An unmanned ship path planning method based on information entropy field guidance

Abstract

The application provides an unmanned ship path planning method based on information entropy field guidance, and relates to the technical field of unmanned ship navigation. The method first acquires a target region image, labels obstacles, and calibrates the starting point and the ending point of the unmanned ship path; then designs an information entropy value function and gives each position in the target region an initial information entropy value to form an initial information entropy field; iteratively searches the unmanned ship path by using an intelligent sampling strategy combined with an RRT algorithm, and completes initial unmanned ship path planning; optimizes the initial unmanned ship path in four steps to obtain an optimized path, combines a hybrid memory module to accumulate historical path experience, then updates the information entropy value by local enhancement and global attenuation, and outputs the historical optimal unmanned ship path after repeated iteration. The application can comprehensively improve the efficiency, accuracy and practicality of the static obstacle path planning of the unmanned ship in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of unmanned vessel navigation technology, and specifically to an unmanned vessel path planning method based on information entropy field guidance. Background Technology

[0002] Currently, unmanned surface vessels (USVs) are increasingly used in tasks such as marine mapping, port patrol, and waterway monitoring. A core challenge in their practical application is real-time and reliable path planning in complex aquatic environments. Existing mainstream methods, such as Rapid Expanding Random Tree (RRT) and its asymptotically optimal variant RRT*, face numerous challenges in practical applications.

[0003] First, the search space exploration is inherently blind. Traditional RRT algorithms employ a globally uniform random sampling strategy, which fails to effectively identify "structurally critical areas" such as obstacle boundaries and narrow passages that decisively influence path connectivity. This results in a large number of sampling points falling into unobstructed open water, causing a severe waste of computational resources and significantly slowing down the effective extension of the random tree to the target point. Second, the feasibility and trackability of the generated paths are poor. Existing methods are mostly based on the geometric shortest path criterion, failing to fully couple the kinematic constraints of the unmanned vessel as an underactuated system. The frequent sharp turns and jagged trajectories in the planning results not only greatly increase the tracking difficulty of the underlying controller but may also cause the actual trajectory to deviate from the planned path due to the vessel's large inertia, posing a potential collision risk. Finally, there is a lack of mechanisms for reusing historical mission experience. In repetitive application scenarios such as fixed-route patrols, traditional algorithms treat each planning as an independent event, lacking a mechanism for storing and reusing the environmental topology information contained in historical successful paths. This "repeated exploration" problem is severely out of sync with the real-time requirements of unmanned vessels for rapid mission response.

[0004] Although existing research attempts to improve traditional path planning algorithms through heuristic functions or post-processing smoothing, most of them can only alleviate single problems locally and fail to make systematic innovations from multiple levels such as environmental cognition, search strategies and experience learning, making it difficult to achieve synergistic optimization of efficiency and quality in complex environments. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide an unmanned vessel path planning method based on information entropy field guidance to quantify environmental uncertainty.

[0006] Technical solution: A path planning method for unmanned surface vessels based on information entropy field guidance, comprising the following steps:

[0007] S1. Acquire the image of the target area and perform binarization processing, mark the obstacles in the target area, and set the starting and ending coordinates of the unmanned vessel path in the target area.

[0008] S2. Design an information entropy function based on the Euclidean distance between each location in the target area and the nearest obstacle, the starting coordinates, and the ending coordinates. Use the information entropy function to assign an initial information entropy value to each location in the target area to form an initial information entropy field.

[0009] S3. An intelligent sampling strategy is adopted, which integrates target bias sampling, historical utilization sampling and information entropy field guided sampling. The RRT algorithm is used to iteratively search the path of the unmanned vessel, generate new nodes and perform collision detection. After the new node is added to the RRT tree, the information entropy value of the surrounding area of ​​the new node is decayed and updated. When the new node enters the preset range of the destination coordinates, the initial unmanned vessel path planning is completed.

[0010] S4. The initial unmanned vessel path is pruned and optimized, and then adaptive smoothing is performed based on information entropy guidance. The smoothed unmanned vessel path is fitted with a B-spline curve to obtain the optimized unmanned vessel path.

[0011] S5. Positively enhance the information entropy value of the area surrounding the optimized unmanned vessel path, while simultaneously decaying and updating the information entropy value within the target area.

[0012] S6. Repeat steps S3 to S5 until the obtained optimized unmanned vessel path meets the index threshold or reaches the maximum number of iterations, and output the historical best unmanned vessel path.

