A special unmanned vehicle autonomous obstacle avoidance method, an unmanned vehicle system and a computer readable storage medium

Abstract

The present application relates to the technical field of unmanned vehicle path planning and autonomous navigation, and particularly relates to a special unmanned vehicle autonomous obstacle avoidance method, which comprises steps of environment modeling, vehicle positioning, path planning, path execution and control, and simultaneously discloses a special unmanned vehicle system for realizing the above-mentioned special unmanned vehicle autonomous obstacle avoidance method, which comprises a vehicle-mounted main controller, a satellite navigation receiver, a driving and executing mechanism; the special unmanned vehicle autonomous obstacle avoidance method effectively solves core problems such as high system cost, poor environmental adaptability, large map data volume, poor planning algorithm performance and substandard path quality through steps of environment modeling, vehicle positioning, path planning, path execution and control, and through integration of sparse environment modeling, satellite precise positioning and hybrid intelligent planning, a complete autonomous obstacle avoidance solution scheme which is completely independent of expensive real-time sensors and has low cost and high reliability is constructed.

Description

Technical Field

[0001] This invention belongs to the field of unmanned vehicle path planning and autonomous navigation technology, specifically relating to a special unmanned vehicle autonomous obstacle avoidance method, unmanned vehicle system and computer-readable storage medium. Background Technology

[0002] In the field of autonomous driving technology, autonomous obstacle avoidance is a core capability for achieving vehicle intelligence and practicality, and its technical approach largely determines the system's cost, complexity, and reliability. Currently, in the field of autonomous navigation and obstacle avoidance technology for autonomous vehicles, the choice of technical path differs from that of conventional autonomous vehicles due to their application scenarios and cost control requirements.

[0003] There are two main approaches to existing technologies. The first approach is obstacle avoidance based on real-time environmental perception. By equipping autonomous vehicles with various sensors such as LiDAR, millimeter-wave radar, and vision cameras, this approach can detect, identify, and track obstacles in the surrounding environment in real time, thereby generating local obstacle avoidance paths. However, in the application of low-cost special-purpose autonomous vehicles, the high cost of high-precision environmental perception sensors (especially LiDAR) conflicts with the cost control requirements of autonomous vehicles. At the same time, optical sensors are easily affected by weather and lighting changes, leading to a decrease in perception capabilities and failing to meet the requirements of all-weather operation, resulting in poor environmental adaptability. In addition, real-time processing of sensor data requires high computing power, which also increases power consumption and cost.

[0004] The second technical approach is a global path planning scheme based on a pre-stored environment map. By constructing a static environment map of the task area beforehand, a collision-free path from the starting point to the destination can be planned for the autonomous vehicle without the need for real-time sensors. Commonly used planning algorithms include A* algorithm, Dijkstra's algorithm, Rapid Random Tree Exploration (RRT) and its variants (such as RRT), as well as biomimetic optimization algorithms such as Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). However, the above algorithms incur huge overhead in terms of map information storage and computation, and their overall performance is often unsatisfactory. Traditional graph search algorithms are inefficient when dealing with complex environments, and the smoothness of the generated paths is not ideal; while algorithms such as RRT can quickly find feasible paths, the initial path quality is low and requires further optimization; metaheuristic algorithms such as GA and PSO are prone to getting trapped in local optima, have slow convergence speeds, and are difficult to generate high-quality paths within a limited time.

[0005] In existing technologies, the autonomous obstacle avoidance methods for special-purpose unmanned vehicles employing the two aforementioned technical approaches suffer from several drawbacks. These include high system costs due to reliance on sensors, poor environmental adaptability, large amounts of pre-stored map data, high hardware resource requirements due to poor planning algorithm performance, and unsatisfactory planned path quality. Therefore, overcoming these technical problems and shortcomings is a key issue that needs to be addressed. Summary of the Invention

[0006] The purpose of this invention is to overcome the defects described in the background art, thereby realizing an autonomous obstacle avoidance method for special unmanned vehicles, and solving the problems of high system cost, poor environmental adaptability, large amount of pre-stored map data, high hardware resource requirements due to poor planning algorithm performance, and unsatisfactory planning path quality in the existing obstacle avoidance methods that rely on sensors.

[0007] To achieve the above-mentioned objectives, the technical solution of this invention is: a method for autonomous obstacle avoidance of a special unmanned vehicle, comprising the following steps:

[0008] S1. Environment Modeling: Construct a sparse binary raster map of the known scene and store the map data using a preset compression storage method.

[0009] S2. Vehicle Positioning: The real-time position coordinates and heading angle of the unmanned vehicle are obtained through satellite navigation equipment. The real-time position coordinates are then mapped onto the sparse binary grid map to obtain the current position of the unmanned vehicle on the map. , and orientation angle .

[0010] S3. Path planning: Based on the current position and the preset target position... A hybrid path planning strategy is adopted to generate a collision-free smooth path. The hybrid path planning strategy includes global coarse path planning, local path fine optimization and path smoothing processing.

[0011] S4. Path Execution and Control: Control the unmanned vehicle to travel along the smooth path, obtain the real-time driving position of the unmanned vehicle through satellite navigation equipment, and perform closed-loop feedback control based on the deviation between the real-time driving position and the smooth path until the unmanned vehicle reaches the preset target position. .

[0012] Specifically, the preset compression storage method in step S1 is the obstacle description method, specifically:

[0013] For each row of a sparse binary raster map, store the row label for that row.

[0014] For each obstacle in the row, store the starting column position of the obstacle and the length of the grid it occupies consecutively.

[0015] If there are no obstructions in the row, only the row flag is stored.

[0016] The satellite navigation device in step S2 is a GPS receiver or a Beidou satellite navigation receiver. The positioning update frequency of the satellite navigation device is ≥1Hz. When mapping the real-time position coordinates to a sparse binary grid map, the latitude and longitude coordinates output by the satellite navigation device are converted into row and column numbers in the sparse binary grid map through coordinate transformation.

