Self-adaptive intelligent obstacle avoidance method for unmanned surface vehicle

Abstract

The invention discloses a self-adaptive intelligent obstacle avoidance method for a water surface unmanned vehicle, which comprises the following steps: establishing a kinematic model of the water surface unmanned vehicle, delimiting an environment boundary, and presetting a plurality of obstacle avoidance algorithms in a system; sensing an obstacle set in a range through a distance-based intelligent local sensing mechanism, modeling a static obstacle, and performing trajectory prediction on a dynamic obstacle; grading the obstacle environment; calculating an overall threat level; based on an obstacle environment grade division result and an overall threat grade, scoring the calculation applicability of each algorithm, and selecting an obstacle avoidance algorithm according to a scoring result and a historical success rate of the algorithm; and generating or updating an obstacle avoidance path according to the selected algorithm. According to the method, the autonomous obstacle avoidance capability with high safety, good real-time performance and relatively low calculation overhead can be provided for the USV under typical complex sea area conditions such as static-dynamic mixing, dense narrow channels and rapid environment change.

Description

Technical Field

[0001] This invention relates to the field of intelligent navigation and control technology for unmanned surface vehicles, and in particular to an adaptive intelligent obstacle avoidance method for unmanned surface vehicles. Background Technology

[0002] In recent years, unmanned surface vehicles (USVs) have been widely used in port security, marine surveying, emergency search and rescue, and surface logistics. USVs need to navigate autonomously in waters with currents, waves, and a mix of static and dynamic obstacles. The core challenge is maintaining path efficiency and attitude stability while ensuring a safe course. To address this need, researchers both domestically and internationally have proposed various obstacle avoidance and path planning schemes based on single algorithms. However, these schemes still suffer from common drawbacks in complex environments, such as limited applicability, slow algorithm switching, and insufficient global-local coordination.

[0003] Most existing USV obstacle avoidance technologies focus on local improvements of a single algorithm: RRT*-smart algorithms have strong global exploration capabilities, but suffer from high replanning overhead in high-frequency dynamic environments; improved DWA emphasizes real-time performance, but is limited by local window size and cost function weights, making it prone to losing global optimality; velocity obstacle models are suitable for multi-vehicle combat, but struggle to handle complex geometric channels and high-speed dynamic obstacles; improved APF enhances the potential field numerical characteristics, but still suffers from parameter sensitivity and trap point issues.

[0004] The root cause of these shortcomings lies in the lack of a unified decision-making framework capable of evaluating and dynamically switching algorithms in real time based on quantifiable metrics of environmental complexity. A single algorithm struggles to simultaneously address global reachability, local real-time performance, and multi-target safety; the key to multi-algorithm hybrid solutions lies in when and how to switch and coordinate them. Current technologies lack a complete system that integrates multiple algorithms such as APF, DWA, RRT, and speed obstacle avoidance through an adaptive scoring mechanism, combined with rolling prediction and local sensor perception. This results in USVs still facing challenges such as path failure, obstacle avoidance oscillations, or excessive computational load in highly dynamic, obstacle-dense, or narrow-channel environments.

[0005] Therefore, there is an urgent need for an adaptive multi-algorithm USV intelligent obstacle avoidance method for complex waters, which can perceive environmental characteristics in real time, predict dynamic threats, and intelligently schedule multiple obstacle avoidance algorithms based on quantitative scoring, so as to make up for the shortcomings of existing technologies in multi-scenario adaptability, algorithm complementarity and global-local integrated decision-making. Summary of the Invention

[0006] To overcome the aforementioned problems in the prior art, this invention proposes an adaptive intelligent obstacle avoidance method for unmanned surface vehicles.

[0007] The technical solution adopted by this invention to solve its technical problem is: an adaptive intelligent obstacle avoidance method for unmanned surface vehicles, comprising the following steps: Step 1: Establish a kinematic model of the unmanned surface vehicle, delineate the environmental boundaries, and pre-set multiple obstacle avoidance algorithms in the system; Step 2: Using a distance-based intelligent local perception mechanism, the set of obstacles within the perception range is used to model static obstacles and predict the trajectory of dynamic obstacles. Step 3: Calculate the obstacle density and the minimum gap between all obstacle pairs within the perception range. Based on the calculation results, classify the obstacle environment into different levels. Step 4: Calculate the potential collision risk between each dynamic obstacle and the surface unmanned vehicle, and calculate the overall threat level by combining the risks of all dynamic obstacles. Step 5: Based on the obstacle environment level classification results in Step 3 and the overall threat level in Step 4, calculate the applicability score for each algorithm, and select the obstacle avoidance algorithm based on the score results and the algorithm's historical success rate. Step 6: Generate or update the obstacle avoidance path based on the algorithm selected in Step 5, verify the safety and feasibility of the path, and record the execution performance until the unmanned surface vehicle reaches the target location.

