A global path planning method and device for deep-sea mining vehicles

By constructing a two-dimensional grid map in the deep-sea mining vehicle path planning, combining obstacle influence factors and dynamic target bias probability thresholds, and integrating multiple forces to calculate the combined force, the system avoids local force jamming, achieving smooth and efficient path operation, adapting to complex seabed environments, and reducing energy consumption.

CN123111302APending Publication Date: 2026-11-17HUNAN GUOTIAN ELECTRONICS TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611622192.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-20
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing global path planning methods for deep-sea mining vehicles do not fully consider the complexity of the seabed environment, have weak obstacle adaptability, and insufficient path smoothness, making it difficult to meet the stability and efficiency requirements of deep-sea mining vehicles.

Method used

By constructing a two-dimensional grid map of the seabed operation area, and combining the obstacle influence factor and the distance between the current planned node and the obstacle to dynamically generate the target bias probability threshold, random or target bias sampling is selected. The resultant force is calculated by integrating the target gravity, distance adaptive repulsion and random growth force. The circular arc bypass is used to avoid local force jamming, and collision detection and iterative expansion of the search tree are performed.

Benefits of technology

It improves the efficiency and smoothness of path planning for deep-sea mining vehicles, enhances their adaptability to complex seabed environments, and reduces energy consumption during vehicle turning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123111302A_ABST
    Figure CN123111302A_ABST
Patent Text Reader

Abstract

This application relates to a method and apparatus for global path planning of deep-sea mining vehicles. The method includes: constructing a two-dimensional grid map of the seabed operating area and determining the starting and target positions of the path planning; using the starting position as the root node of the path search tree, initializing the planning node as the root node, generating random probability values, adaptively switching between random or target-biased sampling to generate sampling points; matching and expanding reference nodes in the existing path search tree based on the sampling points, expanding along the direction of the sampling points by a step size to obtain initial candidate path nodes, and calculating the artificial potential force corresponding to the initial candidate path nodes; when the resultant force is zero, correcting by circling around the target side with the initial candidate node as the center; after the candidate node passes collision detection, iteratively expanding the search tree, and outputting the optimal global path after reaching the target area. This method can improve the efficiency and smoothness of path planning for deep-sea mining vehicles.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of global path planning technology for deep-sea mining vehicles, and in particular to a method and apparatus for global path planning of deep-sea mining vehicles. Background Technology

[0002] As the core equipment of the deep-sea mineral resource mining technology system, the rationality of the path planning of the deep-sea mining vehicle directly affects mining efficiency, equipment safety and energy consumption control.

[0003] In deep-sea mining operations, mining vehicles face extremely complex operating environments, including areas with dense obstacles, narrow passages, irregular terrain, and the need for rapid relocation. They must be able to avoid obstacles, travel stably, and reach the target working area in the shortest possible time. Currently, commonly used global path planning algorithms for deep-sea mining vehicles include the Rapidly-exploring Random Tree (RRT) algorithm and the Rapidly-exploring Random Tree Star (RRT) algorithm. Algorithms such as the Artificial Potential Field Method (APF) are used. Traditional RRT algorithms use random sampling to expand the search tree, which suffers from poor search path quality and low computational efficiency. Some sampling points are far from the target area, resulting in tortuous paths and many redundant nodes, increasing the turning frequency and energy consumption of mining vehicles.

[0004] RRT Although the algorithm achieves asymptotic path optimization by reselecting parent nodes and reconnecting nodes, it still uses a random sampling strategy, resulting in slow convergence and long planning time in complex environments. When the APF algorithm is applied alone, it is prone to problems such as unreachable target points and local minima. When the mining vehicle approaches the target area, the repulsive force of the obstacles may exceed the attractive force of the target point, or the net force on the mining vehicle may be zero, making it impossible to continue moving towards the target point.

[0005] It is evident that existing global path planning methods for deep-sea mining vehicles do not fully consider the complexity of the seabed environment, have weak adaptability to obstacles, and lack path smoothness, making it difficult to meet the stability and efficiency requirements of deep-sea mining vehicles in actual operations. Summary of the Invention

[0006] Therefore, it is necessary to provide a global path planning method and device for deep-sea mining vehicles to address the aforementioned technical problems.