[0017] In the above-mentioned special unmanned vehicle autonomous obstacle avoidance method, when constructing a sparse binary grid map in step S1, the known scene is divided into uniform grids, the size of which is 0.5m-2m; grids with obstacles are marked as 1, and grids without obstacles are marked as 0.

[0018] Preferably, the global path coarse planning in step S3 adopts the RRT* algorithm, and the execution process of the RRT* algorithm includes:

[0019] Step 1: Use the current position of the autonomous vehicle as the root node. Initialize a random tree and set the preset target position. Set as the target area.

[0020] Step 2: Randomly sample points in the free space of the sparse binary raster map to generate sampling points. .

[0021] Step 3: Select the sampling point in the random tree. The nearest node is used as the reference node. With a preset probability Select the sampling point Or preset target location directional growth, expanding to generate new nodes according to a set stepsize. The probability The preferred value is 0.5.

[0022] Step 4: Query the sparse binary raster map to determine the baseline node. With new nodes If the path segment between them crosses an obstacle, then the new node is discarded. Then return to step 2; if the obstacle is not crossed, retain the new node. .

[0023] Step 5: Using the sampling points Using the set step size as the radius and center, find the set of potential parent nodes in the random tree that are within this radius, and calculate the set of parent nodes from the root node. From each potential parent node to the new node Given the path length, select the potential parent node with the shortest path length and no collision as the new node. The parent node.

[0024] Step 6: For the new node Rewire the existing nodes of the random tree within the radius range. If the new node is... Being the parent node of an existing node can shorten the distance from the existing node to the root node. If the path length is known and there are no collisions, then update the parent node of the existing node.

[0025] Step 7: Repeat steps 2 to 6 until a leaf node of the random tree enters the target area, and output an initial path consisting of multiple path points.

[0026] Preferably, the local path optimization in step S3 employs an improved crayfish optimization algorithm, the initialization process of which is as follows:

[0027] The path points in the initial path output by the global coarse-scale path planning are used as the initial population, and the size of the initial population is equal to the number of path points in the initial path. Each individual in the population corresponds to a waypoint. A single individual in the population is represented as a 1×2 matrix containing the x and y coordinates of the corresponding waypoint, i.e., a two-dimensional vector. The entire initial population is represented as follows:

[0028] ;

[0029] For the first The x-coordinates of the path points For the first The y-coordinates of the path points Individuals in a single population are represented individually as .

[0030] Preferably, the improved crayfish optimization algorithm further includes a stage switching and position update process, specifically:

[0031] a. Phase Switching: Calculating Virtual Temperature The calculation formula is:

[0032] ;

[0033] in for Random numbers within a range; when When the temperature exceeds 30℃, the algorithm enters a cooling-off phase.

[0034] when When the temperature is ≤30℃, the algorithm enters the foraging phase.

[0035] b. Location update during the summer retreat phase: Calculate cave locations The calculation formula is:

[0036] ;

[0037] in This represents the historical best position for an individual in the current population. This is the global optimal position for the current population.

[0038] when When the value is ≤0.5, the position of an individual in the population is updated according to the formula:

[0039] ;

[0040] in This represents the current iteration number. This is the number of the next iteration. The maximum number of iterations, the decreasing coefficient. , For the first During the nth iteration Individual in the population The coordinates of the dimension.

[0041] when When the value is greater than 0.5, the position of an individual in the population is updated according to the formula:

[0042] ;

[0043] Random individual index , For the first During the nth iteration Individual in the population The coordinates of the dimension.

[0044] c. Location update during foraging phase: Calculate food size The calculation formula is:

[0045] ;

[0046] in The adjustment factor is set to 3. This represents the fitness value of an individual in the current population. The fitness value for the preset target location.

[0047] Define auxiliary variables .

[0048] according to The size of the population is used to update the position of individual individuals according to the formula:

[0049] ;

[0050] The stride control parameter, i.e., the probability coefficient, is based on the optimal temperature of 30℃ and the normal distribution of intake, and is used to dynamically adjust the step size for updating the position of individuals in the population.

[0051] d. Iteration Termination: The process of switching between iteration phases and updating positions continues until the maximum number of iterations is reached. If the fitness value converges, the optimized path points are output.

[0052] Furthermore, the fitness value The calculation formula is:

[0053] ;

[0054] in: The total path length is calculated using the following formula:

[0055] ;

[0056] Let the dimension be the path point sequence. , The first The x and y coordinates of each path point.

[0057] The formula for calculating path smoothness is:

[0058] ;

[0059] For the first The path point and the first The heading angle of the line connecting the path points is calculated using the following formula:

[0060] .

[0061] The obstacle penalty is calculated as follows:

[0062] ;

[0063] This represents the distance between the pathpoint of an individual in the current population and the nearest obstacle. This is the penalty coefficient. These are weighting coefficients, adjusted according to actual path requirements. If a greater emphasis is placed on path length, the weighting factor increases. Focusing on path smoothness increases .

[0064] Preferably, the path smoothing process in step S3 uses cubic spline interpolation, specifically: a cubic polynomial is constructed between the optimized path points output by the local path optimization, so that the generated smooth path remains continuous in terms of position, first derivative, and second derivative.

[0065] This invention also discloses a special-purpose unmanned vehicle system for implementing the above-described autonomous obstacle avoidance method for special-purpose unmanned vehicles, comprising:

[0066] The onboard main controller is pre-stored with the sparse binary grid map compressed using the obstacle description method, and is equipped with a program module that performs environmental modeling, path planning, and closed-loop feedback control in the autonomous obstacle avoidance method for special unmanned vehicles.

[0067] Satellite navigation receiver: used to acquire the real-time position coordinates and heading angle of the unmanned vehicle, and output the real-time position coordinates and heading angle to the on-board main controller, with an update frequency of ≥1Hz.