[0008] The aforementioned adaptive intelligent obstacle avoidance method for unmanned surface vehicles includes multiple obstacle avoidance algorithms in step 1, such as the improved artificial potential field method, the optimized DWA algorithm, and the spatiotemporal RRT algorithm.

[0009] The aforementioned adaptive intelligent obstacle avoidance method for unmanned surface vessels, specifically the improved artificial potential field method, involves dynamically adjusting the attractive force intensity based on the obstacle density of the local environment. The formula for calculating the target attractive force is as follows: in, For attraction gain coefficient, Let be the vector pointing from the current position of the surface unmanned vehicle to the target. Density adjustment factor, Indicates the number of obstacles; The repulsive force is set as a piecewise function, using different force intensity models for different distance ranges: When adjusting distance hour: when hour: when hour: ; in, The distance from the center of the unmanned surface vehicle to the center of the obstacle is the Euclidean distance. The radius of the obstacle; Let be the unit vector pointing from the obstacle to the unmanned surface vehicle. The obstacle size influence factor ensures that larger obstacles generate stronger repulsive forces; its calculation formula is as follows: It is a distance attenuation factor used for adjusting repulsive force in the mid-range; When the system detects that it has fallen into a local minimum, it activates an intelligent escape mechanism, which searches for the optimal escape path through multi-directional exploration: in, For direction The width of the channel above, Scoring is given based on consistency with the target direction.

[0010] The aforementioned adaptive intelligent obstacle avoidance method for surface unmanned vehicles (UAVs) specifically includes an optimized DWA algorithm: a dynamic window defines the selectable speed and angular velocity range of the UAV within the current control cycle, and the improved dynamic window calculation formula is as follows: in, This is the current speed of the unmanned surface vehicle. The current angular velocity of the unmanned surface vehicle; For maximum linear acceleration, For maximum angular acceleration, To predict the time step; The maximum linear velocity limit for unmanned surface vehicles; Maximum angular velocity limit; The improved integrated cost function contains four main components: in, For speed combination Total cost; These are the weighting coefficients for each component, used to balance the importance of different objectives; heading cost. The consistency between the trajectory and the target direction is evaluated using the following formula: in, To predict the direction angle of the trajectory, The target direction angle; Obstacle Distance Cost An improved exponential model is adopted: in, To predict the shortest distance from any point on the trajectory to the nearest obstacle; Speed ​​cost The calculation formula is: Smoothing Cost The calculation formula is: .

[0011] The aforementioned adaptive intelligent obstacle avoidance method for surface unmanned vehicles, specifically the spatiotemporal RRT algorithm, involves introducing the time dimension into the node definition: in, Let t be the spatial coordinates and t be the timestamp. Pointer to the parent node The cumulative cost from the starting point to this node. To score the risk, For exploration rewards; Virtual obstacles are introduced, and a virtual obstacle is generated for each dynamic obstacle at every time point in the prediction time domain: in, The predicted position of the dynamic obstacle at time t is calculated using the obstacle's motion model. To account for the effective radius of prediction uncertainty; The timestamp corresponding to the virtual obstacle; References to the original dynamic obstacles; Define a spatiotemporal collision detection function that takes into account temporal matching degree: in, t represents the spatial location to be detected; t represents the time to be detected. For virtual obstacles; For time tolerance, slight time deviations are allowed; Spatial distance; The effective radius of the virtual obstacle.

[0012] In the aforementioned adaptive intelligent obstacle avoidance method for unmanned surface vessels, the set of obstacles within the perception range in step 2 is represented as follows: in, This represents the set of local obstacles at time t; This represents the i-th obstacle in the environment; Let be the position coordinates of the USV at time t. Let be the Euclidean distance function. Let i be the radius of the obstacle. For the sensing radius.

[0013] The aforementioned adaptive intelligent obstacle avoidance method for unmanned surface vessels, specifically the static obstacle modeling in step 2, involves: employing a dual-layer geometric structure modeling; the outer layer uses circular boundaries for rapid collision detection, while the inner layer uses random polygons to simulate the shape of the real obstacle. The mathematical representation of the static obstacle is as follows: in, The coordinates of the geometric center of the obstacle are represented, and R is the radius of the outer boundary circle. It is the set of vertex coordinates of the internal polygon.

[0014] The adaptive intelligent obstacle avoidance method for surface unmanned vehicles described above, specifically the trajectory prediction of dynamic obstacles in step 2, involves: predicting the motion trajectory of dynamic obstacles within a future time window using a rolling temporal prediction algorithm. Its future prediction time domain The trajectory sequence within is defined as: in, This represents the predicted trajectory sequence of obstacle j starting from the current time t; This represents the predicted position of obstacle j at a future time. To predict the time step, predict the time domain. Dynamically adjust based on obstacle movement speed and environmental complexity; Considering the uncertainty of the prediction model, the concept of time-varying uncertainty radius is introduced. The formula for calculating the uncertainty radius is as follows: in, Let the radius be the obstacle. The speed at which the obstacle moves.