[0007] A global path planning method for deep-sea mining vehicles, the method comprising: Construct a two-dimensional grid map of the seabed operation area, and determine the starting and target positions of the path planning within the two-dimensional grid map; Using the starting position as the root node of the path search tree, the planning node is initialized as the root node, and a random probability value is generated. Based on the relationship between the random probability value and the target bias probability threshold, random sampling or target bias sampling with the target position as the sampling point is selected to generate sampling points. The target bias probability threshold is calculated based on the obstacle influence factor and the Euclidean distance between the current planning node and the nearest obstacle. Based on the sampling points, the existing path search tree is matched with extended reference nodes, and initial candidate path nodes are obtained by expanding along the sampling point direction by step size. The artificial potential field force corresponding to the initial candidate path node is calculated. The artificial potential field force includes the attractive force towards the target position, the repulsive force whose intensity is dynamically adjusted according to the distance between the node and the obstacle, and the random growth force. If the artificial potential field force is not zero, the initial candidate path node is used as the candidate path node to be detected; otherwise, the initial candidate path node is used as the center and the path is corrected by circling around the target position in an arc to obtain the corrected candidate path node. After the candidate path node to be detected or the corrected candidate path node is verified to be valid by collision detection, it is added to the path search tree and the current planning node is updated. The process is iteratively expanded until the search tree reaches the target area, at which point the optimal global path is output.

[0008] A global path planning device for deep-sea mining vehicles, the device comprising: The map building module is used to construct a two-dimensional grid map of the seabed operation area and determine the starting and target positions of the path planning within the two-dimensional grid map. The sampling point generation module is used to initialize the planning node as the root node of the path search tree with the starting position as the root node, generate random probability values, and select random sampling or target bias sampling with the target position as the sampling point according to the relationship between the random probability values ​​and the target bias probability threshold to generate sampling points; the target bias probability threshold is calculated based on the obstacle influence factor and the Euclidean distance between the current planning node and the nearest obstacle; The resultant force calculation module is used to match and expand reference nodes in the existing path search tree based on sampling points, expand along the sampling point direction by step size to obtain initial candidate path nodes, and calculate the artificial potential force corresponding to the initial candidate path nodes; the artificial potential force includes attractive force towards the target position, repulsive force with dynamic adjustment of intensity according to the distance between the node and the obstacle, and random growth force. The node acquisition module is used to select the initial candidate path node as the candidate path node to be detected if the artificial potential field force is not zero; otherwise, it corrects the candidate path node by taking the initial candidate path node as the center and circling around it in an arc on the side facing the target position to obtain the corrected candidate path node. The result output module is used to add the candidate path node to be detected or the corrected candidate path node to the path search tree after the collision detection verifies its validity, and to update the current planning node. The process is iteratively expanded until the search tree reaches the target area, and then the optimal global path is output.

[0009] The aforementioned global path planning method and device for deep-sea mining vehicles dynamically generates a target bias probability threshold by combining obstacle influence factors and the distance between the current planned node and the obstacle. It then compares the random probability with the threshold to determine the sampling method, reducing invalid sampling far from the target and accelerating the convergence speed of the search tree. By fusing target attraction, distance-adaptive repulsion, and random growth force into a total resultant force for the initial candidate node, it can autonomously adjust the repulsion intensity according to the distance to the obstacle. It continuously pulls nodes closer to the target using target attraction to accelerate convergence, while maintaining global random expansion capability through random growth force to avoid missing feasible areas. When the candidate node is in complete force equilibrium, it regenerates the point by circling around the target in an arc, avoiding the defect of traditional potential field forces causing the node to be stuck and unable to move forward. It continuously iterates and expands the search tree through collision-verified nodes, adapting to deep-sea operating spaces with dense obstacles and irregular terrain. This invention improves the path planning efficiency and smoothness of deep-sea mining vehicles, enhances adaptability to complex seabed environments, and reduces vehicle turning energy consumption. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating the global path planning method for deep-sea mining vehicles in one embodiment; Figure 2 Here is a three-dimensional topographic map of the seabed mining area in one embodiment; Figure 3 This is a two-dimensional raster topographic map of an undersea mining area in one embodiment; Figure 4 This is a schematic diagram of the kinematic model of a mining vehicle in one embodiment; Figure 5 This is a schematic diagram of the workflow of a global path planning method in one embodiment. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0012] In one embodiment, such as Figure 1 As shown, a global path planning method for deep-sea mining vehicles is provided, including the following steps: Step 102: Construct a two-dimensional grid map of the seabed operation area and determine the starting and target positions for path planning within the two-dimensional grid map.

[0013] A two-dimensional raster map is a discrete digital map formed by uniformly dividing a three-dimensional seabed topography into a grid. The starting position of the path planning is the initial spatial coordinates of the deep-sea mining vehicle when it begins planning. The target position of the path planning is the operational data collection coordinate point that the mining vehicle needs to reach. It can be understood that by completely digitizing the entire seabed operational area using a raster method, the topography distribution can be clearly represented, indicating impassable and obstacle areas. This provides a unified coordinate calculation standard for each subsequent iteration and expansion of nodes, ensuring consistency in the path coordinate calculation benchmark throughout the entire process.