[0068] Drive and actuator mechanism: including drive motor and steering mechanism, used to receive control commands output by the on-board main controller, drive the unmanned vehicle to drive and adjust the direction of travel of the unmanned vehicle, so as to realize the unmanned vehicle to travel along a smooth path.

[0069] The present invention also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the autonomous obstacle avoidance method for special unmanned vehicles as described above.

[0070] The special unmanned vehicle autonomous obstacle avoidance method of the present invention has the following beneficial effects:

[0071] 1. The special unmanned vehicle autonomous obstacle avoidance method of the present invention effectively solves the core problems such as high system cost, poor environmental adaptability, large map data volume, poor planning algorithm performance and unsatisfactory path quality through environmental modeling, vehicle localization, path planning, path execution and control. By integrating sparse environment modeling, satellite precise positioning and hybrid intelligent planning, a complete autonomous obstacle avoidance solution that does not rely on expensive real-time sensors and has both low cost and high reliability is constructed.

[0072] 2. This invention employs sparse binary maps and obstacle description methods for environmental characterization, completely eliminating the reliance on costly hardware such as lidar, millimeter-wave radar, and visual cameras. This not only significantly reduces the overall system cost but also substantially improves its environmental adaptability and reliability under complex weather conditions such as rain, snow, fog, haze, and strong light. It achieves stable operation based solely on pre-stored maps and satellite navigation, laying the foundation for the widespread application of low-cost special unmanned vehicles.

[0073] 3. Meanwhile, in terms of path planning, a hybrid strategy is adopted to combine global rapid exploration with local fine optimization, which effectively overcomes the technical defects of traditional single algorithms, such as being prone to getting trapped in local optima, slow convergence speed, and non-smooth generated paths. It can systematically generate high-quality driving paths that are shorter, have smoother turns, and are safer, which greatly improves the traffic efficiency and driving stability of vehicles. At the same time, efficient collision detection and lightweight computing ensure the real-time performance of the algorithm on limited hardware resources.

[0074] 4. In addition, the present invention has achieved optimal allocation of system resources through hardware and software co-optimization. The simplified map storage structure significantly reduces the storage and computing load of the vehicle controller. The closed-loop control of satellite positioning and path tracking ensures high-precision tracking of the vehicle on the planned path. In the end, an autonomous navigation system with excellent performance in terms of cost, performance and reliability has been realized, providing a reliable guarantee for the long-term stable operation of special unmanned vehicles in known scenarios. Detailed Implementation

[0075] The present invention will now be described in more detail through specific embodiments.

[0076] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship shown, and are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0077] Example 1

[0078] This embodiment provides an autonomous obstacle avoidance method for special unmanned vehicles based on pre-stored sparse maps and satellite navigation. This addresses the problems in existing obstacle avoidance methods, such as high system costs and poor environmental adaptability due to reliance on sensors, large amounts of pre-stored map data, poor planning algorithm performance leading to high hardware resource requirements, and unsatisfactory planned path quality.

[0079] The special unmanned vehicle autonomous obstacle avoidance method of this embodiment includes the following steps:

[0080] S1. Environment Modeling: Construct a sparse binary raster map of the known scene and store the map data using a preset compression storage method. The preset compression storage method in step S1 is the obstacle description method.

[0081] In the S1 environment modeling step, the known scene is abstracted into minimal obstacle location information, fundamentally eliminating the reliance on expensive real-time perception sensors such as LiDAR, significantly reducing system hardware costs, and making its environmental adaptability unaffected by weather and lighting conditions, thus enhancing the applicability of this method. Using the obstacle description method, combined with embedded system loading and parsing tests, for an 80km × 80km area, a preset compression storage method can compress the map data size from the GB level of traditional raster maps to approximately 80MB, allowing it to be directly loaded into the memory of a low-cost vehicle controller, thereby significantly improving system efficiency and feasibility.

[0082] S2. Vehicle Positioning: The real-time position coordinates and heading angle of the unmanned vehicle are obtained through satellite navigation equipment. The real-time position coordinates are then mapped onto the sparse binary grid map to obtain the current position of the unmanned vehicle on the map. , and orientation angle .

[0083] The S2 vehicle positioning process relies on satellite navigation equipment and utilizes mature, low-cost satellite positioning technology to achieve global positioning. By converting latitude and longitude coordinates into row and column numbers on a map in real time, it establishes the vehicle's precise pose in the virtual environment. In open terrain, this positioning solution can provide initial position information that meets accuracy requirements for path planning, ensuring that the vehicle can accurately start and execute the predetermined driving route.

[0084] S3. Path planning: Based on the current position and the preset target position... A hybrid path planning strategy is adopted to generate a collision-free smooth path. The hybrid path planning strategy includes global coarse path planning, local path fine optimization and path smoothing processing.

[0085] In this step, the global path coarse planning uses the RRT* algorithm, the local path fine optimization uses the improved crayfish optimization algorithm, and the path smoothing process uses cubic spline interpolation.

[0086] A hybrid strategy is adopted, combining the rapid global exploration capability of the RRT* algorithm with the fine local search capability of the improved biomimetic optimization algorithm. First, the RRT* algorithm quickly generates a collision-free initial path, and then the optimization algorithm smooths and optimizes it. This effectively overcomes the shortcomings of traditional single algorithms, such as being prone to getting trapped in local optima and having uneven paths. Simulation and experimental data show that this strategy can systematically generate higher-quality paths with shorter lengths and smoother turns, which is more conducive to stable vehicle tracking.

[0087] S4. Path Execution and Control: Control the unmanned vehicle to travel along the smooth path, obtain the real-time driving position of the unmanned vehicle through satellite navigation equipment, and perform closed-loop feedback control based on the deviation between the real-time driving position and the smooth path until the unmanned vehicle reaches the preset target position. .

