Path planning method and device for memory parking

Abstract

The invention discloses a path planning method and device for memory parking. The method comprises the steps that environment perception information of a vehicle in the memory parking process is acquired; according to the current position of the vehicle, a hazard source in a preset range is extracted from the environment perception information, and the hazard source comprises at least one of a road boundary and an obstacle; offset processing is carried out on the dangerous source boundary in the direction of increasing the safe driving margin, so that noise is added to the environment perception information; and performing path planning on the vehicle according to the environment perception information subjected to noise processing. Before behavior decision making, noise is added to environment perception information in advance, and an active and preventive security strategy is provided. By actively and directionally introducing uncertainty, a more harsh virtual environment model is constructed, so that a path planning algorithm is guided to generate a track with a larger safety margin, and a more conservative, safer and more stable decision is made.

Description

Technical Field

[0001] This application relates to the field of path planning technology, and in particular to a path planning method and device for memory parking. Background Technology

[0002] Memory parking is an advanced driver assistance technology that assists the driver in completing parking operations. The driver participates in each step from inside the vehicle, without needing to search for a parking space or park manually. The memory parking function can include sub-functions such as memory mapping, intelligent parking, along-route parking space recognition, intelligent parking exit, fault handling during operation, dynamic and static obstacle handling, and video monitoring. Its core is to use parking or some driving sensors to create a local map for memory parking and establish route maps for intelligent parking and intelligent parking exit. After the vehicle reaches a flat surface, the user sets a learning route and completes the learning process. The vehicle can then compare this route with previously stored environmental images to determine if the route is correct.

[0003] In the memory-based parking function, the decision-making system and path planning are the core components for achieving autonomous driving and parking. The former determines what to do, while the latter plans how to do it. Together, they ensure the safe and efficient reproduction of the memorized route and completion of parking in complex parking environments. The decision-making system dynamically judges the vehicle's driving behavior (such as acceleration, braking, steering, and avoidance) by combining pre-memorized data, real-time environmental perception results, and vehicle information. Its core objective is to respond to environmental changes and ensure safety while following the memorized route. Path planning, guided by the decision-making system, generates a specific driving trajectory (including position, direction, and speed curve) for the vehicle. Its core objective is to allow the vehicle to safely and smoothly reach the target parking space from its current position, as close as possible to the memorized route.

[0004] However, perception systems that collect environmental perception information have inherent errors, such as positioning drift and recognition box jitter. If decision-making and path planning are based on raw, uncertain environmental perception information, extreme behaviors may occur, such as driving close to boundaries or obstacles. The fault tolerance is extremely low, and even a slight perception error may lead to a collision. Summary of the Invention

[0005] To address the aforementioned issues, this application proposes a path planning method for memory-based parking, the method comprising: Acquire environmental perception information of the vehicle during the memory parking process; Based on the vehicle's current location, extract potential hazards within a predetermined range from environmental perception information. These potential hazards include at least one of road boundaries and obstacles. The boundary of the hazard source is offset in a direction that increases the margin for safe driving, thereby adding noise to the environmental perception information; Based on the environmental perception information from noise processing, the vehicle performs path planning.

[0006] In one example, the process of shifting the hazard information in a direction that increases the safety margin specifically includes: When the hazard source is the road boundary, the road boundary will be shifted towards the center of the road; When the hazard source is an obstacle, shift the obstacle boundary towards the side with the larger drivable area. In one example, the process of planning a path for the vehicle based on environmental perception information processed by noise treatment specifically includes: The vehicle makes behavioral decisions based on multi-source information during the memory parking process; the multi-source information includes the environmental perception information from the noise processing. When the decision result is to activate the driving-restricted planning strategy, a graph search-based planning algorithm performs a forward path search; When at least one feasible path solution is found, evaluate whether the feasible path meets the preset path quality requirements. If the path quality requirements are met, path planning is performed on the vehicle based on the feasible path and the remembered path.

[0007] In one example, the planning algorithm for graph search is a hybrid A algorithm, and the heuristic cost function of the planning algorithm is calculated based on the following parameters: Lateral deviation between the extension point and the memory path, Euclidean distance between the extension point and the target point, and difference between the heading angle of the extension point and the heading angle of the target point.

