculating the Euclidean distance from the multiple evolution end points to another key point of the path planning, using the greedy method to retain the evolution end point with the shortest distance as the selected key point;

[0043] S35, determining the evolution end point as the selected key point in the reduced dimension map.

[0044] Further, in step S4, the process of smoothing the obtained global route using the Bezier curve based on the slope interpolation includes the following steps:

[0045] S41, the optimal path grid point coordinates planned in the reduced dimension map are converted into corresponding grid point coordinates in the original high-precision grid map, and the formula used is:

[0046] P′ x (i)=P x (i)×μ-(μ-1) / 2

[0047] P′ y (i)=P y (i)×μ-(μ-1) / 2

[0048] In the formula, P x (i) and P y (i) are the optimal path grid point coordinates in the reduced dimension map, μ is the simplified kernel size, P′ x (i) and P′ y (i) are the grid point coordinates in the original high-precision map.

[0049] S42, the selected self-selected key point grid coordinates are added to the path points, and arranged in order from the starting point to the ending point to form the optimal path grid sequence.

[0050] S43, the slope between the second path point and the first and third path points in the optimal path grid sequence is calculated, and then the slope between the third path point and the second and fourth path points is calculated, and so on until the (N-1)th path point, N being the number of path points in the optimal path grid sequence. The formula used is:

[0051] k i1 =(P′ y (i)-P′ y (i-1)) / (P′ x (i)-P′ x (i-1))

[0052] k i2 =(P′ y (i+1)-P′ y (i)) / (P′ x (i+1)-P′ x (i))

[0053] In the formula, k i1 and k i2 represent the slopes of the adjacent two connecting lines of the ith path point, P′ x (i) and P′ y (i) represent the X and Y axis coordinates of the ith path point, respectively.

[0054] S44, when k i1 ≠ k i2At this point, the path points in the optimal path raster sequence are defined as slope abrupt change points, and two path smoothing control points that can affect Bezier smoothing are inserted at both ends of these path points.

[0055]

[0056]

[0057] In the formula, P′ x (i) and P′ y (i) represent the X and Y coordinates of the i-th control point, respectively, and p1(i) and p2(i) represent the interpolation control points at both ends of the i-th control point. Indicates rounding up;

[0058] S45. Insert the path smoothing control points into the optimal path raster sequence according to the path order to obtain the path smoothing control point sequence. Use the Bezier curve to perform smoothing calculation on the path smoothing control point sequence to obtain the smooth path.

[0059] This invention discloses a global planning method for unmanned vehicles in coal mines based on rapid map dimensionality reduction, relating to the fields of environmental map processing and unmanned driving technology. Based on the original high-precision raster map information constructed by sensor sampling, a rapid dimensionality reduction method is used to reduce the dimensionality of the original map while retaining basic information. An inverse automaton is used to compensate for the unavailability of key points caused by map dimensionality reduction, and a Bezier curve path smoothing method based on slope interpolation is used to compensate for the problem of collisions in smooth paths caused by insufficient path points, thereby obtaining a safe and collision-free global smooth path. The specific steps are as follows: A basic static map of the coal mine is recorded using a hybrid sampling-based RRT algorithm, along with the roadway width and turnout dimensions. After obtaining the task, a simplified kernel size and simplified mode are designed based on the recorded minimum roadway width and turnout dimensions. Map dimensionality reduction is then performed using the map dimensionality reduction method described in this invention. In the dimensionality-reduced map, it is determined whether the self-selected key points, including the vehicle's location, the task's location, and the locations of the sub-tasks assigned by the task, are passable grids. If they are passable grids, a path planning method is used to obtain a global route; otherwise, a greedy strategy-based inverse automaton is used to select key points and then perform path planning to obtain a global route. The obtained global route is smoothed using a Bezier curve based on slope interpolation and output to the vehicle's decision control system. This invention can quickly reduce the dimensionality of a map without damaging a high-precision map, instead of using the vehicle environment perception system to perform secondary or multiple sampling and storage of the current basic environment. This can effectively reduce the workload of robots, lower the time cost of path planning, and save the computing power of industrial control computers, which is of great significance for the implementation of unmanned driving technology for underground vehicles in coal mines.