[0013] Specifically, the formula for calculating the information entropy value function is as follows:

[0014]

[0015]

[0016]

[0017] In the formula: The information entropy value function, Let be the distance entropy function. Influence function as the starting point As the endpoint influence function, The coordinates of the position to be calculated, As the starting coordinates, The coordinates of the endpoint , These are the weighting coefficients.

[0018] Specifically, the distance entropy function first calculates the Euclidean distance from each location in the target area to the nearest obstacle using a distance scanning algorithm, and then calculates the distance using a piecewise function based on the range of Euclidean distances. The calculation formula is as follows:

[0019]

[0020] In the formula: The piecewise distance entropy function, For Euclidean distance, The first distance threshold, The second distance threshold is given, and , The basic entropy coefficient, The maximum entropy gain in the mid-range region. This represents the peak entropy value for a long-distance region. This represents the rate of entropy decay over long distances.

[0021] Specifically, the optimized unmanned surface vessel (USV) path is stored using a hybrid memory module, which includes a topology memory module and a path cache pool. The topology memory module identifies path segments with curvature greater than a threshold in the USV path through curvature analysis and stores them as critical path segments. The critical path segments are ranked by value, and the last critical path segment in the ranking is removed from the topology memory module when the storage limit is exceeded. The path cache pool stores each path segment in the optimized USV path and sets an initial confidence probability for each path segment. In subsequent iterations, the confidence probability of path segments that are successfully reused increases, while the confidence probability of path segments that are not reused multiple times decreases. The path segments in the path cache pool are ranked from highest to lowest confidence probability, and the last path segment in the ranking is removed from the path cache pool when the storage limit is exceeded.

[0022] Specifically, ranking the critical path segments by value includes scoring the critical path segments using a comprehensive value assessment formula, as follows:

[0023]

[0024] In the formula: For rating, , , For parameter weights, The average information entropy value of the path segment. The path segment length, The mean curvature.

[0025] Specifically, step S5 also includes: storing the key path segments in the optimized unmanned vessel path into the topology memory module, storing each path segment in the optimized unmanned vessel path into the path cache pool, and reordering them respectively to complete the dynamic update of the hybrid memory module.

[0026] Specifically, the intelligent sampling strategy includes: assigning different sampling probabilities to different sampling methods; generating sampling points directly in the area surrounding the endpoint in target-biased sampling to shorten the search distance at the end of the unmanned vessel path to accelerate path closure; randomly selecting reference nodes from the optimized unmanned vessel path using historical sampling and applying two-dimensional Gaussian perturbations around them to generate new sampling points; and using the inverse transformation of the cumulative distribution function to generate sampling points in information entropy field-guided sampling.

[0027] Specifically, the sampling probability formula for target-biased sampling is:

[0028]

[0029] In the formula: The sampling probability for target-biased sampling. The maximum number of iterations, and It is a preset constant, and ;

[0030] The sampling probability formula for historical sampling is:

[0031]

[0032] In the formula: The sampling probability of historical sampling. For the number of iterations, and It is a preset constant, and ;

[0033] The sampling probability formula for information entropy field-guided sampling is:

[0034]

[0035] In the formula: The sampling probability is used to guide the sampling of the information entropy field.

[0036] Specifically, step S4 includes the following sub-steps:

[0037] S41. Perform topology preservation pruning on the initial unmanned vessel path. Starting from the path start point, directly connect the next node after the original first node. If the distance from the first node to the line connecting the start point and the next node is less than a set threshold, the topology is considered to remain unchanged, and the first node is deleted. This process is repeated to complete the first simplification of the unmanned vessel path length.

[0038] S42. Maintain the pruned unmanned vessel path for the topology, filter nodes based on their turning angles, and while retaining the start and end points, filter out the nodes with the largest turning angles according to a set ratio, and remove the remaining nodes to further simplify the unmanned vessel path structure.

[0039] S43. The adaptive smoothing intensity is dynamically adjusted based on the local information entropy value at the node, using the following formula:

[0040]

[0041] In the formula: For smooth strength, , To set a constant, For nodes Information entropy value;

[0042] The path of the unmanned vessel is smoothed according to the adaptive smoothing intensity;

[0043] S44. Perform B-spline curve fitting on the smoothed unmanned vessel path, and adjust the fitting parameters according to the distribution of local information entropy values ​​of the nodes: use the first fitting order in the region where the information entropy value is higher than the set threshold to preserve path details, and use the second fitting order in the region where the information entropy value is lower than the set threshold to enhance the smoothing effect. The first fitting order is less than the second fitting order. The optimized unmanned vessel path is obtained after fitting.