[0008] In one example, the method further includes: When no feasible path solution is found or the path quality requirements are not met, the graph search-based planning algorithm performs a bidirectional path search in both forward and backward directions. When at least multiple feasible path solutions are found, an optimal feasible path is selected from the multiple feasible path solutions based on a preset optimization evaluation index; If no feasible path solution is found, generate a parking trajectory with a speed of 0 along the previously planned tracking path.

[0009] In one example, the method further includes: When the optimal feasible path includes a reversing operation, the endpoint of the reversing path is set as the waypoint.

[0010] A graph search-based planning algorithm generates a first segmented path from the vehicle's current location to the waypoint, and a second segmented path from the waypoint to the target point. The first segmented path and the second segmented path are concatenated to replace the optimal feasible path.

[0011] In one example, the method further includes: When the decision result is to activate the cruise connection planning strategy, it is determined whether the distance between the end point of the cruise memory path and the target storage location is greater than the preset distance threshold. When the distance exceeds a preset threshold, sampling is performed on the cruise memory path to find the target path point, so that the first trajectory driving direction of the parking path starting from the target path point meets the direction consistency requirement with the trajectory driving direction of the vehicle at the target path point. The target path point is determined as the starting point of the parking path, and the parking path of the vehicle is planned. The cruise memory path or the cruise tracking path planned with the target waypoint as the endpoint is spliced ​​with the parking path.

[0012] In one example, sampling along the cruise memory path to find the target waypoint specifically includes: Using the first sampling distance from the target storage location as the initial sampling value, sample the first path point on the cruise memory path; Determine whether the first trajectory direction of the parking path starting from the first path point meets the direction consistency requirement with the trajectory direction of the vehicle at the first path point. If the condition is not met, the second path point is resampled at a second sampling distance greater than the first sampling distance; Repeat the direction determination until a target path point that meets the direction consistency requirement is found or the number of samplings exceeds a preset threshold.

[0013] In one example, the method further includes: When the distance is less than or equal to a preset distance threshold, a first cruise tracking path is planned from the current position of the vehicle to the path point at the initial sampling distance. Sampling is performed on the first cruise tracking path to find the target path point. or, Plan a second cruise tracking path from the vehicle's current position to a safe starting point, wherein the safe starting point is located in front of the target storage location and maintains a safe distance from the storage location; Using the aforementioned safe starting point as the starting or ending point of the parking path, a parking path is planned and then spliced ​​with the second cruise tracking path.

[0014] On the other hand, embodiments of this application provide a path planning device for memory parking, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a path planning method for memory parking as described above.

[0015] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects: Adding noise to environmental perception information before behavioral decisions provides a proactive, preventative safety strategy. By actively and purposefully introducing uncertainty, a more stringent virtual environment model is constructed, guiding the path planning algorithm to generate trajectories with greater safety margins, thus leading to more conservative, safer, and more stable decisions. Ultimately, this reduces decision-making errors and prevents decisions from being altered by minor fluctuations in environmental perception information, improving the accuracy and smoothness of decision-making. Especially in extreme scenarios such as narrow bends or special scenarios involving patrol and connection, it improves the success rate of path planning while ensuring safety.

[0016] During the parking process, the autonomous driving system can intelligently select the most appropriate planning strategy for path planning based on real-time changes in the environment. This transforms the traditional reactive mode of sensing collision risks, emergency stopping, and passively escaping from trouble into a proactive proactive mode of anticipating infeasible paths, actively planning, and smoothly passing through, thereby fundamentally improving the smoothness of passage and user safety.

[0017] Specifically, by using noise-processed environmental perception information, the data quality on which decision-making relies is higher, resulting in more accurate and reliable behavioral decisions (such as determining whether to activate the driving-restricted planning strategy), reducing false triggers caused by perception errors (such as unnecessary emergency braking or planning switching), and improving the system's stable performance in complex and uncertain environments.