[0014] Step 104: Using the starting position as the root node of the path search tree, initialize the planning node as the root node, generate random probability values, and select random sampling or target bias sampling based on the relationship between the random probability values ​​and the target bias probability threshold to generate sampling points. The target bias probability threshold is calculated based on the obstacle influence factor and the Euclidean distance between the current planning node and the nearest obstacle.

[0015] The root node of the path search tree is the unique initial node of the entire path search tree, and all subsequent nodes are generated from it. The currently planned node is the spatial coordinate of the mining vehicle corresponding to each round of calculating the target bias probability threshold. The random probability value is a value randomly generated within the range of 0 to 1. The target bias probability threshold is the criterion used to switch sampling modes. Target bias sampling is a sampling logic that can dynamically switch between random and directional modes, and the sampling point serves as the directional reference coordinate for node growth. Euclidean distance represents the straight-line distance between two points in space.

[0016] It is understandable that the search tree is initialized with the starting coordinates. In each round, the sampling mode is dynamically matched and adapted according to the obstacle distribution and vehicle position. In open areas of the seabed, the probability of sampling towards the target is automatically increased, and in areas with dense obstacles, the frequency of random sampling across the entire area is increased. This can generate a large number of effective direction references close to the target, reduce invalid sampling that deviates from the target, and speed up the overall speed of the search tree extending towards the target area.

[0017] Step 106: Based on the sampling points, match and expand the baseline nodes in the existing path search tree, expand along the sampling point direction by a step size to obtain initial candidate path nodes, and calculate the artificial potential field forces corresponding to the initial candidate path nodes. The artificial potential field forces include the attractive force towards the target position, the repulsive force whose intensity dynamically adjusts with the distance between the node and the obstacle, and the random growth force.

[0018] The extended baseline node is the closest stored node to the sampling point on the current search tree, serving as the starting point for the outward growth of nodes in this round. The expansion step size is a fixed spatial length for each outward extension. The initial candidate path node is the spatial coordinates initially generated after extending from the extended baseline node, awaiting force verification. The artificial potential force is the total force obtained by superimposing three types of component force vectors. Among them, the target attraction continuously pulls the node closer to the target position, the obstacle repulsion automatically adjusts the magnitude of the force according to the distance between the node and the obstacle, and the random growth force is used to preserve the ability of the search tree to explore the entire domain.

[0019] It is understandable that by using the generated tree nodes to generate nodes to be verified in a directional manner, and simultaneously integrating the three types of forces to comprehensively judge the force state of the nodes, it is possible to simultaneously achieve convergence towards the target, actively move away from obstacles, and conduct full-domain coverage search, thus avoiding the path stagnation caused by a single force.

[0020] Step 108: If the artificial potential field force is not zero, the initial candidate path node is taken as the candidate path node to be detected; otherwise, the initial candidate path node is used as the center and the path is corrected by circling around the target position in an arc to obtain the corrected candidate path node.

[0021] The candidate path nodes to be tested are nodes that are under normal force and do not require adjustment, and can directly participate in the collision verification. A resultant force of zero means that the attractive force of the target and the repulsive force of the obstacle at the node are completely canceled out. The mining vehicle will be unable to move forward if it stays at this point. The arc-shaped detour correction only draws an arc-shaped trajectory on the side closer to the target and reselects the node coordinates. The corrected candidate path node is a new spatial coordinate that is free from the force balance dead zone.

[0022] It is understandable that setting up a one-sided arc-shaped detour adjustment logic for nodes that cannot move forward due to force cancellation, and changing the node position only towards the target side, can quickly get out of the local force balance dead zone, ensuring that the search tree will not stop growing midway and continue to maintain the growth trend of extending towards the target.

[0023] Step 110: After the candidate path node to be detected or the corrected candidate path node is verified to be valid by collision detection, it is added to the path search tree and the current planning node is updated. The process is iteratively expanded until the search tree reaches the target area, and the optimal global path is output.

[0024] Collision detection is used to verify whether the coordinates of candidate nodes fall within obstacle areas. A valid node indicates that the candidate node is within an unobstructed, passable area. Iterative expansion refers to repeatedly executing the entire process of sampling, generating candidate nodes, force correction, and collision verification. The optimal global path is a complete sequence of passable coordinates that connects the starting point to the target and has a smoother node arrangement.