[0044] Specifically, step S5 includes:

[0045] The information entropy value of the area surrounding the optimized unmanned surface vessel path is positively reinforced, and the update formula is as follows:

[0046]

[0047] In the formula: This is the information entropy value after positive reinforcement. The information entropy value before the update. To enhance the learning rate update for the entropy field, For the range of influence parameters, The upper limit of entropy;

[0048] The information entropy value within the target area is decayed and updated using the following formula:

[0049]

[0050] In the formula: To decay the updated information entropy value, The global decay rate of the entropy field. This is the minimum entropy value.

[0051] Beneficial effects: Compared with the prior art, the significant effects of the present invention are:

[0052] 1. This invention guides sampling by constructing an information entropy field within the target area, focusing the search on high-value areas with high information entropy values, avoiding blind searching, and significantly improving the algorithm's convergence speed.

[0053] 2. This invention employs a multi-stage path optimization strategy to ensure a smooth path with no redundant nodes, conforming to the kinematic characteristics of unmanned vessels, reducing control difficulty and energy consumption.

[0054] 3. This invention constructs a hybrid memory module that accumulates historical path experience and dynamically updates the entropy field to adapt to environmental changes, maintaining a high success rate and stable performance in complex scenarios such as narrow waterways, dense obstacles, and real sea areas.

[0055] 4. As the number of trials increases, the overall algorithm provided by this invention can continuously accumulate experience through the hybrid memory module, and the planning efficiency and path quality are continuously optimized, making it suitable for frequently repeated planning scenarios (such as port patrols, fixed route delivery, etc.). Attached Figure Description

[0056] Figure 1 This is a flowchart of the method of the present invention.

[0057] Figure 2 This is a three-dimensional visualization diagram of the initial information entropy field of this invention.

[0058] Figure 3 This is a two-dimensional visualization diagram of the initial information entropy field of this invention.

[0059] Figure 4 This is a schematic diagram of the multi-strategy fusion sampling of the present invention.

[0060] Figure 5 This is a schematic diagram of the path optimization process of the present invention.

[0061] Figure 6 This is a schematic diagram of the simulation results of the E-RRT method of this invention.

[0062] Figure 7 This is a schematic diagram of the simulation results of the RRT method.

[0063] Figure 8 This is a schematic diagram of the simulation results of the RRT* method. Detailed Implementation

[0064] A preferred embodiment of the present invention will be further described below with reference to the accompanying drawings.

[0065] Example 1

[0066] Please see Figure 1 As shown, this embodiment provides a path planning method for unmanned surface vessels based on information entropy field guidance, including the following steps:

[0067] S1. Acquire the image of the target area and perform binarization processing, mark the obstacles in the target area, and set the starting and ending coordinates of the unmanned vessel path in the target area.

[0068] Based on the actual application scenario, input the actual map parameters of the target area to accurately define the latitude and longitude boundaries. Taking a real sea scene as an example, a binary map is obtained by processing the high-resolution nautical chart of the target area (marking non-navigable areas as black and navigable waters as white). At the same time, the obstacle information in the target area is accurately marked, including the location coordinates and physical attributes of the obstacles (such as whether they can be crossed). The spatial positions of the path start and end points are marked in the target area.

[0069] S2. Design an information entropy function based on the Euclidean distance between each location in the target area and the nearest obstacle, the starting coordinates, and the ending coordinates. Use the information entropy function to assign an initial information entropy value to each location in the target area to form an initial information entropy field.

[0070] The information entropy field is the core guiding mechanism of this method, used to quantify the exploration value and uncertainty of each location in the environment, providing a basis for intelligent sampling. Its core idea is that different locations in the environment contribute differently to path planning; therefore, by quantifying this difference through information entropy values, the sampling process is guided to focus on high-value areas, reducing ineffective searches.

[0071] The formula for calculating the information entropy value function can be expressed as:

[0072]

[0073]

[0074]

[0075] In the formula: The information entropy value function, Let be the distance entropy function. Influence function as the starting point As the endpoint influence function, The coordinates of the position to be calculated, As the starting coordinates, The coordinates of the endpoint , In this embodiment, the weighting coefficient is used. , The design that the endpoint has a greater impact weight than the starting point highlights the goal-oriented nature of path planning and helps the algorithm converge to the goal quickly in the later stages of the search.