[0018] Furthermore, the activation condition is a forward-looking judgment. Before the vehicle gets stuck, it's identified that continuing along the current tracking path is likely to cause problems. Based on graph search, a new path is calculated in advance that fully considers the current environment and avoids all potential risks. For example, a turning path that meets curvature requirements is planned.

[0019] Furthermore, the step of assessing whether feasible paths meet the preset path quality requirements ensures that the newly planned path is not only passable, but also of high quality, such as being smooth (continuous curvature) and maintaining a sufficient safe distance from obstacles, thereby greatly reducing the probability of the vehicle having another accident during execution.

[0020] In summary, the vehicle can adjust its posture in advance to smoothly and easily pass through narrow or curved areas. The entire process is seamless, without any noticeable jerking or hesitation.

[0021] In summary, the initial distance threshold judgment effectively prevents incorrect planning in dangerous or narrow environments, improving the overall safety of the system. Before the vehicle reaches the parking space, sampling is performed on the cruise memory path to find the target path point. This ensures that the first trajectory direction of the parking path starting from the target path point matches the vehicle's trajectory direction at the target path point. By guaranteeing directional consistency and path smoothness, a complete path seamlessly transitioning from the current cruise state to the parking state is planned in advance, smoothly, and safely. This achieves a seamless transition between cruise and parking, greatly improving comfort, reducing the number of adjustments required for parking, shortening overall parking time, and increasing parking efficiency. Attached Figure Description

[0022] To more clearly illustrate the technical solution of this application, some embodiments of this application will be described in detail below with reference to the accompanying drawings, in which: Figure 1 A flowchart illustrating a path planning method for memory parking provided in an embodiment of this application; Figure 2 A flowchart illustrating a path planning method in a given scenario, as provided in an embodiment of this application; Figure 3 A flowchart illustrating another path planning method in a different scenario provided by an embodiment of this application; Figure 4 This is a schematic diagram of a path planning device for memory parking provided in an embodiment of this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] Some embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0025] Figure 1 A flowchart illustrating a path planning method for memory parking provided in this application embodiment includes the following steps: S101: Obtain environmental perception information of the vehicle during the memory parking process.

[0026] Environmental perception information can be the real-time detection results of the vehicle's environmental perception sensors (such as lidar, cameras, millimeter-wave radar, etc.). Environmental perception information can include obstacle information (vehicles, pedestrians, etc.), road information where the vehicle is located, etc. Obstacle information includes location, speed, type, etc., and road information includes lane lines, road boundaries, etc.

[0027] S102: Based on the vehicle's current location, extract hazard sources within a predetermined range from environmental perception information, wherein the hazard sources include at least one of road boundaries and obstacles.

[0028] The vehicle's current location can be determined by integrating technologies such as GPS, IMU, wheel speedometer, lidar, and cameras to obtain the vehicle's precise coordinates and orientation (heading angle) in a high-precision map.

[0029] In this example, by setting a predetermined range, the focus can be placed on key areas around the vehicle, thereby reducing computational complexity and allowing attention to the relevant environment.

[0030] The predetermined range is the area formed by a first predetermined distance in front of the vehicle and a second predetermined distance behind the vehicle. For example, the predetermined range could be 35 meters in front of the vehicle and 10 meters behind it. The 35 meters in front of the vehicle can cover sufficient reaction and planning distance in low-to-medium speed scenarios, while the 10 meters behind mainly focuses on safety when reversing. The smaller range is because reversing speeds are usually very slow.

[0031] S103: Offset the boundary of the hazard source in a direction that increases the safety margin for driving, so as to add noise to the environmental perception information.

[0032] Typically, data is denoised to improve accuracy. However, the noise addition method described in this application is a proactive and preventative safety strategy. Its core logic is to construct a more stringent virtual environment model by directionally modifying the perception results. That is, by introducing uncertainty proactively and directionally rather than randomly perturbing the data, noise addition leads to more conservative and safer decisions, guiding the path planning algorithm to instinctively generate trajectories with greater safety margins.

[0033] Road boundaries define the limits of the drivable area, and their boundaries are typically represented as lines or curves. Obstacles are objects that occupy drivable space, and their boundaries are typically represented as the smallest bounding rectangle, convex polygon, or a more refined outline polygon surrounding the object.