[0088] In this step, closed-loop feedback control is performed based on the deviation between real-time satellite positioning and the planned path, forming a stable negative feedback system to ensure that the vehicle can travel accurately and reliably along the predetermined path to the target. Real vehicle test data supports that it can stably control the lateral tracking error within 0.2 meters.

[0089] In this embodiment, a pre-built sparse binary grid map is used for efficient environmental modeling, and satellite navigation is used for accurate positioning. A hybrid path planning strategy is employed to generate a safe and smooth path. Ultimately, an autonomous obstacle avoidance system is achieved that is completely independent of real-time environmental perception sensors, and features low cost, high reliability, and excellent path planning performance.

[0090] Example 2

[0091] The similarities to the above embodiments will not be repeated, the differences are as follows:

[0092] The preset compression storage method in step S1 is the obstacle description method, specifically:

[0093] For each row of a sparse binary raster map, store the row label for that row.

[0094] For each obstacle in a row, store its starting column position and the length of the grid it occupies consecutively. If there are no obstacles in a row, only the row flag is stored.

[0095] In the above-mentioned special unmanned vehicle autonomous obstacle avoidance method, when constructing a sparse binary grid map in step S1, the known scene is divided into uniform grids, the size of which is 0.5m-2m; grids with obstacles are marked as 1, and grids without obstacles are marked as 0.

[0096] In this embodiment, an obstacle description method is adopted. Taking advantage of the sparse distribution of obstacles in the known scene, the data storage structure is changed instead of directly storing the complete two-dimensional grid matrix. Only the effective obstacle information is recorded while ignoring the massive number of obstacle-free blank areas, thereby achieving extreme compression of map data. The grid map is traversed row by row using the above method. For each row, only a row marker is stored for positioning. For each obstacle block in the row, its starting column position and the length of the grid it occupies are recorded. If the entire row is obstacle-free, only the row marker is retained. Its compression efficiency is outstanding.

[0097] Through actual modeling tests on large-scale typical scenarios (such as an 80 km × 80 km training range), test data shows that when an 80 km × 80 km scenario is divided into 1 m × 1 m grids, the original binary grid map requires storing 6,400,000,000 grid values, with each grid occupying 1 bit, resulting in a total data volume of approximately 800 MB. After adopting the obstacle description method, the obstacle coverage in the scenario is approximately 2%-5%. Through actual coding tests and embedded system file read / write tests, it is verified that the data volume after storing all row markers and obstacle description information can be compressed to approximately 20-50 MB, a reduction of 16-40 times. In actual deployment, the ARM Cortex-A53 processor can complete map parsing and loading within 150-300 milliseconds, proving the feasibility of this compression scheme on resource-constrained platforms.

[0098] Furthermore, based on a comprehensive trade-off between real-vehicle passability testing and computational complexity, the grid size was set between 0.5 meters and 2 meters. In real-vehicle verification with a 1-meter grid precision, an autonomous vehicle with a width of 2.2 meters could stably pass through a passage with a minimum width of 3.2 meters. The path planning module's runtime decreased from 1.8 seconds with a fine map (0.2-meter grid) to 0.4 seconds, and the average CPU load on the controller decreased from 68% to below 25%. In addition, the method of recording the obstacle list significantly improved collision detection efficiency. Benchmark tests showed that, in typical scenarios, its collision detection speed was 8-15 times faster than traversing a complete two-dimensional grid map, providing a reliable guarantee for real-time path planning.

[0099] The satellite navigation device in step S2 is a GPS receiver or a Beidou satellite navigation receiver. The positioning update frequency of the satellite navigation device is ≥1Hz. When mapping the real-time position coordinates to a sparse binary grid map, the latitude and longitude coordinates output by the satellite navigation device are converted into row and column numbers in the sparse binary grid map through coordinate transformation.

[0100] In this embodiment, the latitude and longitude coordinates of the unmanned vehicle in the WGS-84 coordinate system are obtained by a GPS or Beidou satellite navigation receiver, and then mapped to the specific row and column positions in a pre-constructed sparse binary raster map by a coordinate transformation algorithm.

[0101] By leveraging the global coverage of satellite navigation, precise positioning is achieved in known environments, while avoiding reliance on local perception sensors such as vision and lidar, which are susceptible to environmental interference, significantly reducing system cost and complexity. Based on the needs of actual path tracking control, real-vehicle testing verified that the positioning update frequency is ≥1Hz. When using the ubloxF9P high-precision GNSS module, in open field environments, its positioning data output frequency can be configured up to 5Hz. Real-world testing shows that at a vehicle speed of 10m / s (36km / h), a 1Hz update frequency ensures that the maximum displacement of the vehicle within the control cycle does not exceed 10 meters. Combined with a PID controller, the path tracking error can be controlled within 0.5 meters. The coordinate transformation is implemented using a planar projection method. A local Cartesian coordinate system is established by selecting the scene center point as the origin. Experimental data shows that within an 80km×80km scene range, the projection distortion error of this method is less than 0.8 meters, fully meeting the accuracy requirements of a 1-meter grid map.

[0102] In practical deployment, calibration tests were conducted using 25 evenly distributed reference points. The root mean square error (RMSE) of this coordinate transformation method was 0.42 grid cells. The positioning scheme demonstrated good stability in a 24-hour continuous vehicle test, with an effective positioning data output rate of 99.2%. The positioning availability in the GPS+BeiDou dual-system mode reached 99.5%, significantly better than the 92.3% of the single GPS mode. Simultaneously, the use of carrier phase smoothing pseudorange technology improved the original positioning accuracy from 2-3 meters to 0.8-1.2 meters, providing a reliable positioning foundation for path tracking control.

[0103] Example 3

[0104] The similarities with the above embodiments and their combinations will not be repeated, the differences being:

[0105] The global path coarse planning in step S3 uses the RRT* algorithm, and the execution process of the RRT* algorithm includes:

[0106] Step 1: Use the current position of the autonomous vehicle as the root node. Initialize a random tree and set the preset target position. Set as the target area.