[0076] The distance entropy function first calculates the Euclidean distance from each location in the target area to the nearest obstacle using a distance scanning algorithm. Experimental studies show that the distance to the obstacle has a non-linear relationship with the exploration value, and linear or exponential functions cannot accurately describe this relationship. Therefore, this invention uses a piecewise function to model the contribution of distance entropy, dividing the distance into extremely close, medium, and long distance ranges, and assigning different entropy calculation logic to each. The calculation formula is as follows:

[0077]

[0078] In the formula: The piecewise distance entropy function, For Euclidean distance, The first distance threshold, The second distance threshold is given, and , The basic entropy coefficient, The maximum entropy gain in the mid-range region. This represents the peak entropy value for a long-distance region. This represents the rate of entropy decay over long distances.

[0079] In this embodiment, , , , , , .

[0080] To ensure path safety and visualization effectiveness, an obstacle masking mechanism is introduced, forcibly setting the information entropy value of the obstacle region to 0. This avoids the algorithm sampling in obstacle regions and prevents obstacles from being occluded during information entropy field visualization. Finally, the global information entropy value is normalized and limited to the [0,1] interval to eliminate the interference of numerical scale differences on subsequent sampling probability calculations, forming the final initial information entropy field.

[0081] Please refer to Figure 2 The image shown is a schematic diagram of the three-dimensional visualization entropy field obtained through the above steps. Figure 3 This is a two-dimensional visualization of the entropy field, which intuitively demonstrates the exploration value of different parts within the target area.

[0082] S3. An intelligent sampling strategy is adopted, which integrates target bias sampling, historical sampling, and information entropy field-guided sampling. The RRT algorithm is used to iteratively search the path of the unmanned vessel, generate new nodes and perform collision detection. After the new node is added to the RRT tree, the information entropy value of the surrounding area of ​​the new node is decayed and updated. When the new node enters the preset range of the destination coordinates, the initial unmanned vessel path planning is completed.

[0083] Please refer to Figure 4 As shown, the sampling strategy is a key factor affecting the performance of the RRT algorithm. This invention designs an intelligent sampling framework that integrates multiple strategies, achieving the collaborative operation of three sampling strategies through dynamic probability allocation. This embodiment assigns different sampling probabilities to different sampling methods, and the specific methods for generating sampling points are as follows:

[0084] Target bias sampling generates sampling points directly in the area surrounding the endpoint, shortening the search distance at the end of the unmanned vessel's path to accelerate path closure.

[0085] Historically, a baseline node was randomly selected from the optimized unmanned surface vessel path using sampling, and a two-dimensional Gaussian perturbation (the perturbation follows a mean of 0 and a covariance matrix of σ) was applied around it. explore2 The distribution of ×I is isotropic, where I is a 2×2 identity matrix, and σ explore2 =8km is the standard deviation of the local exploration radius. New sampling points are generated. This method not only reuses historical successful experience, but also ensures the diversity of exploration through local perturbation, and avoids getting trapped in local optima.

[0086] Information entropy field-guided sampling first transforms the information entropy field into the following probability distribution formula:

[0087]

[0088] The sampling points are generated using the inverse cumulative distribution function transformation method. The purpose of the squaring operation is to further amplify the probability difference between high-entropy and low-entropy regions, making the sampling more focused on high-value exploration areas. At the same time, appropriate random perturbations are added during the sampling process, and search diversity is maintained through periodic global decay, so as to avoid the algorithm from focusing too much on certain regions and ignoring other possible paths.

[0089] The sampling probability formula for target-biased sampling is:

[0090]

[0091] In the formula: The sampling probability for target-biased sampling. The maximum number of iterations, and It is a preset constant, and In this embodiment, , This setup maintains a moderate goal orientation in the early stages of iteration, ensuring the algorithm fully explores the environment; while in the later stages of iteration, it strengthens the goal bias, accelerating path convergence.

[0092] The sampling probability formula for historical sampling is:

[0093]

[0094] In the formula: The sampling probability of historical sampling. For the number of iterations, and It is a preset constant, and In this embodiment, =0.02, As the number of trials increases, the probability of utilizing historical data gradually increases, reflecting the increasing reliance of the algorithm on historical experience; at the same time, the upper limit constraint prevents the algorithm from over-relying on historical experience and losing its ability to explore new environments.

