Path planning based autonomous obstacle avoidance method and system for unmanned vehicle and medium

Abstract

This invention discloses an autonomous obstacle avoidance method, system, and medium for unmanned vehicles based on path planning, relating to the field of intelligent driving technology. The method includes: analyzing obstacles at path nodes according to path planning to establish an obstacle distribution probability map for each path node; performing monitoring and matching based on the obstacle distribution probability map to determine the monitoring target range and obstacle avoidance analysis path strategy; acquiring current monitoring data based on the monitoring target range; matching path factors based on the current monitoring data and the obstacle avoidance analysis path strategy to activate the path analysis strategy and generate autonomous obstacle avoidance control commands. This invention solves the technical problems of traditional unmanned vehicles struggling to accurately identify obstacles and dynamic risks at different locations along the driving route, lacking reasonable response logic in the face of diverse and complex road conditions, and exhibiting insufficient obstacle avoidance accuracy and reliability. It achieves the technical effect of improving the vehicle's obstacle avoidance capability in complex environments, ensuring driving safety, and adhering to a preset driving route.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving technology, and in particular to an autonomous obstacle avoidance method, system and medium for unmanned vehicles based on path planning. Background Technology

[0002] Autonomous vehicles (RVs) have been widely used in transportation, industrial parks, and other fields. Autonomous obstacle avoidance based on path planning is a core technology for ensuring their safe and efficient operation, and accurate assessment of obstacles around path nodes is crucial for obstacle avoidance effectiveness. Currently, the industry mostly adopts conventional perception and basic obstacle avoidance solutions. This approach can achieve basic obstacle avoidance in scenarios with simple road conditions and few obstacles. However, as the application scenarios of RVs continue to expand and operational requirements continue to increase, traditional obstacle avoidance technologies are gradually revealing many limitations. Affected by factors such as the layout of path nodes and the dynamic and ever-changing movement of obstacles, traditional solutions cannot accurately analyze the obstacle distribution status of each path node, making it difficult to comprehensively quantify risks. The collected data is partial and lacks precision, failing to support RVs in conducting precise obstacle avoidance and failing to meet the driving control needs under complex road conditions. Summary of the Invention

[0003] This application provides a path planning-based autonomous obstacle avoidance method, system, and medium for unmanned vehicles, which solves the technical problems of traditional unmanned vehicles having difficulty accurately identifying obstacles and dynamic risks at different locations along the route, lacking reasonable response logic when facing diverse and complex road conditions, and having insufficient obstacle avoidance accuracy and reliability.

[0004] The first aspect of this application provides an autonomous obstacle avoidance method for unmanned vehicles based on path planning. The method includes: analyzing obstacles at path nodes according to path planning to establish an obstacle distribution probability map for each path node; performing monitoring and matching based on the obstacle distribution probability map to determine the monitoring target range and obstacle avoidance analysis path strategy; obtaining current monitoring data according to the monitoring target range; performing path factor matching based on the current monitoring data and the obstacle avoidance analysis path strategy to activate the path analysis strategy and generate autonomous obstacle avoidance control commands.

[0005] A second aspect of this application provides an autonomous obstacle avoidance system for unmanned vehicles based on path planning. The system includes: an obstacle distribution probability map construction module, used to analyze obstacles at path nodes according to path planning and establish an obstacle distribution probability map for each path node; a monitoring target range determination module, used to perform monitoring matching based on the obstacle distribution probability map to determine the monitoring target range and obstacle avoidance analysis path strategy; and an autonomous obstacle avoidance control command generation module, used to obtain current monitoring data based on the monitoring target range, perform path factor matching based on the current monitoring data and the obstacle avoidance analysis path strategy, activate the path analysis strategy, and generate autonomous obstacle avoidance control commands.

[0006] A third aspect of this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the path planning-based autonomous obstacle avoidance method for unmanned vehicles provided in this application.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] This application analyzes obstacles at each node of the planned path and constructs an obstacle distribution probability map. Through processing such as region segmentation, feature analysis, and template matching, it obtains the risk range and obstacle avoidance scheme, extracts real-time driving-related parameters and completes index comparison, and combines preset judgment rules to adapt the strategy, thereby realizing autonomous obstacle avoidance of unmanned vehicles in dynamic environments. This makes the obstacle avoidance operation of unmanned vehicles in complex road conditions more accurate and reliable, achieving the technical effect of improving the obstacle avoidance capability of vehicles in complex environments, ensuring driving safety, and conforming to the preset driving route. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating the autonomous obstacle avoidance method for unmanned vehicles based on path planning provided in this application embodiment.

[0011] Figure 2 This is a schematic diagram of the structure of the autonomous obstacle avoidance system for unmanned vehicles based on path planning provided in the embodiments of this application.

[0012] Figure labeling: Obstacle distribution probability map construction module 1, monitoring target range determination module 2, autonomous obstacle avoidance control command generation module 3. Detailed Implementation

[0013] This application provides a path planning-based autonomous obstacle avoidance method, system, and medium for unmanned vehicles, which solves the technical problems of traditional unmanned vehicles having difficulty accurately identifying obstacles and dynamic risks at different locations along the route, lacking reasonable response logic when facing diverse and complex road conditions, and having insufficient obstacle avoidance accuracy and reliability.

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0015] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0016] Example 1, as Figure 1 As shown, an autonomous obstacle avoidance method for unmanned vehicles based on path planning is described, wherein the method includes:

[0017] Based on the path planning, the obstacles at the path nodes are analyzed, and a probability distribution map of obstacles at each path node is established.

[0018] Specifically, the neighborhood space of the planned path nodes is divided into grids. The occupancy probability of each grid within a set time window is calculated by combining the prior information of the static map with the kinematic model of the dynamic obstacle. Then, sub-probability maps are constructed for different types of dynamic obstacles and weighted and fused to obtain a comprehensive obstacle distribution probability map. The specific implementation process of this step will be described in detail later.

[0019] Based on the obstacle distribution probability map, monitoring and matching are performed to determine the monitoring target range and obstacle avoidance analysis path strategy.

[0020] Optionally, the obstacle distribution probability map is segmented into regions, and regions with probability values ​​exceeding the first threshold are selected as candidate monitoring ranges. The obstacle-related features within the candidate regions are calculated and the optimal obstacle avoidance response template is matched in combination with the current motion state of the unmanned vehicle. This generates an obstacle avoidance parsing path strategy that includes strategy type and key parameters. At the same time, the final monitoring range is locked based on a second threshold that is not lower than the first threshold, and the sensor parameters are adjusted for perception focusing. The specific implementation process of this step will also be described in detail later.