[0107] Step 2: Randomly sample points in the free space of the sparse binary raster map to generate sampling points. .

[0108] Step 3: Select the sampling point in the random tree. The nearest node is used as the reference node. With a preset probability Select the sampling point Or preset target location directional growth, expanding to generate new nodes according to a set stepsize. The probability The preferred value is 0.5.

[0109] Step 4: Query the sparse binary raster map to determine the baseline node. With new nodes If the path segment between them crosses an obstacle, then the new node is discarded. Then return to step 2; if the obstacle is not crossed, retain the new node. .

[0110] Step 5: Using the sampling points Using the set step size as the radius and center, find the set of potential parent nodes in the random tree that are within this radius, and calculate the set of parent nodes from the root node. From each potential parent node to the new node Given the path length, select the potential parent node with the shortest path length and no collision as the new node. The parent node.

[0111] Step 6: For the new node Rewire the existing nodes of the random tree within the radius range. If the new node is... Being the parent node of an existing node can shorten the distance from the existing node to the root node. If the path length is known and there are no collisions, then update the parent node of the existing node.

[0112] Step 7: Repeat steps 2 to 6 until a leaf node of the random tree enters the target area, and output an initial path consisting of multiple path points.

[0113] In this embodiment, the RRT* algorithm constructs a search tree by randomly sampling in the configuration space. Its probabilistic completeness ensures that a feasible path is found within a finite time, and the path quality is gradually improved through a continuous reconstruction and optimization mechanism. Real-vehicle experiments show that shortening the step size to 2 meters increases the planning time by 2.8 times, while increasing the step size to 8 meters leads to a 22% decrease in the throughput of narrow passages (width < 6 meters). Furthermore, comparative experiments verify that setting the probability parameter p=0.5, in a typical scenario with a passage density of 30%, the convergence speed is improved by 37% compared to a purely random strategy, and the local optimum occurrence rate is reduced by 58% compared to a purely goal-oriented strategy. With the path optimization and rewiring mechanism, the path length can be shortened by 12%-28% after thousands of iterations.

[0114] In practical applications, when the control platform equipped with an ARM Cortex-A53 processor processes an 80km×80km scene, the above algorithm can generate an initial path within 0.6±0.2 seconds. After 2500 iterations, the path length converges to within 107% of the theoretical optimal value. The collision detection module utilizes the row-oriented storage structure of a sparse binary map, achieving a single query time of 0.2±0.1 milliseconds, ensuring the real-time performance of the algorithm. This solution has undergone 187 simulated scenarios and 26 real-vehicle tests, achieving a 97.3% success rate in complex environments including continuous obstacle zones and sawtooth-shaped passages, providing a stable and reliable initial path for subsequent optimization stages.

[0115] The performance verification experimental data of the RRT* algorithm are shown in Table 1:

[0116] Table 1

[0117] Example 4

[0118] The similarities with the above embodiments and their combinations will not be repeated, the differences being:

[0119] The local path optimization in step S3 employs an improved crayfish optimization algorithm. The initialization process of the improved crayfish optimization algorithm is as follows:

[0120] The path points in the initial path output by the global coarse-scale path planning are used as the initial population, and the size of the initial population is equal to the number of path points in the initial path. Each individual in the population corresponds to a waypoint. A single individual in the population is represented as a 1×2 matrix containing the x and y coordinates of the corresponding waypoint, i.e., a two-dimensional vector. The entire initial population is represented as follows:

[0121] ;

[0122] For the first The x-coordinates of the path points For the first The y-coordinates of the path points Individuals in a single population are represented individually as .

[0123] In this embodiment, the path planning problem is transformed into a population optimization problem. By utilizing the high-quality initial paths provided by global coarse-macro programming as prior knowledge, the convergence efficiency of the optimization algorithm is significantly improved. The N path points output by the RRT* algorithm are directly mapped to the initial population of the improved crayfish optimization algorithm, and the two-dimensional coordinates of each path point are... Constituting a population of individuals The entire population then forms a matrix.

[0124] The aforementioned initialization strategy provides a starting point for the optimization algorithm, eliminating the need for random searching from scratch. Comparative experiments demonstrate that, under the same maximum number of iterations (T=100), the improved crayfish optimization algorithm using RRT* output as the initial population achieves approximately 85% faster convergence speed compared to the traditional random initialization method, and reduces the number of iterations required to obtain a satisfactory solution (path length not exceeding 110% of the theoretical optimal solution) by approximately 60% (verified through Monte Carlo simulation tests on 50 scenarios with different complexities).

[0125] Meanwhile, real-vehicle test data shows that the above initialization method reduces the probability of the optimization algorithm getting trapped in bad local optima from about 30% with random initialization to below 5%, thus ensuring the stability and reliability of the subsequent path refinement stage. This population initialization method, by utilizing the planning results from the previous stage, effectively solves the classic problem of slow convergence and poor performance in path optimization caused by poor initial solution quality in metaheuristic algorithms, laying the foundation for generating a smooth and safe final path.

[0126] Example 5

[0127] The similarities with the above embodiments and their combinations will not be repeated, the differences being:

[0128] The improved crayfish optimization algorithm also includes a stage switching and position update process, specifically:

[0129] a. Phase Switching: Calculating Virtual Temperature The calculation formula is:

[0130] ;

[0131] in for Random numbers within a range; when When the temperature exceeds 30℃, the algorithm enters a cooling-off phase.

[0132] when When the temperature is ≤30℃, the algorithm enters the foraging phase.

[0133] b. Location update during the summer retreat phase: Calculate cave locations The calculation formula is:

[0134] ;

[0135] in This represents the historical best position for an individual in the current population. This is the global optimal position for the current population.