[0095] The sampling probability formula for information entropy field-guided sampling is:

[0096]

[0097] In the formula: This refers to the sampling probability that guides the sampling of the information entropy field. It undertakes the main exploration task in the early stages of the search, but its proportion gradually decreases as the experiment progresses and historical experience accumulates, achieving a dynamic balance between exploration and utilization.

[0098] After sampling, RRT tree expansion and collision detection are performed. The node closest to the sampling point is quickly found in the RRT tree and used as the reference node for new node expansion. An adaptive step size is calculated dynamically based on the distance between the sampling point and the reference node. When the distance is large, a larger step size is used to accelerate the search; when the distance is small, a smaller step size is used to improve path accuracy, balancing search efficiency and path quality. After new nodes are generated, they are generated along the line connecting the sampling point and the reference node, using an adaptive step size. To ensure the safety of new nodes, a hierarchical collision detection strategy of "fast filtering + precise verification" is adopted. New nodes that pass collision detection without collisions are added to the RRT tree.

[0099] After a new node is successfully added to the RRT tree, it triggers a local information entropy field decay update, which reduces the information entropy value of the area around the node, reduces the focus of subsequent sampling on this area, guides the algorithm to expand to unexplored areas, and improves the comprehensiveness of the search.

[0100] If a new node enters the preset threshold area of ​​the destination (e.g., the distance is less than 5km), it is determined that the target has been successfully reached and the path optimization stage begins; otherwise, the above sampling, expansion and collision detection process is repeated until the maximum number of iterations is reached.

[0101] S4. The initial unmanned vessel path is pruned and optimized, and then adaptive smoothing is performed based on information entropy guidance. The smoothed unmanned vessel path is fitted with a B-spline curve to obtain the optimized unmanned vessel path.

[0102] Please refer to Figure 5 As shown, the original paths generated by the traditional RRT algorithm have problems such as many redundant nodes, excessive turning angles, and poor smoothness, and cannot be directly applied to the actual navigation of unmanned vessels. Therefore, this invention designs a four-stage path optimization strategy of "layer-by-layer progression" to comprehensively improve the feasibility and optimality of the path.

[0103] First, the initial unmanned vessel path is topology-preserving and pruned. Starting from the path's starting point, the path is directly connected to the next node after the first node. If the distance from the first node to the starting point and the next node is less than a set threshold, the topology is considered to remain unchanged, and the first node is deleted. This process is repeated to complete the initial simplification of the unmanned vessel path length while maintaining the basic rationality of the path.

[0104] The unmanned vessel path is preserved after topology pruning. Nodes are filtered based on their turning angles. While retaining the start and end points, nodes with the largest turning angles (such as the first 80%) are selected according to a set ratio, and the remaining nodes are removed. This further simplifies the unmanned vessel path structure while retaining key navigation information.

[0105] Different environments have varying requirements for path smoothness: complex regions (high-entropy regions) need to retain more path details to adapt to environmental constraints, while open regions (low-entropy regions) can benefit from stronger smoothing to improve navigation smoothness. Therefore, this invention dynamically adjusts the adaptive smoothing intensity based on the local information entropy value at each node, using the following formula:

[0106]

[0107] In the formula: For smooth strength, , To set a constant, For nodes The information entropy value.

[0108] The unmanned vessel path is smoothed according to the adaptive smoothing intensity; the smoothing intensity in high-entropy regions is significantly reduced to preserve key details; the smoothing intensity in low-entropy regions is enhanced to improve path smoothness and achieve a balance between smoothness and environmental adaptability.

[0109] B-spline curve fitting is performed on the smoothed unmanned surface vessel (USV) path. The fitting parameters are adjusted according to the distribution of local information entropy values ​​of the nodes: the first fitting order is used in the region where the information entropy value is higher than the set threshold to preserve path details, and the second fitting order is used in the region where the information entropy value is lower than the set threshold to enhance the smoothing effect. The first fitting order is less than the second fitting order. After fitting, a continuous, smooth, and optimized USV path that conforms to the motion characteristics of the USV is obtained, completely eliminating the problems of sawtooth trajectory and sharp turns.

[0110] The optimized unmanned vessel path obtained by this invention is stored through a hybrid memory module, which includes a topology memory module and a path cache pool.

[0111] The topology memory module identifies path segments with curvature exceeding a threshold in the unmanned surface vessel's path through curvature analysis and stores them as critical path segments. These critical path segments are then ranked by value, and any critical path segment exceeding the storage limit is removed from the topology memory module. A comprehensive value assessment formula is used to score the critical path segments, as follows:

[0112]

[0113] In the formula: For rating, , , For parameter weights, The average information entropy value of the path segment. The path segment length, The mean curvature.