[0025] It is understandable that by incorporating collision-free compliant nodes into the search tree and simultaneously updating the baseline coordinates for the next round of sampling calculations, the node growth is continuously completed in a loop until the target range is reached. This process can continuously accumulate all legally passable nodes and ultimately generate a collision-free, coherent, and complete mining travel path.

[0026] In the aforementioned global path planning method for deep-sea mining vehicles, a target bias probability threshold is dynamically generated by combining obstacle influence factors and the distance between the current planned node and the obstacle. The sampling method is determined by comparing the random probability and the threshold, which reduces invalid sampling far from the target and accelerates the convergence speed of the search tree. By fusing target attraction, distance-adaptive repulsion, and random growth force into the initial candidate node to obtain a total resultant force, the repulsion intensity can be autonomously adjusted according to the distance to the obstacle. The target attraction continuously pulls the node closer to the target to accelerate convergence. Simultaneously, random growth force maintains the ability to randomly expand across the entire domain, avoiding the omission of feasible areas. When the candidate node is completely balanced by force, it regenerates the point by circling around the target in an arc, avoiding the defect of traditional potential field forces causing the node to be stuck and unable to move forward. The search tree is continuously expanded iteratively through collision-verified nodes, adapting to deep-sea operating spaces with dense obstacles and irregular terrain. This invention improves the path planning efficiency and smoothness of deep-sea mining vehicles, enhances adaptability to complex seabed environments, and reduces vehicle turning energy consumption.

[0027] In one embodiment, after constructing the two-dimensional grid map, the method further includes: extending the boundaries of obstacles and impassable areas within the two-dimensional grid map outward along the normal direction to form a safe obstacle avoidance boundary; during the collision detection process, using the safe obstacle avoidance boundary as the judgment benchmark, verifying whether a node collides with an obstacle.

[0028] In this embodiment, the solution sets up an obstacle avoidance assistance module to handle obstacle safety boundaries. After the map is constructed, the obstacle avoidance assistance module uniformly expands the outline outward along the normal direction of the obstacle contour to reserve safe obstacle avoidance space.

[0029] in, The coefficient of thermal expansion is the minimum widened dimension required to satisfy the vehicle's steering and positioning errors. The extent of the effect of the obstacle. The value is equal to the safe expansion radius. , The width of the mining vehicle, based on empirical values. It can be set to 0.5 times the vehicle length, and collision detection uses the generated safety obstacle avoidance boundary and the entire impassable area as the basis for determining passage. By unifying and expanding the safety margin of physical obstacles and directly prohibiting passage in geologically weak areas, it can simultaneously avoid two types of risks: vehicle hardware collisions and vehicle getting stuck, thereby improving the safety of path travel in complex underwater environments.

[0030] In one embodiment, the two-dimensional grid map is obtained by planar projection and grid discretization processing of three-dimensional terrain sensing data of the seabed mining area, including the location information of obstacles, impassable areas and workable areas.

[0031] In this embodiment, the 3D terrain sensing data includes seabed depth and obstacle location data. Planar projection is a process of converting the 3D seabed into a 2D plane. Grid discretization is the process of uniformly dividing the plane into a grid to form a digital map. The workable area is a space free of obstacles and accessible to mining vehicles. By standardizing the 3D sensing data to generate a unified 2D grid map, irregular 3D seabed terrain can be transformed into a unified coordinate calculation carrier, providing a standardized calculation benchmark for the entire sampling process and node collision verification.

[0032] Specifically, by collecting obstacle information (including location, size, and projection area) and impassable areas in the seabed mining area through sensor data, a simulated environment map incorporating terrain elements is constructed, and the coordinates of obstacles, target points, and starting points are output. For example... Figure 2 The 3D topographic map of the seabed mining area shown depicts 5 obstacles (black) and 3 impassable areas (gray). Obstacles are solid objects such as hard seabed reefs and protruding rock formations; sensors can collect their location, size, and planar projection area. Impassable areas are areas without solid protrusions, such as soft mud gullies and low-bearing-capacity seabed, which are navigable. A 1000m × 1000m two-dimensional plane is used to simulate the seabed environment; the 2D raster topographic map of the seabed mining area is shown below. Figure 3 As shown.

[0033] In one embodiment, random sampling or target bias sampling is performed based on the relationship between the random probability value and the target bias probability threshold. Generating sampling points includes: when the generated random probability value is greater than the target bias probability threshold, the target location is used as the sampling point; when the generated random probability value is less than or equal to the target bias probability threshold, coordinate points are randomly selected within the entire seabed operation search space as the sampling points for this operation.