[0136] when When the value is ≤0.5, the position of an individual in the population is updated according to the formula:

[0137] ;

[0138] in This represents the current iteration number. This is the number of the next iteration. The maximum number of iterations, the decreasing coefficient. , For the first During the nth iteration Individual in the population The coordinates of the dimension.

[0139] when When the value is greater than 0.5, the position of an individual in the population is updated according to the formula:

[0140] ;

[0141] Random individual index , For the first During the nth iteration Individual in the population The coordinates of the dimension.

[0142] c. Location update during foraging phase: Calculate food size The calculation formula is:

[0143] ;

[0144] in The adjustment factor is set to 3. This represents the fitness value of an individual in the current population. The fitness value for the preset target location.

[0145] Define auxiliary variables .

[0146] according to The size of the population is used to update the position of individual individuals according to the formula:

[0147] ;

[0148] The stride control parameter, i.e., the probability coefficient, is based on the optimal temperature of 30℃ and the normal distribution of intake, and is used to dynamically adjust the step size for updating the position of individuals in the population.

[0149] d. Iteration Termination: The process of switching between iteration phases and updating positions continues until the maximum number of iterations is reached. If the fitness value converges, the optimized path points are output.

[0150] Furthermore, the fitness value The calculation formula is:

[0151] ;

[0152] in: The total path length is calculated using the following formula:

[0153] ;

[0154] Let the dimension be the path point sequence. , The first The x and y coordinates of each path point.

[0155] The formula for calculating path smoothness is:

[0156] ;

[0157] For the first The path point and the first The heading angle of the line connecting the path points is calculated using the following formula:

[0158] .

[0159] The obstacle penalty is calculated as follows:

[0160] ;

[0161] This represents the distance between the pathpoint of an individual in the current population and the nearest obstacle. This is the penalty coefficient. These are weighting coefficients, adjusted according to actual path requirements. If a greater emphasis is placed on path length, the weighting factor increases. Focusing on path smoothness increases .

[0162] In this embodiment, by simulating the heat-avoidance and foraging behaviors of crayfish, an intelligent optimization mechanism is generated that can adaptively balance global exploration and local exploitation capabilities. Its stage switching mechanism is controlled by virtual temperature. for The virtual temperature range is 20~35℃, which is a random number within the range.

[0163] When the temperature exceeds 30℃, the region enters a heat-seeking phase, engaging in localized development by approaching caves formed by the individual's historical best and the global best positions; when the temperature is below or equal to 30℃, the region enters a foraging phase, developing based on the size of the food. The value dynamically selects the search strategy when When Q > 2, conduct large-scale exploration; when Q ≤ 2, conduct localized fine-tuning.

[0164] The above-mentioned two-stage adaptive switching mechanism can effectively avoid the defect of traditional optimization algorithms (such as the standard particle swarm optimization algorithm) being prone to getting trapped in local optima. Comparative experiments based on 85 sets of simulated scenarios show that, compared with the standard particle swarm optimization algorithm, this mechanism can increase the probability of the algorithm escaping local optima by more than 40% (the algorithm is judged as trapped in local optima if the optimal solution is not updated after 50 consecutive iterations).

[0165] Decreasing coefficient in the algorithm ( This represents the current iteration number. (Maximum number of iterations) ensures that the search step size gradually decreases with each iteration, thus achieving a smooth transition from global exploration to local development. Within the maximum number of iterations... With a value of 100, simulation tests show that this design enables the algorithm to converge after about 60 generations, and the stability of the convergence process is improved by about 30% compared to the version without the decreasing strategy.

[0166] In addition, a comprehensive evaluation method is adopted, weighted by path length, smoothness, and obstacle penalty, where the obstacle penalty is based on distance. A penalty is imposed when the distance is less than 1m. When the distance is ≥1m, an inverse squared penalty is applied. Real vehicle test data proves that this method can ensure that the optimized path maintains an average safe distance of more than 1.2 meters from obstacles, while reducing the total curvature change of the path by about 50%.

[0167] Weighting coefficient Adjusted according to actual needs, tests show that when : When set to 2:1, the best balance between path length and smoothness is achieved, so that the final total path length increases by only about 5% compared to the initial path, while the smoothness is improved by about 60%.

[0168] After verification through 85 simulated scenarios and 18 real-vehicle tests, Within the iteration range of 100, the fitness value can converge to a stable state, and the final generated path is on average about 18% shorter than the initial path length provided by RRT*, providing a safe, smooth and near-optimal driving path for autonomous vehicles.

[0169] The performance verification experimental data of the improved crayfish optimization algorithm are shown in Table 2:

[0170] Table 2

[0171] Example 6

[0172] The similarities with the above embodiments and their combinations will not be repeated, the differences being:

[0173] The path smoothing process in step S3 uses cubic spline interpolation, specifically: a cubic polynomial is constructed between the optimized path points output by local path optimization, so that the generated smooth path remains continuous in position, first derivative and second derivative.

[0174] In this embodiment, piecewise cubic polynomial curves are constructed between adjacent path points. Mathematical constraints ensure that adjacent curve segments have the same position, first derivative (tangent direction), and second derivative (curvature) at the connection point, thereby forming a globally smooth path. The generated path is not only continuous and smooth, but also has the mathematical property of second-order continuous differentiability, which directly corresponds to the requirements of position, velocity, and acceleration continuity in vehicle kinematics.

[0175] Real-vehicle testing showed that after employing cubic spline interpolation, the lateral tracking error of the path was reduced by approximately 60% compared to the linear interpolation path, decreasing from an average of 0.50 meters to less than 0.20 meters. Simultaneously, analysis of 85 optimized path points revealed that this method can reduce the maximum curvature of the path (directly related to the second derivative) by approximately 45%, significantly reducing the frequency of adjustments and wear on the vehicle's steering actuators. In 18 real-vehicle verifications, the standard deviation of lateral acceleration when the vehicle travels along the spline path was reduced by approximately 50% compared to the unsmoothed path, directly demonstrating the engineering value of acceleration (second derivative) continuity and effectively improving ride comfort and driving stability.