[0114] The path cache pool stores each path segment in the optimized unmanned vessel path and sets an initial confidence probability for each path segment. In this embodiment, the initial confidence probability is 0.5. In subsequent iterations, the confidence probability of a path segment that is successfully reused is increased by 0.1 using a Bayesian update mechanism. The confidence probability of a path segment that is not reused for multiple consecutive times (e.g., 3 times as a threshold) is decreased by 0.05. The path segments in the path cache pool are sorted in descending order of confidence probability. Once the storage limit is exceeded, the last sorted path segment is removed from the path cache pool.

[0115] The core advantage of the hybrid memory system lies in enabling the algorithm to gradually transition from "entropy field-guided exploration" to "experience-driven utilization," achieving an adaptive balance between exploration and utilization, and significantly improving the algorithm's performance in repetitive planning scenarios.

[0116] S5. Positively enhance the information entropy value of the area surrounding the optimized unmanned vessel path, while simultaneously decaying and updating the information entropy value within the target area.

[0117] After the path optimization is completed, the hybrid memory system module and the global information entropy field are updated to achieve continuous optimization of knowledge accumulation and search guidance.

[0118] Hybrid memory module update: Key path segments for optimizing unmanned vessel paths are incorporated into the topology memory module. Key inflection points and core channels are identified through curvature analysis, enriching the environmental topology feature library. At the same time, the path cache pool is updated, and newly obtained path segments are sorted and stored. A Bayesian update mechanism is used to adjust the confidence probability of path nodes, ensuring that high-value nodes are prioritized for subsequent sampling and reuse.

[0119] Global information entropy field update: Positively enhance the information entropy value of the area surrounding the optimized unmanned vessel path. The update formula is:

[0120]

[0121] In the formula: This is the information entropy value after positive reinforcement. The information entropy value before the update. To enhance the learning rate update for the entropy field, For the range of influence parameters, This serves as the upper limit of entropy; it guides subsequent searches to focus more on these high-value regions, accelerating path convergence.

[0122] The information entropy value within the target area is decayed and updated using the following formula:

[0123]

[0124] In the formula: To decay the updated information entropy value, The global decay rate of the entropy field. This is to achieve the minimum entropy value. It avoids excessive concentration of the entropy field in certain regions, maintaining the algorithm's ability to explore new paths.

[0125] The combination of the aforementioned enhancement and global decay achieves a dynamic balance between experience accumulation and exploration, enabling the information entropy field system to possess intelligent characteristics of continuous optimization.

[0126] S6. Repeat steps S3 to S5 until the obtained optimized unmanned vessel path meets the index threshold or reaches the maximum number of iterations, and output the historical best unmanned vessel path.

[0127] To ensure the optimality and stability of the output path, this invention designs a multi-dimensional convergence judgment mechanism, comprehensively considering three core indicators: path length stability, performance improvement, and computation time volatility.

[0128] Path length stability: Calculate the standard deviation of the path length in the most recent 20 tests. If the standard deviation is ≤ ±5km, it indicates that the path quality has stabilized and there is limited room for further optimization.

[0129] Performance improvement magnitude: Calculate the average path length of the most recent 10 trials and the average length of the previous 10 trials. If the difference is <0.5%, it indicates that the algorithm performance improvement has reached a bottleneck period, and it is difficult to significantly optimize the path quality by continuing to experiment.

[0130] Calculate time volatility: Calculate the standard deviation of the planning time for the most recent 20 trials. If the standard deviation is ≤ ±0.1s, it indicates that the algorithm's running efficiency is stable and there is no significant fluctuation.

[0131] When all three criteria are met simultaneously in 5 consecutive trials, the algorithm is considered to have converged, the trial cycle is terminated, and the historical best unmanned vessel path is output. If the convergence criteria are not met, the next trial cycle is executed until the preset maximum number of trials is reached.

[0132] This embodiment uses a real sea area map near a port to verify the performance of the method of the present invention (hereinafter referred to as the E-RRT method) in a real application scenario.

[0133] The sea area extends from 99.85°E to 100.85°E and from -3.38°S to -2.51°S, forming a rectangular area of ​​approximately 111km × 96km.

[0134] The unmanned vessel's path starts near the port exit (99.90°E, -2.6°S) and ends at the entrance to the outer sea channel (100.7°E, -3.3°S), with a straight-line distance of approximately 118 km between the start and end points.