[0060] The beneficial effects of the present application are:

[0061] (1) The global path planning method based on map fast dimension reduction provided by the present application is different from the conventional method of reducing the calculation cost by improving the algorithm. The method can quickly generate a reduced dimension map based on the original map, without the need to resample the environment again, saving the industrial computer computing power, storage space and vehicle power consumption of the underground vehicle, and is particularly suitable for multi-robot splicing map, coal mine underground and other complex path planning with limited vehicle resources;

[0062] (2) The present application provides a new map fast dimension reduction framework, which reduces the dimension of the grid map while retaining useful information, thereby reducing the calculation amount and path planning time;

[0063] (3) The present application designs a reverse automaton based on a greedy strategy, which combines the automaton with the path planning based on the grid map, including the design of the evolution rule, evolution range and evolution target point of the reverse automaton according to the problem requirements of the key point selection in the coal mine unmanned vehicle;

[0064] (4) The present application provides a Bezier curve smoothing method based on slope interpolation, which inserts control points on both sides of the path point connecting two path segments with different slopes. Compared with other Bezier curve smoothing methods, this method has fewer control points and can effectively generate a collision-free smooth path, making up for the shortcomings of the map fast dimension reduction framework. BRIEF DESCRIPTION OF DRAWINGS

[0065] Figure 1 is a main flowchart of a global path planning method based on map fast dimension reduction of the present application;

[0066] Figure 2 is a map fast dimension reduction flowchart of the present application;

[0067] Figure 3 is a map fast dimension reduction process example diagram of the present application;

[0068] Figure 4 is a comparison diagram of the dimension reduction results of the case map of the present application;

[0069] Figure 5 is a reverse automaton flowchart based on a greedy strategy of the present application;

[0070] Figure 6 is an example diagram of the evolution range selection of the reverse automaton of the present application;

[0071] Figure 7 is an example diagram of the evolution rule of the reverse automaton of the present application;

[0072] Figure 8 An example diagram of the selection strategy of the evolution end point of the reverse automaton of the present application is shown in the figure;

[0073] Figure 9 An example diagram of the selection strategy of the evolution end point of the reverse automaton of the present application is shown in the figure;

[0074] Figure 10 A comparison result diagram of obtaining a smooth path using a common Bezier curve and a slope interpolation based Bezier curve of the present application for the implementation case result is shown in the figure. DETAILED DESCRIPTION

[0075] The present application will be further described in detail in combination with the drawings.

[0076] It should be noted that the terms such as "upper", "lower", "left", "right", "front", "back" and the like cited in the application are only for the convenience of clear description, and are not used to limit the scope of the application, and the change or adjustment of the relative relationship is also regarded as the implementation scope of the application without substantial change in technical content.

[0077] The present application provides a coal mine underground unmanned vehicle global planning method based on map fast dimension reduction, which combines Figure 1 The overall scheme flowchart includes the following steps:

[0078] Step 1, record the basic high-precision static grid map of the coal mine underground using the RRT algorithm based on hybrid sampling, and record the roadway width and turnout size.

[0079] Step 2, after obtaining the task, design the simplified kernel size according to the recorded minimum roadway width and turnout size, and use the simplified kernel to perform fast dimension reduction of the map.

[0080] Exemplarily, the map fast dimension reduction flowchart and example diagram are shown in Figure 2 and Figure 3 . Figure 3 In the figure, the simplified kernel starts from the left upper corner area 1 of the map model to perform full coverage type traversal dimension reduction, when the current area 1 grid area is dimension reduced, then area 2 is selected to continue dimension reduction calculation, and so on, until all traversal dimension reductions are performed. The process of using the simplified kernel to perform fast dimension reduction of the map includes the following steps:

[0081] Step 2-1, determine the simplified kernel size according to the high-precision static grid map of step 1; specifically including:

[0082] Step 2-1-1, calculate the greatest common divisor σ of the number of long and wide grids of the map;

[0083] Step 2-1-2, obtain the recorded minimum roadway width and the number of grid occupied by the turnout ξ;

[0084] Step 2-1-3, calculate the least common multiple μ of σ and ξ, and take μ as the length and width dimensions of the simplified kernel;

[0085] Step 2-2, determine the simplified kernel parameters according to the high-precision static grid map type; specifically including:

[0086] Step 2-2-1, determine the type of static grid map according to the grid map information, when the grid values only contain two values, define the high-precision grid map as a binary map, otherwise define it as a numerical map;

[0087] Step 2-2-2, standardize the high-precision grid map. When the map is a binary map, standardize the passable grid value to 0 and the different passable grid value to 1, at this time the binary map is converted into a 0-1 grid map, the formula used is:

[0088]

[0089] In the formula, G(i,j) represents the value of the i-th row and j-th column grid in the original high-precision grid map, and GN(i,j) represents the value of the i-th row and j-th column grid after standardization.

[0090] When the map is a numerical map, the passable grid value range is normalized to [0, 255] through standardization, at this time the numerical map is standardized to a gray-scale grid map, the formula used is:

[0091]

[0092] In the formula, G max is the maximum value in the numerical map, and G min is the minimum value.

[0093] Step 2-3, after standardizing the map, use the simplified kernel to reduce the dimension of the map. When using the simplified kernel to reduce the dimension, the simplified kernel starts from the top left corner of the map model and traverses the map for dimension reduction, when the current map grid group is reduced in dimension, the nearest non-repeating grid group is selected for fast dimension reduction processing, and the traversal ends to obtain the reduced dimension grid map. For example, Figure 4 shows a comparison chart of the dimension reduction results of the case map. Figure 4 (a) is the original map, which belongs to the commonly used 0-1 grid map, the white grid is the passable grid, and the black grid represents that there is an obstacle in the area, using a 3x3 maximum simplified kernel for dimension reduction can obtain a reduced dimension map as shown in Figure 4 (b). As can be seen from the figure, the original map has a dimension of 60x60, and after dimension reduction by the simplified kernel, the dimension of the reduced dimension map becomes 20x20.

[0094] For example, the simplified kernel includes the following definitions:

[0095] The simplified kernel is a square matrix with size n x n. When the map is normalized as a 0-1 grid map, the formula used by the simplified kernel is:

[0096] f(i,j) = max(max(G(i:(i-s1+1),j:(j-s2+1)))

[0097] where G(i:(i-s1+1),j:(j-s2+1)) represents the grid data of the neighborhood of the i-th row and j-th column grid in the original grid, g i,j represents the simplified data using maximum feature extraction, and s1 and s2 are the row and column numbers of the simplified kernel. When the map is normalized as a gray-scale grid map, the formula used by the simplified kernel is:

[0098]

[0099] Step 3, in the reduced dimension map, determine whether the self-selected key points including the position of the vehicle, the position of the task, and the position of the sub-task of the task assignment are passable grids in the reduced dimension map. If they are passable grids, use the path planning method to plan the path to obtain the global route. Otherwise, use the reverse automaton based on the greedy strategy to calculate the key points and then plan the path to obtain the global route.

[0100] For example, the process of using the reverse automaton based on the greedy strategy to calculate the key points is as shown in Figure 5 , which includes the following steps:

[0101] Step 3-1, determine whether the self-selected key points are reduced to impassable grids in the reduced dimension grid map. The self-selected key points are the grid where the position of the vehicle is located, the grid where the task is located, and the grid where the specified position of the other vehicle must pass through. Otherwise, jump from step 3 and enter step 4. If yes, enter step 3-2;

[0102] Step 3-2, determine the evolution range of the reverse automaton. For example, Figure 6 shows one type of evolution range selection in the implementation case. The black grid is the self-selected key point, and the gray area is the corresponding evolution range obtained by calculation. When the reduced dimension range of the self-selected key point is in the middle of the entire map, the evolution range is composed of the reduced dimension grid and its eight neighborhood reduced dimension grids. When the reduced dimension range of the self-selected key point is in the middle of the upper, lower, left, and right boundaries of the map, the evolution range is composed of the reduced dimension grid and its five neighborhood reduced dimension grids. When the reduced dimension range of the self-selected key point is in the top corner of the map, Figure 4 the evolution range is composed of the reduced dimension grid and its three neighborhood reduced dimension grids.