[0021] The current monitoring data is obtained based on the monitoring target range. The path factor is matched with the obstacle avoidance analysis path strategy based on the current monitoring data, the path analysis strategy is activated, and an autonomous obstacle avoidance control command is generated.

[0022] In one embodiment of this application, current monitoring data is first obtained based on the finally locked monitoring target range. This monitoring data comes from the fusion output of the vehicle-mounted LiDAR and visual camera, and includes the real-time position, speed, and direction of movement of obstacles, as well as the measured occupancy probability of each grid within the monitoring target range. The monitoring data is updated at a frequency of 50 Hz. After each data acquisition, timestamp alignment is performed to ensure that the time error between the LiDAR data and the visual data is less than 10 milliseconds. Then, the data is converted into a unified format in the vehicle coordinate system.

[0023] Next, path factors are extracted from the current monitoring data, including collision time, relative velocity change rate, obstacle movement direction change rate, and measured occupancy probability. Simultaneously, based on the calculated obstacle distribution probability map, the probability decay gradient within the range of each candidate monitoring target is extracted: starting from the probability peak grid, sampling points are taken along the vehicle's current driving direction in 0.5-meter increments until leaving the candidate area boundary; the average occupancy probability at each sampling point is calculated, and a univariate linear regression is used to obtain the slope of probability change with distance. This slope is the one-dimensional probability decay gradient of the area, in units of per meter. A positive gradient indicates that the probability increases with distance, while a negative gradient indicates that the probability decreases with distance; its absolute value reflects the drastic change in risk. These path factors will be used in the cascaded matching process described later, which includes an initial screening stage and a precise authentication stage. Finally, based on the activated path parsing strategy, an autonomous obstacle avoidance control command containing the desired deceleration and desired lateral offset is generated. The specific implementation process of this step is detailed in the corresponding paragraphs below.

[0024] Furthermore, the method provided in this application embodiment includes:

[0025] Based on the path nodes of the path planning, the space within the neighborhood of each path node is divided into several grids; based on the prior information of the static map and the kinematic model of the dynamic obstacles, the probability of each grid being occupied within a set time window is calculated to obtain the probability distribution; based on the probability distribution, sub-probability maps are established for different categories of dynamic obstacles, and a comprehensive obstacle distribution probability map is obtained through weighted fusion.

[0026] Specifically, a grid map method is first used for spatial division. Each path node output by the global path planning is used as the center to determine the neighborhood monitoring range of that node. The radius of the neighborhood monitoring range is determined by the current driving speed of the autonomous vehicle and the emergency braking distance, specifically the sum of the current driving speed multiplied by the braking response time and the safe braking distance, while not exceeding the maximum effective detection range of the onboard sensors. The neighborhood monitoring range is uniformly divided into several square grids of the same size. Each grid represents an independent spatial unit used to characterize the obstacle occupancy status at that location. Specific parameters are shown in Table 1.

[0027] Table 1: Path Node Neighborhood Grid and Calculation Parameters

[0028] Parameter name Parameter values unit Remark Single grid space size 0.5×0.5 m Path Node Neighborhood Partitioning Standard Single probability calculation time window 1 s Grid occupancy probability statistics duration Total number of neighboring grid cells for a single path node 400 indivual 20×20 uniformly distributed effective monitoring grid

[0029] The occupancy probability of each grid cell within a set time window is calculated by combining prior information from the static map with the kinematic model of dynamic obstacles. The prior information from the static map is extracted from the vehicle-mounted high-precision map and includes the precise location coordinates of static obstacles such as curbs, guardrails, fixed buildings, and traffic signs. For grid cells completely covered by static obstacles, an initial occupancy probability of 0.9 is directly assigned. For grid cells completely free of static obstacles, the initial occupancy probability is set to 0.1. For grid cells only partially covered by static obstacles, the occupancy probability is calculated based on the proportion of the covered area; that is, the occupancy probability equals the proportion of the area covered by static obstacles to the total area of ​​the grid multiplied by 0.9, with a value of 0.1 if the result is lower. The motion state of dynamic obstacles is acquired in real time by the vehicle-mounted perception module, including the obstacle's position, speed, direction of motion, and the size of its circumscribed rectangle. A mature constant-velocity motion model from existing technologies is used to predict the trajectory of dynamic obstacles within the set time window, with a time step of 0.1 seconds. The percentage of time each grid cell is covered by the circumscribed rectangle of the predicted trajectory of a dynamic obstacle within a time window is calculated. Coverage is defined as any part of the circumscribed rectangle entering the grid cell. This percentage is multiplied by a dynamic confidence coefficient of 0.8 to obtain the grid cell occupancy probability corresponding to the dynamic obstacle. The dynamic confidence coefficient of 0.8 is determined based on the average detection accuracy of the onboard perception module for dynamic obstacles and can be adjusted to 0.6 in low visibility weather conditions such as rain and fog.

[0030] Subsequently, based on the obstacle classification results output by the vehicle-mounted perception module, dynamic obstacles are divided into three categories: motor vehicles, non-motor vehicles, and pedestrians. An independent sub-probability map is established for each category of dynamic obstacles. The grid structure of the sub-probability map is completely consistent with the aforementioned path node neighborhood grid. Each sub-probability map only contains the grid occupancy probability calculated for the corresponding category of dynamic obstacles. Static probability serves as a basic term and does not participate in the construction of the sub-probability map. A weighted average method is used to fuse the three sub-probability maps, assigning a corresponding fusion weight coefficient to each sub-probability map. The weight coefficients are determined based on the collision risk level of different obstacle categories: motor vehicles have the highest collision energy and are assigned a weight of 0.5, pedestrians are assigned a weight of 0.3, and non-motor vehicles are assigned a weight of 0.2. In pedestrian-dense park scenarios, the pedestrian weight can be adjusted to 0.4, and the non-motor vehicle weight to 0.1. The fused dynamic probability of each grid is obtained by multiplying the probability value of each grid in the three sub-probability maps by the corresponding weight coefficient and then summing the results. The static occupancy probability is superimposed with the fused dynamic probability. The superposition result is then normalized: if the result is greater than 1, it is set to 1; if it is less than 0, it is set to 0. This yields the final occupancy probability of the grid. The final occupancy probabilities of all grids together constitute the comprehensive obstacle distribution probability map.