[0034] Offsetting the boundary of a hazard source refers to performing a geometric transformation on the original hazard source boundary data model to generate a virtual boundary data model for path planning. This virtual boundary, compared to the original boundary, has a changed outline that increases the margin for safe driving. For example, the road may be narrower than it actually is, or the obstacles may be larger.

[0035] Safe driving margin refers to the safe distance or buffer space reserved between a vehicle and a hazard source to cope with various uncertainties and thus prevent collisions.

[0036] For example, uncertainties include: sensor measurements (cameras, radar, lidar) have errors. The detection of an obstacle's position and shape may have an error of 5 centimeters or more. The vehicle's own position judgment also has slight drifts; it may think it's in the center of the lane when it's actually a few centimeters to the left. When the vehicle executes steering, acceleration, and braking commands, it cannot perfectly adhere to the planned ideal path; there will always be slight tracking errors. The system's predictions of surrounding pedestrians are based on probability and may suddenly change direction.

[0037] It's important to note that we typically don't offset every boundary of a hazard. Instead, we selectively offset those boundaries that pose the most direct threat to the vehicle's current driving state. The purpose of this offset is to increase the safety margin; therefore, the offset strategy is intelligent and directional. If we offset every boundary of a hazard, such as expanding the perimeter of an obstacle, the obstacle would become very large in the virtual environment, potentially blocking the road and preventing the path planning algorithm from finding a feasible path. Therefore, we only offset the side most likely to affect vehicle safety.

[0038] In this example, when offsetting the boundary of a hazard source, the offset amount can be determined using any of the following methods or a combination thereof: Random determination methods include uniform distribution and Gaussian distribution.

[0039] A uniform distribution method can be achieved by adding uniform noise, which involves applying a random offset that is uniformly distributed between a preset minimum and maximum value to the offset boundary of the hazard source. The advantage is that it is computationally simple and all offsets have the same probability of occurrence.

[0040] Gaussian distribution can be achieved by adding Gaussian noise, meaning the offsets follow a normal distribution with a preset mean and a predetermined variance, causing most offsets to concentrate around the mean. The advantage is that it better matches the statistical characteristics of many sensor errors, resulting in relatively smooth boundaries.

[0041] Deterministic approaches include a fixed offset, a dynamic offset based on vehicle state, an attribute-related classification offset, and a dynamic offset related to drivable areas.

[0042] Based on a fixed offset, a fixed offset is preset for different types of hazards, and the hazard boundary is shifted by the corresponding fixed offset. The advantages are simple calculation, high operating efficiency, and strong deterministic system behavior.

[0043] Dynamic offset based on vehicle status. For example, speed-related offset: the higher the speed, the greater the offset. At high speeds, a larger safety margin is reserved to cope with longer braking distances. Yaw rate-related offset: when the vehicle is turning (with a large yaw rate), the offset from obstacles on the inside of the curve is increased to compensate for the risks caused by vehicle sway. The advantage is strong adaptability, providing more reasonable safety assurance under different driving conditions.

[0044] Hazard attribute-based classification offsets differentiate offset values ​​based on the hazard type, dynamic attributes, and credibility. For example, dynamic obstacles (such as pedestrians and vehicles) are assigned larger offset values ​​than static obstacles because their behavior is more difficult to predict. The advantages include higher safety, more refined strategies, and effective handling of complex dynamic environments.

[0045] Based on dynamic offsets related to the drivable area, the magnitude of the offset is inversely proportional to the width of the drivable area. The narrower the drivable space, the greater the applied offset. This is because the margin for error is lower in narrow spaces, requiring a more conservative strategy. The advantage is that it can intelligently achieve an optimal balance between passability and safety, avoiding overly conservative approaches in wide scenarios and ensuring safety in narrow scenarios.

[0046] S104: Based on the environmental perception information from noise processing, perform path planning for the vehicle.

[0047] In this example, path planning uses a virtual environment model that has undergone offset processing to search for and optimize vehicle trajectories. All judgments made by the path planning algorithm regarding collision risk and drivable areas are based on this conservative environment with increased safety margins. Therefore, any feasible trajectory generated guarantees a preset safe distance from real hazards.