[0103] Step 3-3, design the reverse automaton evolution rule according to the map type; specifically including:

[0104] Step 3-3-1, when the map is a binary map, set the cell value of the obstacle grid in the evolution range of step 3-2 to -1, select the key point as the evolution starting point, set its cell value to 1, and set the cell value of other grids to 0. When the map is a numerical map, set the cell value of the obstacle grid in the evolution range to -1, select the key point as the evolution starting point, set its cell value to 1, and set the numerical value of other grids in the evolution range using the following formula:

[0105]

[0106] In the formula, A(g, h) is the cell value of the gth column and hth row in the evolution range, and m and n are the length and width dimensions of the high-precision grid map in step 1;

[0107] Step 3-3-2, start the reverse automaton evolution from the evolution starting point of step 3-3-1, the grid with cell value -1 does not participate in the evolution, and the neighborhood cell value of the center evolution point is increased by 1 based on the value of the center evolution point until all the evolution range of the evolution grid is evolved.

[0108] Exemplarily, Figure 7 The evolution process of the reverse automaton in the demonstration case is shown. Figure 7 (a) is the cell space initialized according to the existing grid, the key point grid value is initialized to 1, the obstacle grid is initialized to -1, and the other passable grids (key point) are 0. After 3 times of evolution using the evolution rule, the cell space shown in Figure 7 (b) is generated, which can be further evolved to Figure 7 (c) shows the cell space after 5 times of evolution, and when all the passable grids are evolved, the cell space shown in Figure 7 (d) is generated, the obstacle grid and the key point grid do not change the grid value because of the evolution rule, and remain -1 and 1, and the numerical value of the passable grid with grid value 0 in the original evolution area is changed after evolution. Step 3-4, use the greedy strategy to obtain the evolution end point of the reverse automaton; exemplarily, Figure 8 and Figure 9 The evolution end point selection method in the implementation case is shown. Figure 8(a) shows a complete grid map, where black is obstacle grid, white is passable grid, green grid is the current position of the robot (i.e. the starting point), and red grid is the self-selected end point. In this map, after using the simplification operation, the starting point is still a passable grid, but the self-selected end point will become an impassable grid. According to the evolutionary range determination strategy, the evolutionary range of the selected key point is selected, and it can be seen that the lowermost row of the evolutionary range cannot be selected because if it becomes a selected key point, the robot cannot cross it. According to the existing evolution rule, evolution calculation is performed to obtain Figure 8 (b) shows the evolved cell space, from which it can be seen that the lowermost row is 0, i.e. not evolved, which is the advantage of using the reverse automaton selected key point, which can avoid selecting a key point that cannot be crossed. After evolution, the minimum evolution value of the outermost periphery of the evolution range is found except for the obstacle and the unevolved area, such as Figure 8 (c) shows that in this case, it is 4, i.e. 5 selected key points can be obtained. At this time, the shortest evolution distance from the selected key point to the evolution starting point is deduced in reverse by taking the difference value 1 as the step length, from which a plurality of distance values can be calculated, such as Figure 8 (d) shows that assuming the grid length is 1 cm, then d4=3. It can be known from the greedy method that the fourth selected key point is the shortest distance key point, and the fourth selected key point is selected as the selected key point. Figure 9 (a), when the self-selected end point is at a corner and will also be simplified to an impassable grid, the reverse automaton evolution will encounter Figure 9 (b) shows the evolution result, i.e. two selected key points with the shortest distance. At this time, another greedy strategy is introduced, i.e. calculating the Euclidean distance of the two selected key points to another key point, and selecting the selected key point with the shortest Euclidean distance as the final selected key point. For example, Figure 9 (c), the Euclidean distance D1 of the selected key point at the upper left corner of the evolution range to another key point (the starting point of the robot) is less than the Euclidean distance D2 of the selected key point at the lower right corner of the evolution range to another key point, so the selected key point at the upper left corner of the evolution range is selected as the final selected key point in this case.

[0109] Specifically, it includes:

[0110] Step 3-4-1, select the farthest edge from the evolution starting point in the evolution range, and record the evolved cell with the smallest grid value except for the obstacle and the impassable grid;

[0111] Step 3-4-2, start from the minimum value cell and deduce the cells with decreasing values in reverse, and use the greedy method and the Euclidean distance to retain the minimum distance cell as the evolution end point;

[0112] Step 3-4-3, if the evolution end point is not unique, calculate the Euclidean distance from the evolution end point to the next key point of the path planning, and use the greedy method to retain the evolution end point with the shortest distance as the selected key point;

[0113] Step 3-5, determine the evolution end point as the selected key point in the reduced dimension map.