[0031] Through the steps of dividing the path node neighborhood grid, calculating the grid occupancy probability step by step, and weighted fusion of the sub-probability maps, the risk of obstacle distribution along the path is accurately quantitatively represented, providing reliable basic data support for the generation of subsequent obstacle avoidance strategies.

[0032] Furthermore, the method provided in this application embodiment includes:

[0033] The obstacle distribution probability map is segmented into regions, and several high-probability regions with probability values ​​exceeding a first threshold are extracted as candidate monitoring target ranges. The center position, peak probability, and probability decay gradient with distance of the obstacle distribution within each candidate monitoring target range are calculated. Based on the current motion state of the autonomous vehicle, the decay gradient is matched with several preset obstacle avoidance response templates to determine the response template with the highest matching degree. Based on the response template with the highest matching degree, a corresponding obstacle avoidance parsing path strategy is generated, which includes strategy type and its key parameters. Simultaneously, sub-regions with probability values ​​exceeding a second threshold within the candidate monitoring target range are locked as the final monitoring target range, and the sensor perception parameters are adjusted to focus on the corresponding range, wherein the second threshold is ≥ the first threshold.

[0034] Optionally, firstly, a connected component analysis method is used to segment the obstacle distribution probability map into regions. The comprehensive obstacle distribution probability map is first converted into a binary image, and a first threshold of 0.3 is set. This threshold is determined based on the balance between the average false detection rate and false negative rate of the vehicle-mounted perception module. When the false detection rate is higher than 10%, the first threshold is increased to 0.35; when the false negative rate is higher than 15%, the first threshold is decreased to 0.25. Grids with probability values ​​greater than or equal to the first threshold are marked as foreground, and grids with probability values ​​less than the first threshold are marked as background. An 8-neighborhood connected component analysis algorithm is used to process the binary image, extracting independent regions composed of all interconnected foreground grids. Each independent region is a candidate monitoring target range. Each candidate monitoring target range is numbered, and the coordinates of all grids it contains are recorded, completing the initial screening of candidate monitoring target ranges.

[0035] For each candidate monitoring target range, the obstacle distribution characteristics are calculated. The centroid calculation method is used to determine the center position of the obstacle distribution. The coordinates of the center position are obtained by multiplying the horizontal and vertical coordinates of each grid cell within the region by its occupancy probability, summing the results, and then dividing by the sum of the occupancy probabilities of all grid cells within the region. The maximum occupancy probability of all grid cells within the region is extracted as the probability peak value for that region. When calculating the probability decay gradient with distance, starting from the probability peak grid cell, sampling points are taken along the current driving direction of the vehicle at 0.5-meter steps until leaving the candidate region boundary. The average occupancy probability at each sampling point is calculated, and a univariate linear regression is used to obtain the slope of the probability change with distance. This slope is the one-dimensional probability decay gradient of the region, in units of per meter. When multiple candidate monitoring target ranges exist, the above feature calculations are performed in parallel for each range. The current driving speed and heading angle of the autonomous vehicle are obtained, and the calculated decay gradient is compared with the standard decay gradient in the preset obstacle avoidance response template library.

[0036] The preset obstacle avoidance response template library contains four types of standard attenuation gradients: the basic hierarchical template corresponds to a standard attenuation gradient range of -0.05 to -0.15 per meter; the multi-obstacle avoidance overlapping template corresponds to a standard attenuation gradient range of -0.2 to -0.4 per meter; the obstacle motion posture classification template corresponds to a standard attenuation gradient range of -0.1 to -0.2 per meter with positive and negative fluctuations; and the sudden vehicle trajectory node overlap matching template corresponds to a standard attenuation gradient range of -0.4 to -0.6 per meter. The Euclidean distance method is used to calculate the matching degree between the attenuation gradient of each candidate region and each standard gradient. The smaller the matching degree value, the higher the matching degree. The obstacle avoidance response template with the highest matching degree is selected as the optimal matching template for that candidate region. When multiple candidate regions match successfully at the same time, they are sorted from highest to lowest probability peak value, and the region with the highest probability peak value is processed first.

[0037] Based on the identifier of the optimal matching template, the corresponding strategy type and key parameters are extracted from the preset obstacle avoidance strategy parameter library, and combined to generate an obstacle avoidance parsing path strategy and temporarily stored. In the preset obstacle avoidance strategy parameter library, the key parameters corresponding to the basic hierarchical template include a basic avoidance distance of 1.5 meters to 3 meters and a basic deceleration of 2 m / s² to 4 m / s²; the key parameters corresponding to the multi-obstacle avoidance overlapping template include a priority coefficient of 1 to 3 and a quadratic planning step size of 0.5 meters; the key parameters corresponding to the obstacle motion attitude classification template include an attitude judgment time window of 0.5 seconds and a direction offset threshold of 15 degrees; the key parameters corresponding to the sudden vehicle trajectory node overlap matching template include an emergency deceleration of 8 m / s² to 10 m / s² and an acceleration threshold of 3 m / s².

[0038] Meanwhile, within the selected candidate monitoring target range, a second threshold of 0.6 is set. This threshold is determined based on the collision probability critical value of high-risk areas. In high-speed driving scenarios, the second threshold is increased to 0.7, and in low-speed park scenarios, it is decreased to 0.5. Grids with probability values ​​greater than or equal to the second threshold are re-marked as high-risk foreground. The same 8-neighborhood connected component analysis method is used to extract sub-regions formed by these high-risk foregrounds, and these sub-regions are locked as the final monitoring target range. The perception parameters of the vehicle-mounted sensors are adjusted to focus on the final monitoring target range. For LiDAR, the scan line density within the final monitoring target range is increased to twice its original value, while the scan line density in non-focused areas is reduced to one-third of its original value. For the visual camera, the gimbal angle is adjusted so that the camera's optical axis is aligned with the center of the final monitoring target range, and the image exposure time in this area is appropriately increased to enhance clarity.

[0039] Through the steps of probabilistic map region segmentation, obstacle distribution feature calculation, obstacle avoidance response template matching, and dynamic adjustment of sensor perception parameters, accurate positioning of high-risk areas and optimized allocation of perception resources are achieved, providing accurate monitoring data support for the subsequent verification and execution of obstacle avoidance strategies.