[0015] The aforementioned adaptive intelligent obstacle avoidance method for unmanned surface vessels, specifically step 3 includes: The formula for calculating obstacle density within the sensing range is: In the above formula, The number of static obstacles within the sensing range; The number of dynamic obstacles within the sensing range; Formula for calculating the minimum gap between all pairs of obstacles: in, Let be the distance between the centers of obstacles i and j.

[0016] In the aforementioned adaptive intelligent obstacle avoidance method for unmanned surface vessels, step 4 specifically comprises: Calculate the nearest approach point; the position of the unmanned surface vehicle is... The speed is The location of the obstacle is The speed is The relative position was obtained through calculation. and speed for: The closest approach time of a dynamic obstacle is defined as: The closest approach distance to a dynamic obstacle: Collision risk rating: in, For a safe distance, This is the critical time window; Calculate the overall threat level by considering the risks posed by all dynamic obstacles: in, The sum of the risks of all dynamic obstacles. For maximum relative velocity, This represents the number of path convergence points detected.

[0017] The beneficial effects of this invention are: (1) It realizes an environmental complexity perception and dynamic adaptation mechanism: Existing USV obstacle avoidance technologies generally lack a quantitative assessment mechanism for environmental complexity, making it difficult to flexibly adjust obstacle avoidance strategies according to actual scenarios. This invention constructs an environmental complexity assessment model, comprehensively considering feature parameters of multiple dimensions such as obstacle density, channel width, and dynamic threat level, and dynamically adjusts and adaptively switches obstacle avoidance strategies based on this model, thereby improving the system's adaptability in static-dynamic mixed, dense narrow passages, and dynamic complex scenarios.

[0018] (2) A unified multi-algorithm fusion obstacle avoidance system was constructed: Traditional methods often use a single algorithm to handle obstacle avoidance problems, which makes it difficult to take into account both the global and local real-time performance of path planning. This invention integrates APF, DWA, and RRT obstacle avoidance algorithms and designs an algorithm scheduling module based on a scoring mechanism, so that various algorithms can be called on demand and dynamically coordinated under a unified architecture, thereby improving the obstacle avoidance effect and computational efficiency of the system under various sea conditions.

[0019] (3) Introducing rolling prediction and virtual obstacle modeling mechanism: In view of the problem that traditional obstacle avoidance system has weak prediction ability when facing high-speed dynamic obstacles, this invention introduces a rolling prediction mechanism and a virtual obstacle modeling method. By continuously predicting and modeling the future trajectory of dynamic obstacles, a safety boundary is constructed and potential risks are avoided in advance, which greatly improves the collision avoidance safety in high-speed intersection or violent disturbance environments.

[0020] (4) The intelligent algorithm selection mechanism improves decision-making efficiency and success rate: The intelligent scoring mechanism designed in this invention not only considers the real-time environmental perception results, but also introduces the historical success rate as a decision factor, taking into account both theoretical adaptability and actual performance.

[0021] This invention belongs to the category of intelligent obstacle avoidance methods in navigation and control systems for unmanned surface vessels (USVs). It is directly applied to the path planning and decision-making execution module of the USV autonomous navigation control system. It does not fall under the category of generalized autonomous driving technology or general path search algorithms, nor does it fall under the control design of sensor devices or hardware actuators. Instead, it is a comprehensive decision-making and control strategy method for USVs to dynamically avoid obstacles in complex aquatic environments. It is a path dynamic adjustment and obstacle avoidance decision-making technology adopted to deal with scenarios where static and dynamic obstacles are mixed. It is applicable to the technical implementation of USV intelligent control systems in complex application scenarios such as marine engineering, port operations, environmental monitoring, and water transportation. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the process of this invention; Figure 2 This is a schematic diagram of an obstacle navigation scenario according to the present invention; Figure 3 This is a simulation diagram of the intelligent obstacle avoidance system of the present invention; Figure 4 This is the curve showing the change in heading angle and angular velocity of the unmanned surface vehicle during obstacle avoidance. Figure 5 This is the curve showing the change in distance between the unmanned surface vehicle of this invention and the nearest obstacle. Detailed Implementation

[0023] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] like Figure 1 As shown, this embodiment discloses an adaptive intelligent obstacle avoidance method for unmanned surface vessels (USVs). The system sequentially completes five stages within each control cycle: First, in the environmental perception update stage, sensor data is acquired, static obstacle information is updated, and dynamic obstacles and their predicted trajectories are tracked, thereby improving the local obstacle list. Then, in the environmental analysis and evaluation stage, the system calculates environmental complexity, assesses dynamic threat levels, resolves spatial constraints, and updates the environmental feature vector. Next, in the algorithm adaptability calculation stage, candidate obstacle avoidance algorithms are evaluated for fitness, and the optimal algorithm is selected based on historical performance and algorithm switching stability. Finally, in the path planning and execution stage, the selected algorithm is run to generate or update obstacle avoidance paths, verifying their safety and feasibility, and recording execution performance. The system can provide safe and efficient autonomous obstacle avoidance capabilities for USVs in various complex marine environments.