[0114] Step 4, use the Bezier curve based on slope interpolation to smooth the obtained global route, and output to the decision control system of the vehicle.

[0115] Exemplarily, the process of using the Bezier curve based on slope interpolation to smooth the obtained global route includes the following steps:

[0116] Step 4-1, convert the optimal path grid point coordinates planned in the reduced dimension map into corresponding grid point coordinates in the original high-precision grid map, using the formula:

[0117] P′ x (i)=P x (i)×μ-(μ-1) / 2

[0118] P′ y (i)=P y (i)×μ-(μ-1) / 2

[0119] In the formula, P x (i) and P y (i) are the optimal path grid point coordinates in the reduced dimension map, μ is the simplified kernel size, P′ x (i) and P′ y (i) are the grid point coordinates in the original high-precision map;

[0120] Step 4-2, add the selected self-selected key point grid coordinates to the path points, and arrange them in order from the starting point to the ending point to form an optimal path grid sequence;

[0121] Step 4-3, calculate the slope between the 2nd path point and the 1st and 3rd path points in the optimal path grid sequence, then calculate the slope between the 3rd path point and the 2nd and 4th path points, and so on until the (N-1)th path point, N being the number of path points in the optimal path grid sequence. The formula used is:

[0122] k i1 =(P′ y (i)-P′ y (i-1)) / (P′ x (i)-P′ x (i-1))

[0123] ki2 =(P′) y (i+1)-P′ y (i)) / (P′ x (i+1)-P′ x (i))

[0124] In the formula, k i1 and k i2 P′ represents the slope of the line connecting two adjacent segments of the path point i. x (i) and P′ y (i) represent the X and Y coordinates of the i-th path point, respectively;

[0125] Step 4-4, when k i1 ≠k i2 At this point, the path points in the optimal path raster sequence are defined as slope abrupt change points, and two path smoothing control points that can affect Bezier smoothing are inserted at both ends of these path points.

[0126]

[0127]

[0128] In the formula, P′ x (i) and P′ y (i) represent the X and Y coordinates of the i-th control point, respectively, and p1(i) and p2(i) represent the interpolation control points at both ends of the i-th control point. Indicates rounding up;

[0129] Steps 4-5 involve inserting the path smoothing control points into the optimal path raster sequence according to the path order to obtain the path smoothing control point sequence. Then, using a Bezier curve, the path smoothing control point sequence is smoothed to obtain the smoothed path. For example, Figure 10 The diagram shows a comparison of the smooth paths obtained using a standard Bezier curve and the slope-interpolated Bezier curve of this invention in the implementation examples.

[0130] In summary, the present application proposes a coal mine underground unmanned vehicle global planning method based on map fast dimension reduction, which can effectively utilize the original map information to quickly generate a reduced dimension map, without the need to resample the environment again, saving the industrial computer power, storage space and vehicle power consumption of the underground vehicle, especially suitable for multi-robot splicing map, coal mine underground and other complex, limited vehicle resource path planning, which can quickly realize the global path planning of coal mine underground unmanned vehicle. A new map fast dimension reduction framework is proposed, which reduces the dimension of the grid map while retaining useful information, thereby reducing the computational load and path planning time. A reverse automaton based on greedy strategy is designed, which combines the automaton with the path planning based on the grid map, including the problem requirements of selecting key points in the coal mine underground unmanned vehicle. The evolution rules, evolution range and evolution target point of the reverse automaton are designed, which can make up for the problem of unavailable key points caused by map dimension reduction. A Bezier curve smoothing method based on slope interpolation is proposed, which inserts control points on both sides of the path point connecting two path segments with different slopes. Compared with other Bezier curve smoothing methods, this method has fewer control points, can effectively generate collision-free smooth paths, and uses the Bezier curve path smoothing method based on slope interpolation to make up for the problem of collision of smooth paths caused by fewer path points, and obtain a safe and collision-free global smooth path.