[0040] Furthermore, the method provided in this application embodiment includes:

[0041] The preset obstacle avoidance response templates include: a basic classification template, a multi-obstacle avoidance overlap template, an obstacle motion posture classification template, and a sudden vehicle trajectory node overlap matching template.

[0042] Specifically, the preset obstacle avoidance response template library includes four types of differentiated obstacle avoidance response templates: basic hierarchical templates, multi-obstacle avoidance overlapping templates, obstacle motion posture classification templates, and sudden vehicle trajectory node overlap matching templates. The basic hierarchical templates are designed for scenarios with different risk levels of a single obstacle and serve as the basic logical unit for all obstacle avoidance strategies. The obstacle motion posture classification templates are designed for different obstacle motion characteristics to adapt to the complex motion changes of dynamic obstacles. The multi-obstacle avoidance overlapping templates are used to handle scenarios where multiple high-risk areas overlap spatially, determining avoidance priorities through joint risk assessment and generating cascaded obstacle avoidance trajectories. The sudden vehicle trajectory node overlap matching templates are used to handle scenarios where high-speed sudden vehicles approach, determining the conflict type through spatiotemporal node matching and executing corresponding emergency avoidance actions. The detailed implementation processes of the latter two types of templates will be explained in subsequent sections.

[0043] The basic grading template is divided into three sub-templates based on obstacle occupancy probability and relative distance, each corresponding to a different handling strategy. The two judgment conditions are logically ANDed; when the two conditions conflict, the obstacle occupancy probability condition takes precedence. Template 1 corresponds to low-probability long-distance scenarios, with judgment conditions of obstacle occupancy probability less than 0.3 and relative distance greater than 10 meters, executing a smooth deceleration and path fine-tuning strategy. Smooth deceleration uses linear deceleration control, with a deceleration not exceeding 0.1g. Path fine-tuning is achieved by generating a cubic polynomial lateral offset trajectory, with a lateral offset not exceeding 0.3 meters. The boundary conditions for the cubic polynomial lateral offset trajectory are: the starting point is the vehicle's current position, with lateral velocity and lateral acceleration of 0; the ending point is the preset offset target position, with lateral velocity and lateral acceleration of 0. During trajectory generation, the acceleration is continuously changed to avoid vehicle impact.

[0044] Template 2 corresponds to medium-probability, medium-distance scenarios. The criteria are an obstacle occupancy probability between 0.3 and 0.6 and a relative distance between 5 and 10 meters. Active lane changing or local replanning strategies are implemented. Deceleration is controlled between 0.1g and 0.3g, and lateral offset is controlled between 0.3 meters and 1.0 meter. Local replanning starts from the current path node and generates the shortest collision-free trajectory within the available space. Template 3 corresponds to high-probability, short-distance scenarios. The criteria are an obstacle occupancy probability greater than 0.6 and a relative distance less than 5 meters. Emergency braking or a preset avoidance trajectory strategy is implemented. Deceleration is no less than 0.5g. During emergency braking, the steering wheel remains straight. The preset avoidance trajectory allows temporary use of adjacent lanes, with the trajectory endpoint set at a position more than 2 meters away from the obstacle within the adjacent lane.

[0045] The obstacle motion posture classification template extracts the motion posture features of each obstacle through event flow. These features include motion pattern and angular velocity change rate. The angular velocity change rate is calculated by averaging the differences in the obstacle's angular velocity across 10 consecutive frames of detection data. The motion pattern is determined based on the continuity of the obstacle's position change. Based on these motion posture features, obstacles are categorized into three types: deterministic motion, intermittent motion, and non-cooperative motion. The criteria for deterministic motion are an angular velocity change rate of less than 5 degrees per second and a continuous and stable motion pattern. It matches either template one or template two in the basic classification template, employs a smooth detour trajectory, and maintains a minimum safe distance of 1.5 meters between the detour trajectory and the obstacle.

[0046] Intermittent motion is defined as an angular velocity change rate between 5 and 20 degrees per second with a clear start-stop interval. A two-stage observation-decision template is applied, initially maintaining the current path for 0.5 seconds while doubling the monitoring sampling rate. After the observation period, if the obstacle remains stationary or its direction of movement is perpendicular to the autonomous vehicle's direction and shows no intersection tendency, a low-speed detour is initiated, with a speed not exceeding 5 km / h. If the obstacle begins to move towards the autonomous vehicle's path or enters within 5 meters in front of the vehicle, a stop and wait is initiated. Non-cooperative motion is defined as an angular velocity change rate greater than 20 degrees per second with no obvious pattern in the trajectory. Template three in the basic grading template is forcibly applied, abandoning the original path tracking and prioritizing stopping. After stopping, the vehicle remains in braking mode until the obstacle leaves the monitoring range.

[0047] By constructing a multi-layered obstacle avoidance response template system that includes basic classification and motion posture classification, differentiated and precise handling of obstacles with different risk levels and motion characteristics can be achieved, thereby improving the safety and smoothness of the obstacle avoidance process of unmanned vehicles.

[0048] Furthermore, the method provided in this application embodiment includes:

[0049] When two or more high-probability areas are detected to overlap in space, a cascaded avoidance strategy is dynamically generated from the basic hierarchical template. The cascaded avoidance strategy includes: calculating the joint risk contour line of the overlapping areas, determining the avoidance priority according to the distance from the vehicle and the probability peak; prioritizing the execution of the basic template action matched by the high-priority obstacle, and then performing secondary planning on the remaining risk space to generate a composite obstacle avoidance trajectory.

[0050] Specifically, when two or more high-probability regions are detected to overlap spatially, the execution process of the multi-obstacle avoidance overlap template is triggered. The criteria for determining a high-probability region are connected components with an obstacle occupancy probability greater than 0.6. The criteria for determining spatial overlap are that the bounding rectangles of two high-probability regions intersect, or the minimum distance between the two regions is less than 1 meter. First, the grid coordinates and occupancy probability data of all overlapping high-probability regions are extracted to prepare for subsequent joint risk assessment.