[0176] Smoothing, as the final step in path planning, transforms discrete optimized path points into a continuous trajectory truly suitable for vehicle tracking. Its processing time is tested to be an average of 8 milliseconds, ensuring the real-time performance of the entire planning process and ultimately outputting a physically executable, high-quality path.

[0177] Meanwhile, this embodiment discloses a computer-readable storage medium storing a computer program that, when executed by a processor, implements the autonomous obstacle avoidance method for special unmanned vehicles as described above.

[0178] Example 7

[0179] The similarities with the above embodiments and their combinations will not be repeated, the differences being:

[0180] This embodiment discloses a special-purpose unmanned vehicle system for implementing the above-described autonomous obstacle avoidance method for special-purpose unmanned vehicles, including:

[0181] The onboard main controller is pre-stored with the sparse binary grid map compressed using the obstacle description method, and is equipped with a program module that performs environmental modeling, path planning, and closed-loop feedback control in the autonomous obstacle avoidance method for special unmanned vehicles.

[0182] Satellite navigation receiver: used to acquire the real-time position coordinates and heading angle of the unmanned vehicle, and output the real-time position coordinates and heading angle to the vehicle main controller. The update frequency is ≥1Hz. The satellite navigation receiver supports GPS + Beidou dual system.

[0183] Drive and actuator mechanism: including drive motor and steering mechanism, used to receive control commands output by the on-board main controller, drive the unmanned vehicle to drive and adjust the direction of travel of the unmanned vehicle, so as to realize the unmanned vehicle to travel along a smooth path.

[0184] In this embodiment, a complete autonomous navigation system, independent of real-time environmental perception sensors, is formed through the collaborative operation of three core components: the vehicle-mounted main controller, the satellite navigation receiver, and the drive actuator. This hardware-software integration achieves extremely low system cost and high environmental adaptability. The vehicle-mounted main controller's pre-stored sparse binary grid map, compressed using obstacle description methods, enables smooth operation on low-cost processors such as the ARM Cortex-A53. The path planning module's total computation cycle is 0.6-1.0 seconds, fully meeting real-time requirements. The satellite navigation receiver employs a GPS / BeiDou dual-system design. Real-vehicle testing verifies its positioning update frequency ≥5Hz, positioning accuracy 0.8-1.2 meters, and system availability 99.5%, providing stable and reliable pose feedback for the control system. The drive and actuator stably control the lateral error of path tracking within 0.5 meters, ensuring accurate tracking of the smooth path generated by cubic spline interpolation.

[0185] The entire system underwent 85 simulated scenarios and 18 real vehicle tests, achieving a 98.1% autonomous obstacle avoidance success rate in complex environments including U-shaped obstacles and narrow passages. At the same time, the system's hardware cost was reduced by approximately 70% compared to solutions equipped with lidar, fully demonstrating the feasibility and superiority of the integrated system in this embodiment for low-cost special unmanned vehicle applications.

[0186] It should be noted that, in actual implementation, the structure described in this specification is not a fixed or unchanging embodiment. The components of the embodiments of the present invention described and shown herein can be arranged and designed in various different configurations. These are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Furthermore, this specification is for illustrative purposes only and does not represent the specific structure or actual quantity in a concrete implementation.

[0187] Unless otherwise defined, the technical or scientific terms used herein should be understood in their ordinary sense as would be understood by one of ordinary skill in the art to which this invention pertains. The use of terms such as "a" or "an" in this specification and claims does not necessarily indicate a limitation of quantity. Terms such as "comprising" or "including" mean that the element or component preceding the word encompasses the element or component listed following the word and its equivalents, without excluding other elements or components. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0188] The exemplary embodiments of the present invention have been described in detail above with reference to preferred embodiments. However, those skilled in the art will understand that various modifications and alterations can be made to the above specific embodiments without departing from the concept of the present invention, and various combinations can be made to the various technical features and structures proposed in the present invention without exceeding the protection scope of the present invention.

Claims

1. A method for autonomous obstacle avoidance of a special unmanned vehicle, characterized in that, Includes the following steps: S1. Environment Modeling: Construct a sparse binary raster map of the known scene and store the map data using a preset compression storage method; the preset compression storage method is the obstacle description method; Specifically: for each row of the sparse binary raster map, store the row identifier; for each obstacle in the row, store the starting column position of the obstacle and the length of the raster it occupies consecutively; if there are no obstacles in the row, only store the row identifier. S2. Vehicle Positioning: The real-time position coordinates and heading angle of the unmanned vehicle are obtained through satellite navigation equipment. The real-time position coordinates are then mapped onto the sparse binary grid map to obtain the current position of the unmanned vehicle on the map. , and orientation angle ; S3. Path planning: Based on the current position and the preset target position... A hybrid path planning strategy is adopted to generate a collision-free smooth path. The hybrid path planning strategy includes global coarse path planning, local path fine optimization and path smoothing processing in sequence. The local path optimization in step S3 employs an improved crayfish optimization algorithm. The initialization process of the improved crayfish optimization algorithm is as follows: The path points in the initial path output by the global coarse-scale path planning are used as the initial population, and the size of the initial population is equal to the number of path points in the initial path. Each individual in the population corresponds to a waypoint. A single individual in the population is represented as a 1×2 matrix containing the x and y coordinates of the corresponding waypoint, i.e., a two-dimensional vector. The entire initial population is represented as follows: ； For the first The x-coordinates of the path points For the first The y-coordinates of the path points ; Individuals in a single population are represented as ; Fitness value in the improved crayfish optimization algorithm The calculation formula is: ； in: The total path length is calculated using the following formula: ； Let the dimension be the path point sequence. , The first The x and y coordinates of each path point; The formula for calculating path smoothness is: ； For the first The path point and the first The heading angle of the line connecting the path points is calculated using the following formula: ； The obstacle penalty is calculated as follows: ； This represents the distance between the pathpoint of an individual in the current population and the nearest obstacle. The penalty coefficient is... These are weighting coefficients, adjusted according to actual path requirements. If a greater emphasis is placed on path length, the weighting factor increases. Focusing on path smoothness increases ; S4. Path Execution and Control: Control the unmanned vehicle to travel along the smooth path, obtain the real-time driving position of the unmanned vehicle through satellite navigation equipment, and perform closed-loop feedback control based on the deviation between the real-time driving position and the smooth path until the unmanned vehicle reaches the preset target position. .