[0135] Initialization phase: Input real sea area map parameters and obstacle information to construct the initial information entropy field and hybrid memory module.

[0136] In the multi-trial iterative search phase, intelligent sampling, RRT tree expansion, collision detection, and path optimization are performed. After each trial, the hybrid memory system and global entropy field are updated. When the convergence condition is met, the optimal path is output, and performance metrics such as path length, planning time, and success rate are recorded. The algorithm converges after 170 trials. Please refer to [reference needed]. Figure 6 As shown, the path length was optimized from the original path of about 140km to about 124km, an improvement of 11.4%; the planning time was reduced from the initial peak of 3s to 0.75s, a reduction of 75% in calculation time; and the standard deviation of the path length after convergence was controlled within ±3km, demonstrating strong stability.

[0137] To further demonstrate the improvement of the E-RRT method compared to existing technologies, we applied the traditional RRT and RRT* methods for unmanned surface vessel path planning in the same implementation scenario. Please refer to [link / reference]. Figure 7 and Figure 8 As shown, compared with the traditional RRT and RRT* methods, in the same implementation scenario, the E-RRT method provided by this invention significantly shortens the path length and greatly increases the exploration efficiency.

[0138] Table 1 RRT 178km 1.10s 95% RRT* 163km 2.04s 100% E-RRT 124km 0.75s 100%

[0139] Please refer to Table 1 above. Specifically, compared with the traditional RRT, the E-RRT method provided by this invention reduces the path length by 30% (from 178km to 124km), reduces the computation time by 31% (from 1.10s to 0.75s), and increases the success rate from 95% to 100%. Compared with RRT*, the path length is reduced by 24% (from 163km to 124km), the computation efficiency is increased by 63% (from 2.04s to 0.75s), and the success rate remains at 100%.

[0140] Experimental results show that the algorithm of this invention still exhibits significant comprehensive performance advantages in complex real-world application scenarios such as real sea areas, achieving dual optimization of path quality and computational efficiency, and meeting the actual needs of autonomous navigation for unmanned vessels.

Claims

1. A path planning method for unmanned surface vessels based on information entropy field guidance, characterized in that, Includes the following steps: S1. Acquire the image of the target area and perform binarization processing, mark the obstacles in the target area, and set the starting and ending coordinates of the unmanned vessel path in the target area. S2. Based on the Euclidean distance between each location in the target area and the nearest obstacle, as well as the starting point coordinates and the ending point coordinates, design an information entropy function. Use this information entropy function to assign an initial information entropy value to each location within the target area, thus forming an initial information entropy field. The calculation formula for the information entropy function is as follows: In the formula: The information entropy value function, Let be the distance entropy function. Influence function as the starting point As the endpoint influence function, The coordinates of the position to be calculated, As the starting coordinates, The coordinates of the endpoint , These are the weighting coefficients; The distance entropy function first calculates the Euclidean distance from each location in the target area to the nearest obstacle using a distance scanning algorithm, and then calculates it using a piecewise function based on the range of Euclidean distances. The calculation formula is as follows: In the formula: The piecewise distance entropy function, For Euclidean distance, The first distance threshold, The second distance threshold, and , The basic entropy coefficient, The maximum entropy gain in the mid-range region. This represents the peak entropy value for a long-distance region. The rate at which the entropy decays over long distances; S3. An intelligent sampling strategy is adopted, which integrates target bias sampling, historical utilization sampling and information entropy field guided sampling. The RRT algorithm is used to iteratively search the path of the unmanned vessel, generate new nodes and perform collision detection. After the new node is added to the RRT tree, the information entropy value of the surrounding area of ​​the new node is decayed and updated. When the new node enters the preset range of the destination coordinates, the initial unmanned vessel path planning is completed. S4. The initial unmanned vessel path is pruned and optimized, and then adaptive smoothing is performed based on information entropy guidance. The smoothed unmanned vessel path is fitted with a B-spline curve to obtain the optimized unmanned vessel path. S5. Positively enhance the information entropy value of the area surrounding the optimized unmanned vessel path, while simultaneously decaying and updating the information entropy value within the target area. S6. Repeat steps S3 to S5 until the obtained optimized unmanned vessel path meets the index threshold or reaches the maximum number of iterations, and output the historical best unmanned vessel path.