[0051] First, the joint risk contour lines for the overlapping areas are calculated. For each grid cell within the overlapping area, its joint risk value is calculated. The joint risk value is equal to the sum of the occupancy probabilities of all overlapping areas at that grid cell location. If the calculated joint risk value is greater than 1, then 1 is taken as the final joint risk value. Three joint risk contour lines are drawn with thresholds of 0.3, 0.6, and 0.9 respectively, dividing the overlapping areas into three levels: low risk, medium risk, and high risk. The joint risk contour lines can intuitively reflect the risk distribution under the combined effect of multiple obstacles, providing spatial constraints for the subsequent generation of avoidance paths.

[0052] Next, the avoidance priority of each obstacle is determined based on its distance from the vehicle and its peak probability. The peak probability of each high-probability area and its relative distance to the vehicle are calculated. A weighted scoring method is used to calculate the priority score, which is equal to the peak probability multiplied by 10 plus 10 divided by the relative distance. A higher score indicates a higher avoidance priority. When the difference in priority scores between two obstacles is less than 0.5, the obstacle closer to the vehicle is given higher priority. All obstacles are sorted from highest to lowest priority score to determine the execution order of the cascading avoidance.

[0053] This embodiment does not employ a simple serial execution method, but instead uses a cascaded avoidance strategy based on dynamic rolling time-domain optimization. The specific steps are as follows:

[0054] First, all overlapping areas are treated as a composite risk field, generating joint risk contour lines. Then, in each control cycle (50 milliseconds), the following dynamic programming is executed: starting from the current vehicle state (including position, speed, and heading), and ending at the next path node of the original path, a local trajectory is generated using model predictive control, considering the predicted trajectories of all obstacles at the current moment, satisfying dynamic constraints and with a minimum distance of 1 meter from all obstacles. Dynamic constraints include lateral acceleration not exceeding 0.3g and deceleration not exceeding 0.5g. The cost function of this trajectory includes avoidance weights for both high-priority and low-priority obstacles, with higher cost coefficients for high-priority obstacles and lower cost coefficients for low-priority obstacles, and a penalty term for deviation from the original path is added.

[0055] If the above model predictive control solution is successful, the generated composite obstacle avoidance trajectory will be executed; if the solution fails, i.e., it is impossible to avoid all obstacles at the same time, the approach will be downgraded to performing emergency braking only on the highest priority obstacle, i.e., template three in the basic classification template, while keeping the steering wheel straight to ensure that no collision occurs.

[0056] Through this rolling temporal optimization, the vehicle can respond in real time to changes in other risk spaces while avoiding high-priority obstacles, without waiting for the process to complete before replanning, thus achieving safe and smooth passage in dynamic multi-obstacle environments.

[0057] Furthermore, the method provided in this application embodiment includes:

[0058] When a sudden vehicle is detected approaching at high speed, the system performs spatiotemporal overlap matching between the vehicle's path nodes and the sudden vehicle's target trajectory nodes, and calculates the estimated arrival time difference for each node. If the conflict node is unique and located far ahead of the vehicle, a turning avoidance template is matched to generate an offset trajectory away from the direction of travel of the sudden vehicle. If there are multiple consecutive conflict nodes that cover the current node, a stop avoidance template is matched to immediately brake to a stop and keep the steering wheel straight. If the conflict node is located behind the vehicle and the estimated arrival time difference continues to decrease, an acceleration avoidance template is matched to accelerate away from the conflict area when conditions permit.

[0059] Specifically, when a sudden vehicle is detected approaching at high speed, the execution process of the sudden vehicle trajectory node overlap matching template is triggered. The criteria for determining a sudden vehicle approaching at high speed are either: a relative speed greater than 30 km / h and a straight-line distance of less than 20 meters from the vehicle, or a relative speed greater than 50 km / h and a straight-line distance of less than 50 meters from the vehicle. First, all path nodes on the vehicle's current planned path are extracted. Each path node contains position coordinates and the vehicle's estimated arrival time. The estimated arrival time is calculated as follows: starting from the vehicle's current position, the travel time between adjacent path nodes is calculated at the current constant speed, and the estimated arrival time of each path node is obtained by summing these values, while also adding the effect of the current acceleration for correction. Simultaneously, the current position, speed, and direction of motion of the sudden vehicle are extracted. A constant-speed motion model is used to predict the trajectory of the sudden vehicle within the next 3 seconds. A trajectory node is generated every 0.5 seconds on the predicted trajectory, and each trajectory node contains position coordinates and the sudden vehicle's estimated arrival time.

[0060] Perform spatiotemporal overlap matching between the vehicle's path nodes and the target trajectory nodes of the sudden vehicle. Traverse all vehicle path nodes and sudden vehicle trajectory nodes, calculating the spatial distance and time difference between the two nodes. When the spatial distance between two nodes is less than 1.5 meters and the absolute value of the time difference is less than 1 second, the node is determined to be a spatiotemporal conflict node. Record the location of all spatiotemporal conflict nodes, the estimated arrival time of the vehicle, and the estimated arrival time of the sudden vehicle. Calculate the estimated arrival time difference for each conflict node, which is equal to the estimated arrival time of the vehicle minus the estimated arrival time of the sudden vehicle.

[0061] The corresponding avoidance template is matched based on the distribution characteristics of the spatiotemporal conflict nodes. If the conflict node is unique and the distance between the node and the current position of the vehicle is greater than 10 meters, a turning avoidance template is matched. The turning avoidance template generates an offset trajectory away from the direction of travel of the sudden vehicle. The rule for determining the direction of departure is: calculate the vertical vector of the direction of movement of the sudden vehicle, and select the vertical direction opposite to the position of the sudden vehicle relative to the vehicle as the direction of departure. If the sudden vehicle is approaching laterally from the left side of the vehicle, the direction of departure is the right side; if it is approaching laterally from the right side, the direction of departure is the left side; if it is approaching from behind in the same lane, the side with fewer obstacles in the two lanes is selected as the direction of departure. The offset trajectory is generated using a cubic polynomial, with the lateral offset controlled between 1.0 meter and 1.5 meters, the deceleration controlled between 0.3g and 0.5g, and the trajectory endpoint set at a safe position 2 meters behind the conflict node.

[0062] If there are three or more consecutive conflict nodes, including one of the nodes on the vehicle's current path, a stop-and-avoidance template will be matched. The stop-and-avoidance template will immediately execute an emergency braking operation with a deceleration of at least 0.8g. During braking, the steering wheel will remain straight, and no lateral steering will be performed until the vehicle comes to a complete stop. After the vehicle comes to a stop, the braking state will be maintained until the vehicle leaves the monitoring range and there are no other potential risks.