[0034] In this embodiment, the random probability value is a value randomly generated between 0 and 1. The target bias probability threshold is the criterion for switching sampling modes. Sampling points are only used as a reference for node growth direction and are not directly added to the path search tree. By comparing the random probability value and the threshold to adopt two sampling modes, namely directional target selection and global random selection, the number of invalid sampling points far from the target can be reduced, the convergence speed of the search tree extending towards the target area can be accelerated, and the defects of blind sampling and redundant and detours in the traditional RRT algorithm can be improved.

[0035] In one embodiment, the steps for calculating the obstacle impact factor include: obtaining the total projected area of ​​all obstacles in the two-dimensional grid map on the plane and the total projected area of ​​the two-dimensional grid map; and calculating the ratio between the total projected area of ​​the obstacles and the total projected area of ​​the map to obtain the obstacle impact factor.

[0036] In this embodiment, the obstacle influence factor is used to quantify the density of obstacles within the entire map. The total projected area of ​​obstacles is the sum of the areas occupied by all obstacles in the two-dimensional plane, and the total projected area of ​​the entire map is the planar area of ​​the entire grid map. By quantifying the degree of environmental congestion through the ratio of these two factors to generate the influence factor, the sampling judgment criteria can be automatically adapted to the density of obstacles on the seabed. In areas with dense obstacles, the frequency of random sampling across the entire area is increased, while in open areas, the proportion of directional sampling is increased, thereby improving the sampling adaptability to complex seabed environments.

[0037] In one embodiment, matching an extended reference node in the existing path search tree based on the sampling point and expanding along the sampling point direction by a step size to obtain the initial candidate path node includes: traversing the nodes of the current path search tree, selecting the node whose spatial coordinates are closest to the current sampling point as the extended reference node; and taking the extended reference node as the starting point, extending linearly along the vector direction pointing to the sampling point according to a preset expansion step size to generate the initial candidate path node.

[0038] In this embodiment, the extended reference node is the legally stored node closest to the sampling point on the search tree, and it is the starting point for the node's outward growth. The preset extension step size is a fixed spatial length for a single node extension, and the initial candidate path nodes are the points to be verified under force after extension. By traversing all tree nodes to select the reference and generating the points to be verified along the sampling vector with a fixed step size, a standardized node generation logic can be formed, stably producing compliant points to be tested, providing a unified coordinate object for subsequent potential force calculations.

[0039] In one embodiment, the artificial potential force is calculated through a resultant force function; the resultant force function includes a target gravitational component, an obstacle repulsive component, and a random growth component; the obstacle repulsive component includes a repulsive coefficient; the repulsive coefficient changes synchronously with the distance between the node and the obstacle, the closer the node is to the obstacle, the larger the value of the repulsive coefficient, and when the node is far away from the obstacle, the repulsive coefficient drops back to the preset basic repulsive coefficient.

[0040] In this embodiment, the algorithm fusion calculation module is set as the core processing module, integrating the improved RRT algorithm, target bias sampling strategy, and improved artificial potential field algorithm. Node growth guidance is completed by relying on potential field force calculation. The resultant force function is used to superimpose the three types of component forces to obtain the total force on the point. The target gravitational component continuously pulls the node towards the target, the obstacle repulsive component is used to drive away obstacles, the repulsive force coefficient is an adjustable parameter controlling the strength of the repulsive force, the base repulsive force coefficient is the default repulsive force value in an unobstructed scene, and the random growth component is used to retain the global search capability. By coordinating the three forces and adaptively adjusting the repulsive force coefficient according to the obstacle distance, it can continuously guide the node towards the target without hindering convergence due to excessively strong obstacles. Simultaneously, it can also accommodate the exploration of concealed feasible areas such as narrow passages, alleviating the problems of force imbalance and target difficulty in traditional single potential fields.

[0041] In one embodiment, the modified candidate path node is obtained by using an arc-shaped path around the initial candidate path node as the center and moving towards the target location. This process involves: drawing a semi-circular virtual path around the initial candidate path node as the center and using the sum of the track width and track gauge of the deep-sea mining vehicle as the radius on one side close to the target location; and selecting coordinate points along the semi-circular virtual path as the modified candidate path node.

[0042] In this embodiment, the sum of the track width and track gauge of the deep-sea mining vehicle is the lateral characteristic dimension of the entire vehicle. The semi-circular virtual detour trajectory is drawn only on the side closest to the target, and the corrected candidate path nodes are new points that have escaped the force balance dead zone. When the node's force is completely canceled and it cannot move forward, a new point is selected along a single-sided semi-circle, which can quickly escape the local minimum dead zone of the artificial potential field, ensuring that the path search tree continues to grow outward and avoiding stagnation in the planning process.