[0048] It should be noted that, although the embodiments in this application are based on... Figure 1 Steps S101 to S104 will be described sequentially, but this does not mean that steps S101 to S104 must be performed in a strict order. The reason this embodiment follows this order is... Figure 1 The order in which steps S101 to S104 are described is provided to facilitate understanding of the technical solutions of the embodiments of this application by those skilled in the art. In other words, in the embodiments of this application, the order of steps S101 to S104 can be appropriately adjusted according to actual needs.

[0049] pass Figure 1 This method pre-adds noise to environmental perception information before behavioral decisions are made, providing a proactive and preventative safety strategy. By actively and directionally introducing uncertainty, a more stringent virtual environment model is constructed, guiding the path planning algorithm to generate trajectories with a larger safety margin, thus making more conservative, safer, and more stable decisions. Ultimately, this reduces decision-making errors and prevents decisions from being altered by minor fluctuations in environmental perception information, improving the accuracy and smoothness of decision-making. Especially in extreme scenarios such as narrow curves or special scenarios involving patrol and connection, it improves the success rate of path planning while ensuring safety.

[0050] based on Figure 1 In addition to the method described herein, this application also provides some specific implementation schemes and extension schemes of the method, which will be further explained below.

[0051] In one example, offsetting the boundary of a hazard source in a direction that increases the safety margin for driving may include the following steps: When the hazard source is the road boundary, the road boundary is shifted towards the road center. That is, the road becomes narrower. For example, the left boundary is shifted to the right and / or the right boundary is shifted to the left.

[0052] When the hazard source is an obstacle, the obstacle boundary is shifted towards the side with the larger drivable area. In other words, the obstacle becomes larger.

[0053] A larger drivable area indicates greater safety for vehicle passage, and path planning tends to favor this side. Therefore, obstacles are expanded on the drivable side. For example, if an obstacle is in the middle of the road with space on both its left and right sides, but the right lane is wider, then the side with more drivable space is the right side, and the right boundary of the obstacle is simulated to expand to the right.

[0054] Furthermore, the offset can be determined by adding randomness to uniform noise. That is, for road boundaries, noise is added to the right from the left boundary, and noise is added to the left from the right boundary; for obstacles, noise is added to the side with the larger drivable area.

[0055] In summary, adding unilateral uniform noise can not only accurately build a safety buffer zone where it is most needed, but also improve computational efficiency, thus meeting the real-time requirements of autonomous driving systems.

[0056] In one example, particularly in confined driving spaces (narrow roads, wide roads with obstacles) and curves, due to environmental perception errors, vehicle positioning errors, and differences between the human-driven steering boundaries and the autonomous driving steering boundaries when remembering the path (these differences may cause the curvature of the remembered path to be greater than the vehicle's maximum curvature), the vehicle cannot smoothly navigate intersections. Often, the perceived obstacle forces the vehicle to stop, requiring it to replan its escape trajectory before continuing. This not only impacts the user experience but also makes users feel insecure.

[0057] Figure 2 This is a flowchart illustrating a path planning method in a given scenario, provided by an embodiment of this application. Based on this, the process of planning a path for the vehicle according to environmental perception information obtained from noise processing may include the following steps: S201: Make behavioral decisions based on multi-source information during the vehicle's parking process.

[0058] The multi-source information includes environmental perception information for noise processing, memory paths, functional state machine information, and vehicle status information.

[0059] A memory path refers to a pre-generated memory path for a vehicle in an automatic parking scenario, based on the target parking space information.

[0060] Functional state machine information refers to the state of an autonomous driving system in different operating modes, such as waiting to start, cruising, following another vehicle, parking, and completed. Vehicle state information includes geographical location information (latitude and longitude, etc.), speed, acceleration, yaw angle, heading angle, gear position, etc.

[0061] Behavioral decision-making can utilize traditional rule-based decision-making algorithms, or learning-based neural networks or deep learning algorithms, to generate path planning strategies and behavioral assistance instructions activated in the current scenario. In other words, behavioral decision-making outputs a combined decision to guide the vehicle's motion planning and control.