2. The special unmanned vehicle autonomous obstacle avoidance method according to claim 1, characterized in that, The satellite navigation device in step S2 is a GPS receiver or a Beidou satellite navigation receiver. The positioning update frequency of the satellite navigation device is ≥1Hz. When mapping the real-time position coordinates to a sparse binary grid map, the latitude and longitude coordinates output by the satellite navigation device are converted into row and column numbers in the sparse binary grid map through coordinate transformation.

3. The special unmanned vehicle autonomous obstacle avoidance method according to claim 1 or 2, characterized in that: In step S1, when constructing a sparse binary grid map, the known scene is divided into uniform grids, the size of which is 0.5m-2m; grids with obstacles are marked as 1, and grids without obstacles are marked as 0.

4. The special unmanned vehicle autonomous obstacle avoidance method according to claim 1, characterized in that, The global path coarse planning in step S3 uses the RRT* algorithm, and the execution process of the RRT* algorithm includes: Step 1: Use the current position of the autonomous vehicle as the root node. Initialize a random tree and set the preset target position. Set as the target area; Step 2: Randomly sample points in the free space of the sparse binary raster map to generate sampling points. ; Step 3: Select a location from the sampling point in the random tree. The nearest node is used as the reference node. With a preset probability Select the sampling point Or preset target location directional growth, expanding to generate new nodes according to a set step size. The probability It is 0.5; Step 4: Query the sparse binary raster map to determine the baseline node. With new nodes If the path segment between them crosses an obstacle, then the new node is discarded. Then return to step 2; if the obstacle is not crossed, retain the new node. ; Step 5, using the sampling points Using the set step size as the radius and center, find the set of potential parent nodes in the random tree that are within this radius, and calculate the set of parent nodes from the root node. From each potential parent node to the new node Given the path length, select the potential parent node with the shortest path length and no collision as the new node. The parent node; Step 6, for the new node Rewire the existing nodes of the random tree within the radius range. If the new node is... Being the parent node of an existing node can shorten the distance from the existing node to the root node. If the path length is known and there are no collisions, then update the parent node of the existing node; Step 7: Repeat steps 2 to 6 until a leaf node of the random tree enters the target area, and output an initial path consisting of multiple path points.

5. The special unmanned vehicle autonomous obstacle avoidance method according to claim 1 or 4, characterized in that, The improved crayfish optimization algorithm also includes a stage switching and position update process, specifically: a. Phase switching; calculating virtual temperature The calculation formula is: ； in for Random numbers within a range; when When the temperature exceeds 30℃, the algorithm enters a cooling-off phase; when When the temperature is ≤30℃, the algorithm enters the foraging phase; b. Location update during the summer retreat phase; calculate cave locations. The calculation formula is: ； in This represents the historical best position for an individual in the current population. This represents the globally optimal position for the current population. when When the value is ≤0.5, the position of an individual in the population is updated according to the formula: ； in This represents the current iteration number. This is the number of the next iteration. The maximum number of iterations, the decreasing coefficient. , For the first During the nth iteration Individual in the population Dimensional coordinates; when When the value is greater than 0.5, the position of an individual in the population is updated according to the formula: ； Random individual index , For the first During the nth iteration Individual in the population Dimensional coordinates; c. Location update during foraging phase; Calculate food size The calculation formula is: ； in The adjustment factor is set to 3. This represents the fitness value of an individual in the current population. The fitness value for the preset target location; Define auxiliary variables ; according to The size of the population is used to update the position of individual individuals according to the formula: ； The stride control parameter, i.e., the probability coefficient, is based on the normal distribution of the optimal temperature of 30℃ and the intake, and is used to dynamically adjust the step size for updating the position of individuals in the population. d. Iteration termination; the process of switching phases and updating positions is repeated until the maximum number of iterations is reached. If the fitness value converges, the optimized path points are output.

6. The special unmanned vehicle autonomous obstacle avoidance method according to claim 1, characterized in that, The path smoothing process in step S3 uses cubic spline interpolation, specifically: a cubic polynomial is constructed between the optimized path points output by the local path optimization, so that the generated smooth path is continuous in position, first derivative and second derivative.

7. A special-purpose unmanned vehicle system for implementing the autonomous obstacle avoidance method for special-purpose unmanned vehicles according to any one of claims 1-6, characterized in that, include: The onboard main controller is pre-stored with the sparse binary grid map compressed using the obstacle description method, and is equipped with a program module that performs environmental modeling, path planning, and closed-loop feedback control in the autonomous obstacle avoidance method for special unmanned vehicles. Satellite navigation receiver: used to acquire the real-time position coordinates and heading angle of the unmanned vehicle, and output the real-time position coordinates and heading angle to the on-board main controller, with an update frequency of ≥1Hz; Drive and actuator mechanism: including drive motor and steering mechanism, used to receive control commands output by the on-board main controller, drive the unmanned vehicle to drive and adjust the direction of travel of the unmanned vehicle, so as to realize the unmanned vehicle to travel along a smooth path.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the autonomous obstacle avoidance method for special unmanned vehicles as described in any one of claims 1-6.

Images