[0043] In one embodiment, when the search tree reaches the target area, outputting the optimal global path includes: starting from the tree node corresponding to the target location, backtracking along the parent node link of the search tree to the root node to obtain the initial global path node sequence; removing redundant collinear nodes in the initial global path node sequence; and performing smooth interpolation on the path after removing redundant nodes to generate the optimal global path with continuous curvature.

[0044] In this embodiment, the solution includes a path output module to perform post-processing. Parent node links represent the hierarchical relationships between nodes within the search tree. Redundant collinear nodes are excess transition points where three points lie on the same straight line. Smooth interpolation is used to optimize path inflection points, and the curvature-continuous path adapts to the turning and driving characteristics of the tracked mining vehicle. By backtracking the complete path, removing redundant nodes, and performing smooth interpolation, the number of turns the mining vehicle makes during its journey is reduced, driving energy consumption is lowered, path smoothness is improved, and the shortcomings of traditional algorithms in terms of path twisting and bumping are addressed. Simultaneously, this solution includes a kinematic model module, establishing a coordinate system with the mining vehicle's center of mass as the origin. A schematic diagram of the mining vehicle's kinematic model is shown below. Figure 4 As shown, C1 represents CENTER (vehicle geometric center), and C2 represents COM (vehicle center of mass). The kinematic state-space equations are constructed using the speed difference between the left and right tracks. The state-space equations are as follows:

[0045] Where L represents the sum of the track width and track gauge of the mining vehicle. and These represent the linear velocities of the left and right wheels, respectively. It is the steering angle when the mining vehicle turns along the ICR (instantaneous steering center). This equation is used to verify the feasibility of the path movement and finally outputs the optimal global path that satisfies the differential steering constraint of the mining vehicle and the curvature continuity. It can reduce the number of turns of the mining vehicle during the journey, reduce driving energy consumption, and solve the defects of the traditional algorithm path that is tortuous and bumpy.

[0046] In one specific embodiment, such as Figure 5 The diagram illustrates the workflow of a global path planning method. The global path planning method of this invention includes the following steps: Step 1: Environmental Modeling and Parameter Initialization. A grid map of the seabed mining area is constructed using the environmental modeling module, determining the starting point. Target point Initialize algorithm parameters, including the security threshold. Expanding step size Basic gravitational coefficient, basic repulsive coefficient Target bias probability threshold The sum of the width of the mining vehicle's tracks and the track gauge, etc.

[0047] Table 1. Parameter meanings and numerical examples

[0048] Step 2: Target bias sampling. Generate random probability values ​​between 0 and 1. ,like Then the sampling point is set as the target point; if Sampling points are then randomly generated in the search space. The target bias probability threshold is used as an example. Related to the distance to obstacles, the calculation formula is:

[0049] The obstacle influence factor, It is the Euclidean distance between the obstacle and the position of the mining vehicle.

[0050]

[0051] It is the area of ​​the obstacle projected onto a two-dimensional plane. It is the area of ​​the environment projected onto a two-dimensional plane.

[0052] Traverse all nodes of the current path search tree and filter out the spatial coordinates distance to the sampling points. The nearest node is used as the extended baseline node. To expand the baseline node Starting from the point, proceed along the path pointing to the sampling point. The vector direction, according to the preset expansion step size Perform linear extension to generate initial candidate path nodes. .

[0053] Step 3: Improve the calculation of artificial potential field forces. Generate new nodes at the sampling points. Then, calculate the resultant force on the node. Including random growth function ( Obstacle repulsion function and target gravitational function ,Right now:

[0054] In the formula: To determine the resultant force function of new nodes in the random tree during the mining vehicle path search; A stochastic growth function for the new location of the mining vehicle; Let the repulsive force function of the obstacle be denoted as . Let be the target gravitational function.

[0055] The repulsion function introduces an adaptive repulsion coefficient. , The value increases when the mining vehicle approaches an obstacle and returns to its default value when it moves away, to avoid excessive repulsive force affecting path convergence; the gravity function guides nodes to approach the target point.

[0056] The formulas for calculating each function are as follows:

[0057]

[0058]

[0059] In the formula: Step size; It is an adaptive repulsive force coefficient; This is the basic gravity coefficient, which defaults to 1. This represents the effective range of an obstacle's influence, indicating the safe threshold distance at which an obstacle affects the path. Let be the vector pointing from the obstacle to the mining vehicle's location node; It is a unit vector pointing to the current node; This is a unit vector pointing to the target; Mining vehicle location node and target node The absolute value of the Euclidean distance.