[0025] This embodiment takes the autonomous obstacle avoidance of a medium-sized USV (5 meters long and 3 meters wide) in a complex marine environment as an example to explain in detail the specific implementation of each technical module.

[0026] Step 1: System Overall Architecture and Initialization The USV intelligent obstacle avoidance system in this embodiment adopts a modular design architecture, mainly including an environmental perception module, an algorithm library module, an intelligent decision-making module, and an execution control module. During system initialization, a USV kinematic model is first established, defining the USV state as a quintuple USVState(x, y, yaw, v, ...). ), where (x, y) are the position coordinates, yaw is the heading angle, and v is the linear velocity. The maximum speed of the USV is set to 10.0 m / s, the maximum angular velocity to 40° / s, and the local sensing range radius to 120 meters. These parameters are determined based on the typical effective detection range of small and medium-sized USV radars and lidars.

[0027] The environmental boundary is set as a rectangular water area of ​​500×500 meters, with the starting point set at (50, 50) and the target point set at (450, 450), forming a diagonal navigation task. The system uses a fixed time step Δt=0.5 seconds for discretized simulation, with the maximum simulation time set to 200 seconds to ensure that the USV has sufficient time to complete the obstacle avoidance task.

[0028] Step 2: Specific Implementation of Environmental Modeling and Sensing System (e.g.) Figure 2 (As shown) To facilitate obstacle avoidance, all obstacles are expanded into regular circles. Static obstacles are modeled using a double-layered geometric structure. The outer layer uses circular boundaries for fast collision detection, while the inner layer uses random polygons to simulate the realistic obstacle shape. The specific implementation process is as follows: First, the system randomly generates static obstacles within seven predefined areas according to a preset obstacle distribution strategy. The ranges of each area are as follows: Area 1 (100, 200, 100, 200), Area 2 (300, 400, 100, 200), Area 3 (100, 200, 300, 400), Area 4 (300, 400, 300, 400), Central Area (200, 300, 200, 300), Additional Area 1 (50, 150, 250, 350), and Additional Area 2 (350, 450, 250, 350). The obstacle radii are randomly generated using a uniform distribution, ranging from 12 to 22 meters.

[0029] The mathematical representation of a static obstacle is: in, Represents the geometric center coordinates of the obstacle. The radius of the outer boundary circle is used for fast collision detection calculations. It is a set of vertex coordinates of an internal polygon. The algorithm automatically generates various shapes such as triangles, pentagons, hexagons or irregular polygons to meet the actual obstacle shape.

[0030] Dynamic obstacle modeling introduces a time dimension to static obstacle modeling, enabling precise description of obstacle movement trajectories. The system supports multiple motion modes, including linear, circular, and zigzag motion. The mathematical representation of dynamic obstacles is as follows: in, This is a parameterized path function, where t is a time variable. This is a set of path parameters. Different motion modes correspond to different path function forms.

[0031] Trajectory prediction for dynamic obstacles directly impacts the accuracy and safety of obstacle avoidance decisions. The system, through a rolling time-domain prediction algorithm, can predict the trajectory of dynamic obstacles within a future time window. Its future prediction time domain The trajectory sequence within is defined as: in, Indicates obstacles From the current moment The initial predicted trajectory sequence; This represents the predicted position of obstacle j at a future time. The prediction time step is set to 0.2 seconds, predicting the time domain. The value is dynamically adjusted based on the obstacle's movement speed and environmental complexity, ranging from 30 to 50 seconds.

[0032] Considering the uncertainty of the prediction model, the system introduces the concept of a time-varying uncertainty radius. As the prediction time increases, the prediction accuracy gradually decreases, therefore the safety margin needs to be increased accordingly. The formula for calculating the uncertainty radius is: in, Let the radius be the obstacle. Given the obstacle's speed, the first parameter, 5, represents the basic uncertainty radius. Firstly, considering the USV's physical dimensions of approximately 5 meters long and 3 meters wide, sufficient safety distance needs to be reserved. Secondly, environmental factors such as sea waves may cause the USV's position to shift, typically by 1-2 meters. Finally, considering sensor measurement errors and control system response delays, additional safety margins are provided. Therefore, a setting of 5.0 meters ensures safety while avoiding excessive conservatism that could reduce path efficiency. The second parameter... The third item reflects the impact of speed on forecast uncertainty. This demonstrates the impact of time on prediction accuracy.