[0063] If the conflict node is located behind the vehicle's current position and the estimated arrival time difference continuously decreases over three consecutive calculation cycles, an acceleration avoidance template is applied. The acceleration avoidance template first determines whether current road conditions permit acceleration. The criteria are: no other obstacles within 50 meters in front of the vehicle, a road curvature of less than 0.01 radians per meter within 50 meters in front of the vehicle, and a continuous straight section of at least 60 meters. If these conditions are met, acceleration is executed, with the acceleration controlled between 2 m / s² and 3 m / s², accelerating until the relative distance to the conflicting vehicle is greater than 30 meters, after which normal driving speed is resumed. If road conditions do not permit acceleration, the vehicle automatically downgrades to a stop avoidance template.

[0064] By matching the spatiotemporal nodes of autonomous vehicles with those of sudden vehicles and executing differentiated avoidance templates, the system achieves rapid and accurate response in high-speed sudden vehicle scenarios, effectively reducing the risk of cross-collision and rear-end collisions and improving the emergency response capabilities of autonomous vehicles in complex traffic environments.

[0065] Furthermore, the method provided in this application embodiment includes:

[0066] Path factors are extracted from the current monitoring data, including collision time, relative velocity change rate, obstacle movement direction change rate, and measured occupancy probability. The path factors are compared with the preset trigger thresholds in each obstacle avoidance path strategy to calculate the matching degree. If there is a unique match, the corresponding path strategy is activated directly. If there are multiple matches, the path strategy with the highest priority is activated according to priority. If there is no match, the current path strategy is maintained. Based on the activated path strategy, a corresponding autonomous obstacle avoidance control command is generated, which includes the desired deceleration and the desired lateral offset.

[0067] In one embodiment, the step of determining the obstacle avoidance parsing path strategy based on obstacle distribution probability map monitoring and matching, and the step of activating the path parsing strategy based on path factor matching according to current monitoring data, are sequentially linked, as follows:

[0068] The first step, the initial screening stage, involves generating one or more candidate obstacle avoidance path strategies based on the obstacle distribution probability map, through region segmentation, decay gradient calculation, and matching with a preset template. Each candidate strategy includes a strategy type, key parameters, and a corresponding preset trigger threshold vector. This threshold vector consists of a collision time threshold, a relative velocity change rate threshold, an obstacle movement direction change rate threshold, and an occupancy probability threshold. If only one candidate strategy is generated, it is sent to the next step as the strategy to be activated; if multiple candidate strategies are generated, all are retained and sent to the next step.

[0069] The second step, the precise authentication stage: After acquiring the current monitoring data, real-time path factors are extracted, including collision time, relative velocity change rate, obstacle movement direction change rate, and measured occupancy probability. The actual values ​​are then matched with the preset trigger threshold vectors of each candidate strategy obtained in the first step using a weighted Euclidean distance to calculate the matching degree. The matching degree is calculated as follows: For each path factor, the absolute value of the difference between the measured value and the corresponding threshold is divided by the threshold to obtain the normalized deviation; this deviation is then multiplied by weight coefficients: collision time has a weight of 0.4, measured occupancy probability has a weight of 0.3, relative velocity change rate has a weight of 0.2, and obstacle movement direction change rate has a weight of 0.1; the products are then summed to obtain the matching degree score, with a lower score indicating a better match.

[0070] Then, the matching degree is used to activate the strategy: if only one candidate strategy has a matching degree of less than 0.2, then the strategy is activated directly; if multiple candidate strategies have a matching degree of less than 0.2, then the strategy with the highest priority is activated according to the following priority: the sudden vehicle trajectory node overlap matching template has the highest priority, followed by the multi-obstacle avoidance overlap template, then the obstacle motion posture classification template, and finally the basic classification template. Among them, the sub-templates within the basic classification template have the following priority: template 3 is higher than template 2 is higher than template 1; if the matching degree of all candidate strategies is greater than or equal to 0.2, then the current path strategy is maintained, that is, no new obstacle avoidance strategy is activated, and the original plan is followed.

[0071] This cascading approach utilizes the global risk prediction capabilities of probabilistic graphs and dynamically verifies the prediction results using real-time data, ensuring that the final activated strategy aligns with both the environmental situation and the current measured risks.

[0072] Next, corresponding autonomous obstacle avoidance control commands are generated based on the activated path resolution strategy. These control commands include the desired deceleration and the desired lateral offset, generated as follows:

[0073] If the activated strategy is Template 1 in the basic grading template, the expected deceleration is dynamically calculated based on the current collision time: when the collision time is greater than 3 seconds, the expected deceleration is 0.05g; when the collision time is between 1.5 seconds and 3 seconds, the expected deceleration is 0.1g; and when the collision time is less than 1.5 seconds, the expected deceleration is 0.15g. The expected lateral offset is determined based on the lateral position of the obstacle: when the obstacle center deviates from the vehicle's centerline by less than 0.5 meters, the offset is 0.3 meters; when the deviation is between 0.5 meters and 1 meter, the offset is 0.2 meters; and when the deviation is greater than 1 meter, the offset is 0.1 meters.

[0074] If the activated strategy is Template 2 in the basic hierarchical template, the expected deceleration is between 0.2g and 0.3g, with the specific value inversely proportional to the relative distance: the smaller the relative distance, the greater the deceleration. The expected lateral offset is between 0.5m and 1.0m, with the specific value directly proportional to the obstacle occupancy probability: the higher the occupancy probability, the greater the lateral offset.

[0075] If the activated strategy is template three in the basic grading template, the desired deceleration is between 0.5g and 0.8g, with the upper limit of 0.8g when the relative distance is less than 2 meters; the desired lateral offset is 0, that is, keeping the steering wheel straight and only performing emergency braking.

[0076] If the activated strategy is a multi-obstacle avoidance overlapping template, the expected deceleration and expected lateral offset are obtained by model predictive control. The expected deceleration ranges from 0.1g to 0.5g, and the expected lateral offset ranges from 0.3m to 1.2m. The specific values ​​are automatically determined by the optimization algorithm based on the joint risk contour lines.