[0060] repulsive force coefficient The calculation formula is:

[0061] In the formula The basic repulsion coefficient represents the magnitude of an obstacle's influence by default, typically set to 1.

[0062] Calculate the resultant force using the parallelogram law. The direction of the resultant force is calculated, and then the final extension direction of the initial candidate nodes is corrected using the resultant force vector direction to ensure that the nodes continue to extend towards the target point, guaranteeing a smooth path. If a new obstacle is detected during the extension process, the resultant force direction of the current node is recalculated, the extension path is adjusted, and the obstacle is actively avoided. This indicates that the node is not trapped in a local minimum. They are directly used as candidate path nodes to be detected.

[0063] Step 4: Local Minimum Handling. If a new node... The combined force Then, with that node as the center, A virtual semicircle is set for the radius, and the sampling reference point is selected and corrected by extending the arc trajectory. , and then Starting from the direction Extend the direction by the step size to generate corrected candidate path nodes. , escape the local minimum region.

[0064] Step 5: Collision Detection and Node Expansion. Identify new nodes. or Check if there is a collision with an obstacle (including the expanded boundary). If there is no collision, add it to the search tree; if there is a collision, repeat step two.

[0065] Step Six: Path Generation and Optimization. Repeat steps two through five until the search tree expands to the target area. Use the path output module to remove redundant nodes and smooth the path to generate the optimal global path. If the target point is not reached after reaching the maximum number of iterations, adjust the parameters and replan.

[0066] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0067] In one embodiment, a global path planning device for a deep-sea mining vehicle is provided, comprising: The map building module is used to construct a two-dimensional raster map of the seabed operation area and determine the starting and target positions for path planning within the two-dimensional raster map. The sampling point generation module is used to initialize the planning node as the root node of the path search tree with the starting position as the root node, generate random probability values, and select random sampling or target bias sampling with the target position as the sampling point based on the relationship between the random probability values ​​and the target bias probability threshold to generate sampling points; the target bias probability threshold is calculated based on the obstacle influence factor and the Euclidean distance between the current planning node and the nearest obstacle; The resultant force calculation module is used to match and expand reference nodes in the existing path search tree based on sampling points, expand along the sampling point direction by step size to obtain initial candidate path nodes, and calculate the artificial potential field force corresponding to the initial candidate path nodes; the artificial potential field force includes the attractive force towards the target position, the repulsive force whose intensity is dynamically adjusted according to the distance between the node and the obstacle, and the random growth force. The node acquisition module is used to select the initial candidate path node as the candidate path node to be detected if the artificial potential field force is not zero; otherwise, it corrects the candidate path node by taking the initial candidate path node as the center and moving around it in an arc on the side facing the target position to obtain the corrected candidate path node. The results output module is used to add candidate path nodes to be detected or corrected candidate path nodes to the path search tree after collision detection verifies their validity, and to update the current planning node. The process is iteratively expanded until the search tree reaches the target area, at which point the optimal global path is output.

[0068] Specific limitations regarding the global path planning device for deep-sea mining vehicles can be found in the limitations of the global path planning method for deep-sea mining vehicles described above, and will not be repeated here. Each module in the aforementioned global path planning device for deep-sea mining vehicles can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0069] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0070] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for global path planning of a deep-sea mining vehicle, characterized in that, The method includes: Construct a two-dimensional grid map of the seabed operation area, and determine the starting and target positions of the path planning within the two-dimensional grid map; Using the starting position as the root node of the path search tree, the planning node is initialized as the root node, and a random probability value is generated. Based on the relationship between the random probability value and the target bias probability threshold, random sampling or target bias sampling with the target position as the sampling point is selected to generate sampling points. The target bias probability threshold is calculated based on the obstacle influence factor and the Euclidean distance between the current planning node and the nearest obstacle. Based on the sampling points, the existing path search tree is matched with extended reference nodes, and initial candidate path nodes are obtained by expanding along the sampling point direction by step size. The artificial potential field force corresponding to the initial candidate path node is calculated. The artificial potential field force includes the attractive force towards the target position, the repulsive force whose intensity is dynamically adjusted according to the distance between the node and the obstacle, and the random growth force. If the artificial potential field force is not zero, the initial candidate path node is used as the candidate path node to be detected; otherwise, the initial candidate path node is used as the center and the path is corrected by circling around the target position in an arc to obtain the corrected candidate path node. After the candidate path node to be detected or the corrected candidate path node is verified to be valid by collision detection, it is added to the path search tree and the current planning node is updated. The process is iteratively expanded until the search tree reaches the target area, at which point the optimal global path is output.