[0131] The above is only the preferred embodiment of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical solutions falling within the concept of the present application shall be considered within the protection scope of the present application. It should be noted that, for ordinary skilled persons in the art, some improvements and refinements without departing from the principles of the present application shall be considered within the protection scope of the present application.

Claims

1. A coal mine underground unmanned vehicle global planning method based on map fast dimension reduction, characterized in that, The method comprises the following steps: S1, record a high-precision static grid map of a coal mine underground, and record a roadway width and a turnout size using a hybrid sampling-based RRT algorithm; S2, after obtaining a task, design a simplified kernel size according to the recorded minimum roadway width and turnout size, and use the simplified kernel to perform fast dimension reduction on the map; The simplified kernel comprises the following definitions: The simplified kernel is a square matrix with a size of n*n, and when the map is normalized to a 0-1 grid map, the formula used by the simplified kernel is: f(i,j)=max(max(G(i:(i-s1+1),j:(j-s2+1))) In the formula, G(i:(i-s1+1),j:(j-s2+1)) represents the neighborhood grid data of the i-th row and j-th column grid in the original grid, f(i,j) represents the data after simplification using the maximum feature extraction, and s1 and s2 are the row and column numbers of the simplified kernel, respectively; When the map is normalized to a gray-scale grid map, the formula used by the simplified kernel is: S3, in the reduced dimension map, determine whether the self-selected key points including the position of the vehicle, the position of the task and the position of the subtask of the task are passable grids in the reduced dimension map, if yes, use a path planning method to plan a global route, otherwise, use a reverse automaton calculation machine to select the key points and then plan a global route using a path planning method; S4, use a Bezier curve based on slope interpolation to smooth the obtained global route, and output the global route to a decision control system of the vehicle.

2. The coal mine underground unmanned vehicle global planning method based on map quick dimension reduction according to claim 1, characterized in that, In step S2, the process of using the simplified kernel to perform fast dimension reduction on the map comprises the following steps: S21, determine the size of the simplified kernel according to the high-precision static grid map; specifically comprising: S211, calculate the greatest common divisor σ of the number of long and wide grids of the map; S212, obtain the recorded minimum roadway width and the number of grids occupied by the turnout ξ; S213, calculate the least common multiple μ of σ and ξ, and take μ as the size of the long and wide dimensions of the simplified kernel; S22, determine the simplified kernel parameters according to the type of the high-precision static grid map; specifically comprising: S221, determine the type of the static grid map according to the grid map information, when the grid values only contain two values, define the high-precision grid map as a binary map, otherwise, define it as a numerical map; S222, perform standardization processing on the high-precision grid map; when the map is a binary map, standardize the passable grid values to 0 and the different passable grid values to 1, at this time, the binary map is converted into a 0-1 grid map, and the formula used is: In the formula, G(i,j) represents the value of the i-th row and j-th column grid in the original high-precision grid map, and GN(i,j) represents the value of the i-th row and j-th column grid in the map after standardization processing; When the map is a numerical map, through standardization processing, the passable grid value range is normalized to [0, 255], at this time, the numerical map is normalized to a gray-scale grid map, and the formula used is: where G max is the maximum value in the numerical map, G min is the minimum value; S23, after the map is standardized, the simplified kernel is used to reduce the dimension of the map; when the simplified kernel is used to reduce the dimension, the simplified kernel starts from the upper left corner of the map model to traverse and reduce the dimension of the map; when the current map grid group is reduced in dimension, the nearest non-repeating grid group is selected for fast dimension reduction processing; the traversal is ended to obtain the reduced dimension grid map.