[0077] If the activated strategy is the observation-decision two-stage template in the obstacle motion posture classification template, the expected deceleration is 0.1g and the expected lateral offset is 0 during the observation stage; if low-speed detour is chosen during the decision stage, the expected deceleration is 0.2g and the expected lateral offset is 0.8m; if stop and wait is chosen, the expected deceleration is 0.4g and the expected lateral offset is 0.

[0078] If the activated strategy is the turning avoidance template in the sudden vehicle trajectory node overlap matching template, the expected deceleration is 0.3g to 0.5g, and the expected lateral offset is 1.0m to 1.5m, with the specific values ​​being proportional to the relative speed of the sudden vehicle; if the stop avoidance template is activated, the expected deceleration is 0.8g, and the expected lateral offset is 0; if the acceleration avoidance template is activated, the expected deceleration is a negative value, i.e., acceleration, which is 2m / s² to 3m / s², and the expected lateral offset is 0.

[0079] After generating the autonomous obstacle avoidance control command, an effectiveness verification is required: if the desired deceleration exceeds the vehicle's maximum braking capacity, it is limited to the vehicle's maximum braking capacity value; if the desired lateral offset exceeds the lateral offset limit corresponding to the vehicle's maximum steering capacity, it is limited to the limit value, and the deceleration is increased accordingly to ensure safety.

[0080] In summary, the path planning-based autonomous obstacle avoidance method for unmanned vehicles provided in this application has the following technical effects:

[0081] This application constructs an obstacle distribution probability map by dividing the neighborhood of path nodes, and generates an obstacle avoidance parsing path strategy through region segmentation, feature operation, and template matching. Multiple path factors are extracted to complete threshold comparison and matching degree calculation. Corresponding strategies are activated hierarchically according to rules and control commands are output. This approach can adapt to complex scenarios such as multiple obstacles and sudden vehicle encounters, making autonomous vehicle obstacle avoidance judgment more accurate, effectively improving driving safety and operational stability. It achieves the technical effect of enhancing the vehicle's obstacle avoidance capability in complex environments, ensuring driving safety, and adhering to the preset driving route.

[0082] Example 2, as Figure 2 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides an autonomous obstacle avoidance system for unmanned vehicles based on path planning. The system includes:

[0083] Obstacle distribution probability map construction module 1 is used to analyze obstacles at path nodes according to path planning and establish an obstacle distribution probability map for each path node.

[0084] The monitoring target range determination module 2 determines the monitoring target range and obstacle avoidance analysis path strategy based on the obstacle distribution probability map.

[0085] The autonomous obstacle avoidance control command generation module 3 is used to obtain current monitoring data based on the monitoring target range, perform path factor matching based on the current monitoring data and the obstacle avoidance parsing path strategy, activate the path parsing strategy, and generate autonomous obstacle avoidance control commands.

[0086] Furthermore, the obstacle distribution probability map construction module 1 is used to perform the following steps:

[0087] Based on the path nodes of the path planning, the space within the neighborhood of each path node is divided into several grids; based on the prior information of the static map and the kinematic model of the dynamic obstacles, the probability of each grid being occupied within a set time window is calculated to obtain the probability distribution; based on the probability distribution, sub-probability maps are established for different categories of dynamic obstacles, and a comprehensive obstacle distribution probability map is obtained through weighted fusion.

[0088] Furthermore, the monitoring target range determination module 2 is used to perform the following steps:

[0089] The obstacle distribution probability map is segmented into regions, and several high-probability regions with probability values ​​exceeding a first threshold are extracted as candidate monitoring target ranges. The center position, peak probability, and probability decay gradient with distance of the obstacle distribution within each candidate monitoring target range are calculated. Based on the current motion state of the autonomous vehicle, the decay gradient is matched with several preset obstacle avoidance response templates to determine the response template with the highest matching degree. Based on the response template with the highest matching degree, a corresponding obstacle avoidance parsing path strategy is generated, which includes strategy type and its key parameters. Simultaneously, sub-regions with probability values ​​exceeding a second threshold within the candidate monitoring target range are locked as the final monitoring target range, and the sensor perception parameters are adjusted to focus on the corresponding range, wherein the second threshold is ≥ the first threshold.

[0090] Furthermore, the monitoring target range determination module 2 is used to perform the following steps:

[0091] The preset obstacle avoidance response templates include: a basic classification template, a multi-obstacle avoidance overlap template, an obstacle motion posture classification template, and a sudden vehicle trajectory node overlap matching template.

[0092] Furthermore, the monitoring target range determination module 2 is used to perform the following steps:

[0093] When two or more high-probability areas are detected to overlap in space, a cascaded avoidance strategy is dynamically generated from the basic hierarchical template. The cascaded avoidance strategy includes: calculating the joint risk contour line of the overlapping areas, determining the avoidance priority according to the distance from the vehicle and the probability peak; prioritizing the execution of the basic template action matched by the high-priority obstacle, and then performing secondary planning on the remaining risk space to generate a composite obstacle avoidance trajectory.

[0094] Furthermore, the monitoring target range determination module 2 is used to perform the following steps:

[0095] When a sudden vehicle is detected approaching at high speed, the system performs spatiotemporal overlap matching between the vehicle's path nodes and the sudden vehicle's target trajectory nodes, and calculates the estimated arrival time difference for each node. If the conflict node is unique and located far ahead of the vehicle, a turning avoidance template is matched to generate an offset trajectory away from the direction of travel of the sudden vehicle. If there are multiple consecutive conflict nodes that cover the current node, a stop avoidance template is matched to immediately brake to a stop and keep the steering wheel straight. If the conflict node is located behind the vehicle and the estimated arrival time difference continues to decrease, an acceleration avoidance template is matched to accelerate away from the conflict area when conditions permit.

[0096] Furthermore, the autonomous obstacle avoidance control command generation module 3 is used to perform the following steps:

[0097] Path factors are extracted from the current monitoring data, including collision time, relative velocity change rate, obstacle movement direction change rate, and measured occupancy probability. The path factors are compared with the preset trigger thresholds in each obstacle avoidance path strategy to calculate the matching degree. If there is a unique match, the corresponding path strategy is activated directly. If there are multiple matches, the path strategy with the highest priority is activated according to priority. If there is no match, the current path strategy is maintained. Based on the activated path strategy, a corresponding autonomous obstacle avoidance control command is generated, which includes the desired deceleration and the desired lateral offset.