2. The method of claim 1, wherein, The step of selecting random sampling or target bias sampling based on the relationship between the random probability value and the target bias probability threshold, and generating sampling points, includes: When the generated random probability value is greater than the target bias probability threshold, the target position is used as the sampling point; When the generated random probability value is less than or equal to the target bias probability threshold, a coordinate point is randomly selected as the sampling point in the entire seabed operation search space.

3. The method of claim 1, wherein, The calculation steps for the obstacle impact factor include: Obtain the total projected area of ​​all obstacles in the 2D grid map on the plane and the total projected area of ​​the entire 2D grid map; The obstacle influence factor is obtained by calculating the ratio between the total projected area of ​​the obstacles and the total projected area of ​​the entire map.

4. The method of claim 1, wherein, Based on the sampling points, the existing path search tree is matched with extended baseline nodes, and the initial candidate path nodes are obtained by expanding along the sampling point direction by a step size, including: Traverse the nodes of the current path search tree and select the node whose spatial coordinates are closest to the current sampling point as the extended reference node; Starting from the extended reference node, the path extends linearly along the vector direction pointing to the sampling point according to the preset extension step size to generate the initial candidate path node.

5. The method according to claim 1, characterized in that, The force in the artificial potential field is calculated using the resultant force function. The resultant force function includes a target gravitational component, an obstacle repulsive component, and a random growth component; the obstacle repulsive component includes a repulsive coefficient; the repulsive coefficient changes synchronously with the distance between the node and the obstacle, the closer the node is to the obstacle, the larger the value of the repulsive coefficient, and when the node is far away from the obstacle, the repulsive coefficient drops back to the preset basic repulsive coefficient.

6. The method according to claim 1, characterized in that, The correction is performed by using the initial candidate path node as the center and circling around it in an arc on the side facing the target position. The corrected candidate path nodes include: Using the initial candidate path node as the center and the sum of the track width and track gauge of the deep-sea mining vehicle as the radius, draw a semi-circular virtual detour trajectory on one side close to the target position. Select coordinate points along the semi-circular virtual trajectory as candidate path nodes after correction.

7. The method according to claim 1, characterized in that, When the search tree reaches the target region, the optimal global path output includes: Starting from the tree node corresponding to the target location, backtrack along the parent node link of the search tree to the root node to obtain the initial global path node sequence; Remove redundant collinear nodes from the initial global path node sequence; The path after removing redundant nodes is smoothed by interpolation to generate the optimal global path with continuous curvature.

8. The method according to claim 1, characterized in that, After constructing the two-dimensional raster map, the following steps are also included: The boundaries of obstacles and impassable areas within a two-dimensional grid map are extended outward along the normal direction to form a safe obstacle avoidance boundary. During the collision detection process, the safety obstacle avoidance boundary is used as the judgment benchmark to verify whether the node collides with the obstacle.

9. The method according to claim 1, characterized in that, The two-dimensional grid map is obtained by planar projection and grid discretization of three-dimensional terrain sensing data of the seabed mining area, and includes the location information of obstacles, impassable areas and workable areas.

10. A global path planning device for a deep-sea mining vehicle, characterized in that, The device includes: The map building module is used to construct a two-dimensional grid map of the seabed operation area and determine the starting and target positions of the path planning within the two-dimensional grid map. The sampling point generation module is used to initialize the planning node as the root node of the path search tree with the starting position as the root node, generate random probability values, and select random sampling or target bias sampling with the target position as the sampling point according to the relationship between the random probability values ​​and the target bias probability threshold to generate sampling points; the target bias probability threshold is calculated based on the obstacle influence factor and the Euclidean distance between the current planning node and the nearest obstacle; The resultant force calculation module is used to match and expand reference nodes in the existing path search tree based on sampling points, expand along the sampling point direction by step size to obtain initial candidate path nodes, and calculate the artificial potential force corresponding to the initial candidate path nodes; the artificial potential force includes attractive force towards the target position, repulsive force with dynamic adjustment of intensity according to the distance between the node and the obstacle, and random growth force. The node acquisition module is used to select the initial candidate path node as the candidate path node to be detected if the artificial potential field force is not zero; otherwise, it corrects the candidate path node by taking the initial candidate path node as the center and circling around it in an arc on the side facing the target position to obtain the corrected candidate path node. The result output module is used to add the candidate path node to be detected or the corrected candidate path node to the path search tree after the collision detection verifies its validity, and to update the current planning node. The process is iteratively expanded until the search tree reaches the target area, and then the optimal global path is output.