[0033] The system employs a distance-based intelligent local sensing mechanism, which simulates the sensor sensing capabilities of a real USV. The local sensing region is defined as having a sensing radius centered on the USV's current location. Considering the actual detection range of radar and lidar in marine environments, a circular area of ​​120 meters represents a typical effective detection range for small to medium-sized USV sensors. This sensing radius selection takes into account factors such as the USV's speed, algorithm computation time, and obstacle avoidance safety distance. The set of obstacles within the sensing range is represented as follows: in, Indicates at time A collection of local obstacles; This represents the i-th obstacle in the environment; Let be the position coordinates of the USV at time t. Let be the Euclidean distance function. Let be the radius of obstacle i. This definition ensures that obstacles near the boundary can also be detected correctly.

[0034] Step 2: Design and Implementation of Multi-Algorithm Obstacle Avoidance Strategies 1) Improved artificial potential field method While the traditional artificial potential field method is conceptually simple and computationally efficient, it is prone to getting trapped in local minima in complex environments, especially in environments with dense obstacles or narrow passages. This study makes several in-depth improvements to the APF algorithm.

[0035] First, traditional APF uses a fixed target attraction force, which can easily cause the USV to move too aggressively toward the target in environments with dense obstacles, thus increasing the risk of collision. The improved APF dynamically adjusts the attraction force intensity based on the obstacle density of the local environment. The formula for calculating the target attraction force is: in, For attraction gain coefficient, This is the vector pointing from the current position of the USV to the target. Density adjustment factor. Calculated based on the number of local obstacles: When the number of local obstacles is small Approaching version 1.0, maintaining strong goal orientation; when obstacles are dense, Reduce or diminish the attractiveness of the target, and avoid overly aggressive behavior.

[0036] Secondly, traditional repulsive force functions typically employ a simple inverse proportional relationship, which is insufficient for effective repulsion at close range. The improved repulsive force uses a piecewise function, applying different force intensity models across different distance ranges: When adjusting distance hour: when hour: when hour: .

[0037] in, The Euclidean distance from the center of the USV to the center of the obstacle; The radius of the obstacle; Let $\mathbf{v}$ be the unit vector pointing from the obstacle to the $V. The obstacle size influence factor ensures that larger obstacles generate stronger repulsive forces; its calculation formula is as follows: This is a distance attenuation factor used for adjusting repulsive forces in the mid-range: Furthermore, when the system detects that it has fallen into a local minimum (the magnitude of the resultant force) (And the distance to the nearest obstacle is less than 50 meters), triggering the intelligent escape mechanism. This mechanism searches for the optimal escape path through multi-directional exploration: in, For direction The width of the channel above, To ensure consistency with the target direction, the system evaluates the direction in 12 equally spaced directions and selects the escape direction with the widest channel and closest to the target direction.

[0038] 2) Optimization of dynamic window method Dynamic windowing has a natural advantage in handling dynamic obstacles, but there is still room for improvement in the computational accuracy and security of traditional DWA in complex environments. This study has comprehensively optimized the DWA algorithm.

[0039] First, for refined dynamic window construction, the dynamic window defines the range of selectable velocity and angular velocity for the USV within the current control cycle. Window construction requires comprehensive consideration of the USV's physical motion constraints and safety constraints. The improved dynamic window calculation formula is as follows: The reachability window is based on kinematic constraints, where: in, This is the current speed of the USV; The current angular velocity of the USV; For maximum linear acceleration, For maximum angular acceleration, To predict the time step; This is the maximum linear velocity limit for the USV; This is a limit on the maximum angular velocity.

[0040] The cost function is the core of the DWA algorithm, directly determining the quality of path selection. The improved comprehensive cost function consists of four main components: in, For speed combination Total cost; These are the weighting coefficients for each component, used to balance the importance of different objectives; heading cost. The consistency between the trajectory and the target direction is evaluated using the following formula: in, To predict the direction angle of the trajectory, The target direction angle.

[0041] The obstacle distance cost uses an improved exponential model: in, To predict the shortest distance from any point on the trajectory to the nearest obstacle.

[0042] Speed ​​costs incentivize higher speeds of progress and improve task execution efficiency: The added smoothness reduces the cost of sharp speed and angular velocity changes: Weights set to .