[0062] Path planning strategies are used to generate specific driving trajectories, including cruise planning strategies, driving-constrained planning strategies, parking connection planning strategies, and parking planning strategies. Each path planning strategy is designed for a specific driving scenario.

[0063] Cruise planning strategies are used in scenarios such as open road environments. Parking planning strategies are used to complete the final parking or parking maneuver.

[0064] In this application, a driving-restricted planning strategy is set up for scenarios such as narrow passages or when the curvature of the memorized path exceeds the vehicle's maximum turning capability. A parking transition planning strategy is set up to address the planning problem of the transition phase from driving on the road to the parking area.

[0065] By designing a dedicated path planning strategy for driving-restricted scenarios, the risk of being forced to stop in extreme situations is reduced, and the system's safety redundancy is improved.

[0066] Behavioral assist commands are used to constrain planning objectives, including obstacle avoidance, following another vehicle, pulling over, left turn, right turn, and emergency stop. The obstacle avoidance command generates a trajectory to bypass obstacles ahead. The following vehicle command maintains a safe distance and follows the vehicle in front. The pulling over command locates a safe area on the side of the road and generates a parking trajectory. Left and right turns are executed at intersections. The emergency stop command immediately generates a trajectory for safe deceleration to a stop.

[0067] It should be noted that two independent decision trees can be used to handle path planning strategy selection and behavioral instruction decision-making respectively, which is more modular and interpretable than traditional single-path decision-making. This decoupling can improve maintainability, flexibility, and transparency.

[0068] S202: When the decision result is to activate the driving-restricted planning strategy, a graph search-based planning algorithm is used to perform a forward path search.

[0069] In this example, activation conditions include the drivable space width being less than a preset width threshold and the curvature of the memory path point corresponding to the curve being greater than the vehicle's maximum curvature.

[0070] It should be noted that during the memory path process, the memory path is formed because the driver drives the vehicle. However, during autonomous driving, the maximum curvature of the vehicle is usually smaller than that during non-autonomous driving. Therefore, the curvature of the memory path may be greater than the maximum curvature of the vehicle during parking.

[0071] Graph search planning algorithms discretize complex, continuous environments, abstracting them into a network graph composed of nodes and edges. Nodes represent the possible states of a vehicle, which can be simple (x, y) coordinate positions or more complex states, such as position and heading angle. Edges connect two nodes with actions or path segments. An edge represents the operation performed to transition from one state to another (e.g., moving forward 1 meter, turning 30 degrees to the left, etc.). Each edge has a cost, such as distance, time, energy consumption, or risk.

[0072] Graph search planning algorithms start from the initial node, explore all possible actions (moving along edges), generate new nodes, and continuously evaluate which node is most likely to reach the target node. The search process records the best path and cumulative cost to each node. When the target node is reached, the optimal or feasible path from the starting point to the target node can be obtained by backtracking.

[0073] For example, graph search planning algorithms include breadth-first search, Dijkstra's algorithm, A algorithm, and hybrid A algorithm.

[0074] Hybrid A* algorithm is a high-level algorithm that combines discrete graph search (finding paths in a grid like A* algorithm) with continuous vehicle kinematics models (simulating realistic turning, forward, and backward movements of the vehicle), making it particularly suitable for generating feasible paths in confined spaces. Its nodes are not grid centers, but rather continuous vehicle states, including position and heading. The edges (actions) between nodes are not simple up, down, left, or right movements, but rather continuous, smooth circular paths forward or backward generated through vehicle kinematics models. The heuristic function may combine the search cost of traditional A* algorithm on a discrete grid (considering obstacles) with the distance to the target node to better guide the search.

[0075] It should be noted that the vehicle performs path planning within a preset time interval; that is, it acquires multi-source information about the vehicle within this preset time interval. Furthermore, each planning iteration re-acquires multi-source information and plans a tracking path within a fixed range ahead, striving to closely resemble the memorized path. The target point is also the endpoint of the fixed range corresponding to this planning, for example, 50 meters ahead of the vehicle.

[0076] In other words, the vehicle's automatic planning range is fixed when no range modification is made.