3. The coal mine underground unmanned vehicle global planning method based on map quick dimension reduction according to claim 1, characterized in that, In step S3, the process of using the reverse automaton computer based on the greedy strategy to select the key point includes the following steps: S31, confirming whether the self-selected key point is reduced in dimension to an impassable grid in the reduced dimension grid map, the self-selected key point being a grid where the underground vehicle is located, a grid where the task is located and a grid where the other vehicle must pass through, otherwise, the step S3 is skipped and step S4 is entered, if yes, step S32 is entered; S32, determining the evolution range of the reverse automaton; when the reduced dimension range of the self-selected key point is in the middle of the whole map, the evolution range is composed of the reduced dimension grid and eight adjacent reduced dimension grids around the self-selected key point; when the reduced dimension range of the self-selected key point is in the middle of the upper, lower, left and right boundaries of the map, the evolution range is composed of the reduced dimension grid and five adjacent reduced dimension grids around the self-selected key point; when the reduced dimension range of the self-selected key point is in the four top corners of the map, the evolution range is composed of the reduced dimension grid and three adjacent reduced dimension grids around the self-selected key point; S33, designing the evolution rule of the reverse automaton according to the type of the map, specifically including: S331, when the map is a binary map, setting the cell value of the obstacle grid in the evolution range in step S32 to-1, setting the self-selected key point as the evolution starting point, setting the cell value of the self-selected key point to 1, and setting the cell value of other grids to 0; when the map is a numerical map, setting the cell value of the obstacle grid in the evolution range to-1, setting the self-selected key point as the evolution starting point, setting the cell value of the self-selected key point to 1, and setting the numerical value of other grids in the evolution range by using the following formula: In the formula, A(g, h) is the cell value of the gth column and hth row in the evolution range, and m and n are the length and width dimensions of the high-precision grid map in step S1. S332, starting the evolution of the reverse automaton from the evolution starting point in step S331, the grid with the cell value of-1 does not participate in the evolution, and the adjacent cell smaller than the center evolution point is added by 1 based on the numerical value of the center evolution point until all the evolution grids in the evolution range are evolved; S34, using the greedy strategy to obtain the evolution end point of the reverse automaton; specifically including: S341, selecting the farthest edge from the evolution starting point in the evolution range, and recording the evolution end cell with the smallest grid value except the obstacle and the non-evolution grid; S342, starting from the minimum value cell to deduce the cells with decreasing numerical values in reverse, and using the greedy method and the Euclidean distance to retain the minimum distance cell as the evolution end point; S343, if the recorded evolution end point is not unique, calculating the Euclidean distance from the multiple evolution end points to another key point of the path planning, using the greedy method to retain the evolution end point with the shortest distance as the machine-selected key point; S35, determining the evolution end point as the machine-selected key point in the reduced dimension map.

4. The coal mine underground unmanned vehicle global planning method based on map quick dimension reduction according to claim 1, characterized in that, In step S4, the process of smoothing the obtained global route using the Bezier curve based on the slope interpolation includes the following steps: S41, convert the optimal path grid point coordinates planned in the reduced dimension map into corresponding grid point coordinates in the original high-precision grid map, using the formula: P' x (i) = P x (i) x μ - (μ - 1) / 2 P' y (i) = P y (i) x μ - (μ - 1) / 2 where P x (i) and P y (i) are the optimal path grid point coordinates in the reduced dimension map, μ is the simplified kernel size, and P′ x (i) and P′ y (i) are the grid point coordinates in the original high-precision map. S42, add the selected self-selected key point grid coordinates to the path points, and arrange them in order from the starting point to the ending point to form an optimal path grid sequence; S43, calculate the slope between the second path point and the first and third path points in the optimal path grid sequence, then calculate the slope between the third path point and the second and fourth path points, and so on until the (N-1)th path point, where N is the number of path points in the optimal path grid sequence, using the formula: k i1 = (P' y (i) - P' y (i - 1)) / (P' x (i) - P' x (i - 1)) k i2 = (P' y (i+1) - P' y (i)) / (P' x (i+1) - P' x (i)) where k i1 and k i2 denotes the slope of the line segment connecting the two adjacent path points, P' x (i) and P' y (i) denote the X and Y coordinates of the ith path point, respectively. S44, when k i1 ≠ k i2 define the path point in the optimal path grid sequence as a slope mutation point, and insert two path smooth control points which can affect Bezier smoothing at both ends of the path point where P x (i) and P y (i) represent the X and Y axis coordinates of the i-th path point, and p1(i) and p2(i) represent the interpolation control points at both ends of the i-th path point, represents rounding up. S45, insert the path smoothing control points into the optimal path grid sequence in order to obtain a path smoothing control point sequence, and use the Bezier curve to perform smoothing calculation on the path smoothing control point sequence to obtain a smoothed path.

Images