[0043] 3) Improved Fast Random Tree The Restricted Range (RRT) algorithm exhibits excellent global search capabilities in complex environments, but traditional RRT cannot effectively handle dynamic obstacles. This study proposes a Spatiotemporal RRT (ST-RRT) algorithm, which significantly improves RRT performance in dynamic environments. For the design of ST-RRT nodes, since traditional RRT nodes only contain spatial coordinate information and cannot handle time-varying dynamic obstacles, ST-RRT introduces the time dimension into the node definition: in, Let t be the spatial coordinates and t be the timestamp. Pointer to the parent node The cumulative cost from the starting point to this node. To score the risk, To explore rewards.

[0044] Traditional random sampling (RRT) uses uniform random sampling, which has low search efficiency. The improved RRT implements a dynamic combination of five sampling strategies: target-oriented, exploratory, intermediate target, uniform random, and sampling away from crowded areas.

[0045] The weights of each strategy are dynamically adjusted as the search progresses, with the probability of goal-oriented sampling decreasing with each iteration: Exploratory sampling probability increases: in This represents the current iteration number. This represents the maximum number of iterations.

[0046] To handle the predicted trajectories of dynamic obstacles, the system introduces the concept of virtual obstacles. For each dynamic obstacle, a virtual obstacle is generated at each time point within the prediction time domain. in, For dynamic obstacles at any time The predicted position is calculated using the motion model of the obstacle; The effective radius, which takes into account the uncertainty of the forecast, increases with the forecast time. The timestamp corresponding to the virtual obstacle; References to the original dynamic obstacles, used for tracing and analysis.

[0047] For each dynamic obstacle, the system generates a series of virtual obstacles at fixed time intervals within its prediction time domain. If the prediction time domain is 50 seconds, then 50 virtual obstacles will be generated for each dynamic obstacle.

[0048] To check if the USV collides with an obstacle and to predict collisions, a spatiotemporal collision detection function is defined, which takes into account temporal matching degree: in, t represents the spatial location to be detected; t represents the time to be detected. For virtual obstacles; For time tolerance, slight time deviations are allowed; Spatial distance; The effective radius of the virtual obstacle.

[0049] Step 3: Intelligent Algorithm Selection Mechanism The core of intelligent algorithm selection lies in accurately assessing the characteristics of the current environment. The system establishes a multi-dimensional environmental analysis framework that can comprehensively quantify environmental complexity. Obstacle density is a crucial indicator of environmental complexity, directly impacting the difficulty of path planning. Before obstacle avoidance, the system calculates the obstacle density within its perception range: In the above formula, The number of static obstacles within the sensing range; The number of dynamic obstacles within the perception range.

[0050] The channel width reflects the available maneuverability of the USV and is an important factor in algorithm selection. In addition to calculating obstacle density, the system also calculates the minimum clearance between all obstacle pairs: in, For obstacles and The center distance; when When the size is large and the environment is relatively open, simple algorithms such as APF can handle it; when When the size is small, algorithms with precise search capabilities, such as RRT, are required.

[0051] Based on a comprehensive analysis of density and channel width, the obstacle environment is divided into three complexity levels, starting with a simple environment: and Medium environment and Complex environment and The grading criteria are set based on the following considerations: the threshold for simple environments ensures sufficient maneuverability; the threshold for medium environments allows for moderate obstacle density but still maintains basic passage capability; and the threshold for complex environments covers all other situations, including environments with high density or narrow passages.

[0052] This hierarchical classification provides an intuitive basis for determining the environment type for algorithm selection. In addition to the distribution of environmental complexity levels, the threat assessment of dynamic obstacles is also a key factor in algorithm selection. Therefore, the system establishes a threat assessment model based on the principles of physical kinematics.

[0053] For each dynamic obstacle, the system calculates its potential collision risk with the USV. First, it calculates the nearest approach point, assuming the USV's position is... The speed is The location of the obstacle is The speed is The relative position and velocity were calculated as follows: The closest approach time of a dynamic obstacle is defined as: The closest approach distance to a dynamic obstacle: Collision risk rating: in, For a safe distance, This is the critical time window.

[0054] Finally, the system integrates the risks of all dynamic obstacles to calculate the overall threat level: in, The sum of the risks of all dynamic obstacles. For maximum relative velocity, This represents the number of path convergence points detected.

[0055] Based on the results of environmental analysis and threat assessment, the system calculates an applicability score for each algorithm, with a score range of 0-1. The higher the score, the more suitable the algorithm is for the current environment.