2. The unmanned vessel path planning method based on information entropy field guidance according to claim 1, characterized in that, The optimized unmanned surface vessel (USV) path is stored through a hybrid memory module, which includes a topology memory module and a path cache pool. The topology memory module identifies path segments with curvature greater than a threshold in the USV path through curvature analysis and stores them as critical path segments. The critical path segments are ranked by value, and the last critical path segment in the ranking is removed from the topology memory module when the storage limit is exceeded. The path cache pool stores each path segment in the optimized USV path and sets an initial confidence probability for each path segment. In subsequent iterations, the confidence probability of a path segment that is successfully reused increases, while the confidence probability of a path segment that is not reused multiple times decreases. The path segments in the path cache pool are ranked from highest to lowest confidence probability, and the last path segment in the ranking is removed from the path cache pool when the storage limit is exceeded.

3. The unmanned vessel path planning method based on information entropy field guidance according to claim 2, characterized in that, The value ranking of critical path segments includes: scoring the critical path segments using a comprehensive value assessment formula, as follows: In the formula: For rating, , , For parameter weights, The average information entropy value of the path segment. The path segment length, The mean curvature.

4. The unmanned vessel path planning method based on information entropy field guidance according to claim 2, characterized in that, Step S5 further includes: storing the key path segments in the optimized unmanned vessel path into the topology memory module, storing each path segment in the optimized unmanned vessel path into the path cache pool, and reordering them respectively to complete the dynamic update of the hybrid memory module.

5. The unmanned surface vessel path planning method based on information entropy field guidance according to claim 1, characterized in that, The intelligent sampling strategy includes: assigning different sampling probabilities to different sampling methods; the target bias sampling directly generates sampling points in the area surrounding the endpoint to shorten the search distance at the end of the unmanned vessel path and accelerate path closure; the historical utilization sampling randomly selects a reference node from the optimized unmanned vessel path and applies a two-dimensional Gaussian perturbation around it to generate new sampling points; and the information entropy field guided sampling uses the inverse cumulative distribution function method to generate sampling points.

6. The unmanned surface vessel path planning method based on information entropy field guidance according to claim 5, characterized in that, The sampling probability formula for the target bias sampling is: In the formula: The sampling probability for target-biased sampling. The maximum number of iterations, and It is a preset constant, and ; The sampling probability formula for the historical sampling is: In the formula: The sampling probability of historical sampling. For the number of iterations, and It is a preset constant, and ; The sampling probability formula for the information entropy field-guided sampling is: In the formula: The sampling probability is used to guide the sampling of the information entropy field.

7. The unmanned surface vessel path planning method based on information entropy field guidance according to claim 1, characterized in that, Step S4 includes the following sub-steps: S41. Perform topology preservation pruning on the initial unmanned vessel path. Starting from the path start point, directly connect the next node after the original first node. If the distance from the first node to the line connecting the start point and the next node is less than a set threshold, the topology is considered to remain unchanged, and the first node is deleted. This process is repeated to complete the first simplification of the unmanned vessel path length. S42. Maintain the pruned unmanned vessel path for the topology, filter nodes based on their turning angles, and while retaining the start and end points, filter out the nodes with the largest turning angles according to a set ratio, and remove the remaining nodes to further simplify the unmanned vessel path structure. S43. Dynamically adjust the adaptive smoothing intensity based on the local information entropy value at the node, using the following formula: In the formula: For smooth strength, , To set a constant, For nodes Information entropy value; The path of the unmanned vessel is smoothed according to the adaptive smoothing intensity; S44. Perform B-spline curve fitting on the smoothed unmanned vessel path, and adjust the fitting parameters according to the distribution of local information entropy values ​​of the nodes: use the first fitting order in the region where the information entropy value is higher than the set threshold to preserve path details, and use the second fitting order in the region where the information entropy value is lower than the set threshold to enhance the smoothing effect. The first fitting order is less than the second fitting order. The optimized unmanned vessel path is obtained after fitting.

8. The unmanned surface vessel path planning method based on information entropy field guidance according to claim 1, characterized in that, Step S5 includes: The information entropy value of the area surrounding the optimized unmanned surface vessel path is positively reinforced, and the update formula is as follows: In the formula: This is the information entropy value after positive reinforcement. The information entropy value before the update. To enhance the learning rate update for the entropy field, For the range of influence parameters, The upper limit of entropy; The information entropy value within the target area is decayed and updated using the following formula: In the formula: To decay the updated information entropy value, The global decay rate of the entropy field. This is the minimum entropy value.

Images