[0098] The path planning-based autonomous obstacle avoidance system for unmanned vehicles provided in the embodiments of the present invention can execute the path planning-based autonomous obstacle avoidance method for unmanned vehicles provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0099] In embodiment three, this application also provides a computer-readable storage medium that can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the path planning-based autonomous obstacle avoidance system for unmanned vehicles in the embodiments of this application, thereby realizing the above-mentioned path planning-based autonomous obstacle avoidance method for unmanned vehicles.

[0100] It should be understood that the embodiments disclosed in this application and the above description enable those skilled in the art to implement this application. However, this application is not limited to the embodiments mentioned above. It should be understood that those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this invention, and should all be included within the protection scope of this application.

Claims

1. An autonomous obstacle avoidance method for unmanned vehicles based on path planning, characterized in that, include: Based on the path planning, the obstacles at the path nodes are analyzed, and a probability distribution map of obstacles at each path node is established. Based on the obstacle distribution probability map, monitoring and matching are performed to determine the monitoring target range and obstacle avoidance analysis path strategy; The current monitoring data is obtained based on the monitoring target range. The path factor is matched with the obstacle avoidance analysis path strategy based on the current monitoring data, the path analysis strategy is activated, and an autonomous obstacle avoidance control command is generated.

2. The autonomous obstacle avoidance method for unmanned vehicles based on path planning according to claim 1, characterized in that, Based on the path planning, obstacles at each path node are analyzed, and an obstacle distribution probability map is established for each path node, including: Based on the path nodes in the path planning, the space within the neighborhood of each path node is divided into several grids. Based on the prior information of the static map and the kinematic model of the dynamic obstacles, the probability of each grid being occupied within a set time window is calculated to obtain the probability distribution; Based on the probability distribution, sub-probability maps are established for different categories of dynamic obstacles, and a comprehensive obstacle distribution probability map is obtained by weighted fusion.

3. The autonomous obstacle avoidance method for unmanned vehicles based on path planning according to claim 1, characterized in that, Based on the obstacle distribution probability map, monitoring and matching are performed to determine the monitoring target range and obstacle avoidance analysis path strategy, including: The obstacle distribution probability map is segmented into regions, and several high-probability regions with probability values ​​exceeding a first threshold are extracted as candidate monitoring target ranges. Calculate the center position, peak probability, and probability decay gradient of the obstacle distribution within the range of each candidate monitoring target. Based on the current motion state of the unmanned vehicle, match the decay gradient with several preset obstacle avoidance response templates to determine the response template with the highest matching degree. Based on the response template with the highest matching degree, a corresponding obstacle avoidance parsing path strategy is generated. The obstacle avoidance parsing path strategy includes the strategy type and its key parameters. Simultaneously, the sub-regions in the candidate monitoring target range whose probability values ​​exceed the second threshold are locked as the final monitoring target range, and the sensor sensing parameters are adjusted to focus on the corresponding range, wherein the second threshold is greater than or equal to the first threshold.

4. The autonomous obstacle avoidance method for unmanned vehicles based on path planning according to claim 3, characterized in that, The preset obstacle avoidance response templates include: a basic classification template, a multi-obstacle avoidance overlap template, an obstacle motion posture classification template, and a sudden vehicle trajectory node overlap matching template.

5. The autonomous obstacle avoidance method for unmanned vehicles based on path planning according to claim 4, characterized in that, The multi-obstacle avoidance overlapping template includes: When two or more high-probability regions are detected to overlap in space, a cascaded avoidance strategy is dynamically generated from the basic hierarchical template. The cascaded avoidance strategy includes: calculating the joint risk contour line of the overlapping regions and determining the avoidance priority according to the distance from the vehicle and the probability peak. Prioritize the execution of basic template actions matched with high-priority obstacles, and then perform secondary planning on the remaining risk space to generate composite obstacle avoidance trajectories.

6. The autonomous obstacle avoidance method for unmanned vehicles based on path planning according to claim 4, characterized in that, The sudden vehicle trajectory node overlap matching template includes: When a sudden vehicle is detected approaching at high speed, the spatiotemporal overlap matching of the vehicle's path nodes and the sudden vehicle's target trajectory nodes is performed, and the estimated arrival time difference of each node is calculated. If the conflict node is unique and located far in front of the vehicle, then the turning avoidance template is matched to generate an offset trajectory that is far away from the direction of travel of the sudden vehicle. If there are multiple consecutive conflicting nodes that cover the current node, then match the stop avoidance template, immediately brake to a stop and keep the steering wheel straight; If the conflict node is located behind the vehicle and the expected arrival time difference continues to decrease, then the acceleration avoidance template is matched, and acceleration is performed to leave the conflict area when conditions permit.

7. The autonomous obstacle avoidance method for unmanned vehicles based on path planning according to claim 1, characterized in that, Based on the current monitoring data and the obstacle avoidance path analysis strategy, path factors are matched, the path analysis strategy is activated, and autonomous obstacle avoidance control commands are generated, including: Extract path factors from the current monitoring data, including collision time, relative velocity change rate, obstacle movement direction change rate, and measured occupancy probability; The path factor is compared with the preset trigger threshold in each obstacle avoidance parsing path strategy to calculate the matching degree; If there is a unique match, the corresponding path strategy is activated directly; if there are multiple matches, the path strategy corresponding to the highest priority is activated according to priority; if there is no match, the current path strategy is maintained. The corresponding autonomous obstacle avoidance control command is generated based on the activated path resolution strategy. The autonomous obstacle avoidance control command includes the desired deceleration and the desired lateral offset.

8. An autonomous obstacle avoidance system for unmanned vehicles based on path planning, characterized in that: The system is used to implement the path planning-based autonomous obstacle avoidance method for unmanned vehicles according to any one of claims 1-7, the system comprising: The obstacle distribution probability map construction module is used to analyze obstacles at path nodes based on path planning and establish an obstacle distribution probability map for each path node. The target range determination module performs monitoring matching based on the obstacle distribution probability map to determine the target range and obstacle avoidance analysis path strategy. The autonomous obstacle avoidance control command generation module is used to obtain current monitoring data based on the monitoring target range, perform path factor matching based on the current monitoring data and the obstacle avoidance parsing path strategy, activate the path parsing strategy, and generate autonomous obstacle avoidance control commands.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the path planning-based autonomous obstacle avoidance method for unmanned vehicles as described in any one of claims 1 to 7.

Images