[0056] The APF algorithm is suitable for simple, static environments. The scoring model is as follows: The base score of 0.5 ensures that each algorithm has a certain probability of being selected, and the design of each component is as follows: The complexity impact factor reflects the APF's sensitivity to environmental complexity: The static environment preference factor reflects the APF's ability to handle dynamic obstacles: The target distance factor takes into account the advantage of APF when approaching the target: The threat level factor reflects the APF's sensitivity to dynamic threats: The DWA algorithm is suitable for dynamic environments of moderate complexity. The preference for moderate complexity reflects the applicability of DWA: Dynamic environmental adaptability is the core advantage of DWA: The RRT algorithm is suitable for complex, high-threat environments. The RRT algorithm's adaptability score is as follows: The preference for complex environments demonstrates RRT's global search capabilities: High threat adaptability reflects RRT's predictive planning capabilities: Narrow channel preference reflects RRT's precise search capability: The final algorithm selection formula, taking into account historical performance weights, is as follows: in, This design takes into account both the algorithm's historical success rate and its applicability to the current environment.

[0057] Based on the method disclosed in this embodiment, obstacle avoidance simulation is performed on the unmanned surface vehicle. The obstacle avoidance simulation diagram is as follows. Figure 3 As shown, the blue triangle represents the starting point of the unmanned surface vehicle (UAV), the red pentagram represents the midpoint of the UAV, the green dots represent the UAV itself, the green line segments represent the trajectory of the UAV switching to the DWA algorithm for obstacle avoidance, the purple line segments represent the trajectory of the UAV switching to the RRT algorithm for obstacle avoidance, the gray polygons represent static obstacles, and the colored polygons and corresponding dashed lines represent the dynamic obstacles and their predicted trajectories. Figure 3 As can be seen from this, the obstacle avoidance method disclosed in this embodiment can effectively avoid obstacles and safely reach the target point.

[0058] Figure 4 and Figure 5 The figures show the curves of the unmanned surface vessel's heading angle and angular velocity during the simulation process, as well as the curve of the unmanned surface vessel's distance from the nearest obstacle. As can be seen from the figures, the heading angle and angular velocity of the unmanned surface vessel changed relatively smoothly during obstacle avoidance, and the steering control was stable without any violent swaying. Figure 5 Recent changes in the distance to obstacles show that the unmanned surface vessel (USV) and obstacles remain within a safe range most of the time. Although there are occasional approaches to dangerous areas, it can adjust and avoid them in time, which is consistent with actual motion control.

[0059] The above embodiments are merely exemplary embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art can make various modifications or equivalent substitutions to the present invention within its scope and spirit, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of the present invention.

Claims

1. An adaptive intelligent obstacle avoidance method for unmanned surface vehicles, characterized in that, Includes the following steps: Step 1: Establish a kinematic model of the unmanned surface vehicle, delineate the environmental boundaries, and pre-set multiple obstacle avoidance algorithms in the system; Step 2: Using a distance-based intelligent local perception mechanism, the set of obstacles within the perception range is used to model static obstacles and predict the trajectory of dynamic obstacles. Step 3: Calculate the obstacle density and the minimum gap between all obstacle pairs within the perception range. Based on the calculation results, classify the obstacle environment into different levels. Step 4: Calculate the potential collision risk between each dynamic obstacle and the surface unmanned vehicle, and calculate the overall threat level by combining the risks of all dynamic obstacles. Step 5: Based on the obstacle environment level classification results in Step 3 and the overall threat level in Step 4, calculate the applicability score for each algorithm, and select the obstacle avoidance algorithm based on the score results and the algorithm's historical success rate. Step 6: Generate or update the obstacle avoidance path based on the algorithm selected in Step 5, verify the safety and feasibility of the path, and record the execution performance until the unmanned surface vehicle reaches the target location.

2. The adaptive intelligent obstacle avoidance method for unmanned surface vehicles according to claim 1, characterized in that, The various obstacle avoidance algorithms in step 1 include the improved artificial potential field method, the optimized DWA algorithm, and the spatiotemporal RRT algorithm.

3. The adaptive intelligent obstacle avoidance method for unmanned surface vehicles according to claim 2, characterized in that, The improved artificial potential field method specifically involves dynamically adjusting the attractive force intensity based on the obstacle density of the local environment. The formula for calculating the target attractive force is as follows: in, For attraction gain coefficient, Let be the vector pointing from the current position of the surface unmanned vehicle to the target. Density adjustment factor, Indicates the number of obstacles; The repulsive force is set as a piecewise function, and different force intensity models are used for different distance ranges: When adjusting distance hour: when hour: when hour: ; in, The distance from the center of the unmanned surface vehicle to the center of the obstacle is the Euclidean distance. The radius of the obstacle; Let be the unit vector pointing from the obstacle to the unmanned surface vehicle. The obstacle size influence factor ensures that larger obstacles generate stronger repulsive forces; its calculation formula is as follows: It is a distance attenuation factor used for adjusting repulsive force in the mid-range; When the system detects that it has fallen into a local minimum, it activates an intelligent escape mechanism, which searches for the optimal escape path through multi-directional exploration: in, For direction The width of the channel above, Scoring is given based on consistency with the target direction.

4. The adaptive intelligent obstacle avoidance method for unmanned surface vehicles according to claim 2, characterized in that, The optimized DWA algorithm specifically defines the range of selectable speeds and angular velocities for the surface unmanned vehicle within the current control cycle using a dynamic window. The improved dynamic window calculation formula is as follows: in, This is the current speed of the unmanned surface vehicle. The current angular velocity of the unmanned surface vehicle; For maximum linear acceleration, For maximum angular acceleration, To predict the time step; The maximum linear velocity limit for unmanned surface vehicles; Maximum angular velocity limit; The improved integrated cost function contains four main components: in, For speed combination Total cost; These are the weighting coefficients for each component, used to balance the importance of different objectives; heading cost. The consistency between the trajectory and the target direction is evaluated using the following formula: in, To predict the direction angle of the trajectory, The target direction angle; Obstacle Distance Cost An improved exponential model is adopted: in, To predict the shortest distance from any point on the trajectory to the nearest obstacle; Speed ​​cost The calculation formula is: Smoothing Cost The calculation formula is: 。 5. The adaptive intelligent obstacle avoidance method for unmanned surface vehicles according to claim 2, characterized in that, The spatiotemporal RRT algorithm specifically incorporates the time dimension into the node definition: in, Let t be the spatial coordinates and t be the timestamp. Pointer to the parent node The cumulative cost from the starting point to this node. To score the risk, For exploration rewards; Virtual obstacles are introduced, and a virtual obstacle is generated for each dynamic obstacle at every time point in the prediction time domain: in, The predicted position of the dynamic obstacle at time t is calculated using the obstacle's motion model. To account for the effective radius of prediction uncertainty; The timestamp corresponding to the virtual obstacle; References to the original dynamic obstacles; Define a spatiotemporal collision detection function that takes into account temporal matching degree: in, t represents the spatial location to be detected; t represents the time to be detected. For virtual obstacles; For time tolerance, slight time deviations are allowed; Spatial distance; The effective radius of the virtual obstacle.

6. The adaptive intelligent obstacle avoidance method for unmanned surface vehicles according to claim 1, characterized in that, The set of obstacles within the sensing range in step 2 is represented as follows: in, This represents the set of local obstacles at time t; This represents the i-th obstacle in the environment; Let be the position coordinates of the USV at time t. Let be the Euclidean distance function. Let i be the radius of the obstacle. For the sensing radius.

7. The adaptive intelligent obstacle avoidance method for unmanned surface vessels according to claim 1, characterized in that, The static obstacle modeling in step 2 specifically involves: using a double-layered geometric structure, with the outer layer employing circular boundaries for rapid collision detection, and the inner layer using random polygons to simulate the shape of the real obstacle. The mathematical representation of the static obstacle is as follows: in, The coordinates of the geometric center of the obstacle are represented, and R is the radius of the outer boundary circle. It is the set of vertex coordinates of the internal polygon.

8. The adaptive intelligent obstacle avoidance method for unmanned surface vehicles according to claim 1, characterized in that, The trajectory prediction of the dynamic obstacle in step 2 specifically involves: predicting the motion trajectory of the dynamic obstacle within a future time window using a rolling time-domain prediction algorithm. Its future prediction time domain The trajectory sequence within is defined as: in, This represents the predicted trajectory sequence of obstacle j starting from the current time t; This represents the predicted position of obstacle j at a future time. To predict the time step, predict the time domain. Dynamically adjust based on obstacle movement speed and environmental complexity; Considering the uncertainty of the prediction model, the concept of time-varying uncertainty radius is introduced. The formula for calculating the uncertainty radius is: in, Where is the radius of the obstacle. The speed at which the obstacle moves.

9. The adaptive intelligent obstacle avoidance method for unmanned surface vehicles according to claim 1, characterized in that, Step 3 specifically includes: The formula for calculating obstacle density within the sensing range is: In the above formula, The number of static obstacles within the sensing range; The number of dynamic obstacles within the sensing range; Formula for calculating the minimum gap between all pairs of obstacles: in, Let be the distance between the centers of obstacles i and j.

10. The adaptive intelligent obstacle avoidance method for unmanned surface vehicles according to claim 1, characterized in that, Step 4 specifically involves: Calculate the nearest approach point; the position of the unmanned surface vehicle is... The speed is The location of the obstacle is The speed is The relative position was obtained through calculation. and speed for: The closest approach time of a dynamic obstacle is defined as: The closest approach distance to a dynamic obstacle: Collision risk rating: in, For a safe distance, This is the critical time window; Taking into account the risks of all dynamic obstacles, calculate the overall threat level: in, The sum of the risks of all dynamic obstacles. For maximum relative velocity, This represents the number of path convergence points detected.

Images