[0077] Forward search refers to the direction of expansion from the vehicle's current position toward the target node. For example, a node expansion method with vehicle kinematic constraints using a hybrid A algorithm can be adopted.

[0078] S203: When at least one feasible path solution is found, evaluate whether the feasible path meets the preset path quality requirements.

[0079] In this example, the search can be exited once a feasible path solution is found. This setting can be adjusted according to actual needs.

[0080] Therefore, when multiple feasible path solutions exist, the optimal solution can be found based on preset forward optimization metrics. For example, forward optimization metrics include the curvature change of the feasible path, the distance between the feasible path and obstacles or road boundaries, etc. For instance, checking whether the curvature of the feasible path changes smoothly (avoiding sudden steering wheel turns). Checking whether the maximum curvature of the path is still within the vehicle's physical limits. Calculating the distance between each point on the feasible path and the nearest obstacle or road boundary to ensure the entire path remains at a safe distance.

[0081] Assessing whether a feasible path meets the preset path quality requirements may include at least one of the following: determining the continuity of the feasible path curvature, whether the feasible path curvature conforms to vehicle kinematic constraints, whether the distance between the feasible path and obstacles is greater than a first safety threshold, and whether the distance between the feasible path and road boundaries is greater than a second safety threshold.

[0082] S204: If the path quality requirements are met, perform path planning for the vehicle based on the feasible path and the remembered path.

[0083] In this example, the generated feasible path may differ from the memory path. If the activation condition reflects a problem where the memory path is clearly infeasible (such as excessive path curvature), the feasible path is determined as the tracking path, allowing the vehicle to travel along the feasible path.

[0084] If the activation condition does not reflect the infeasibility of the remembered path, the optimal solution can be found from the feasible paths and the remembered paths based on the aforementioned preset forward optimization index. Alternatively, path planning can be performed directly on the vehicle based on the feasible paths. However, when the driving-restricted planning strategy is activated, it is usually because the driving environment has changed, and the remembered path has become unsafe or infeasible.

[0085] It's important to note that when initially activating the restricted driving planning strategy, the vehicle may need to perform multiple route planning steps to reach the restricted driving area. However, after these multiple planning steps, the restricted driving area will eventually be included in the planning range. Furthermore, if the restricted area is included in the planning range the first time, it is likely to be included again in subsequent planning steps. Because the planning time intervals are relatively short—for example, planning a tracking path 50 meters ahead of the vehicle at time A and planning the same path at time B, with a one-minute difference between time A and time B—the distance the vehicle can travel within that one minute is typically limited.

[0086] In summary, during the parking memory process, the autonomous driving system can intelligently select the most appropriate planning strategy for path planning based on real-time changes in the environment. This transforms the traditional reactive mode of sensing collision risks, emergency stopping, and passively escaping from difficult situations into a proactive proactive mode that anticipates path infeasibility, pre-plans, and smoothly navigates through the obstacle, thereby fundamentally improving the smoothness of passage and user safety.

[0087] Specifically, by using noise-processed environmental perception information, the data quality on which decision-making relies is higher, resulting in more accurate and reliable behavioral decisions (such as determining whether to activate the driving-restricted planning strategy), reducing false triggers caused by perception errors (such as unnecessary emergency braking or planning switching), and improving the system's stable performance in complex and uncertain environments.

[0088] Furthermore, the activation condition is a forward-looking judgment. Before the vehicle gets stuck, it's identified that continuing along the current tracking path is likely to cause problems. Based on graph search, a new path is calculated in advance that fully considers the current environment and avoids all potential risks. For example, a turning path that meets curvature requirements is planned.

[0089] Furthermore, the step of assessing whether feasible paths meet the preset path quality requirements ensures that the newly planned path is not only passable, but also of high quality, such as being smooth (continuous curvature) and maintaining a sufficient safe distance from obstacles, thereby greatly reducing the probability of the vehicle having another accident during execution.

[0090] In summary, the vehicle can adjust its posture in advance to smoothly and easily pass through narrow or curved areas. The entire process is seamless, without any noticeable jerking or hesitation.

[0091] In one example, since the Hybrid A algorithm is particularly well-suited for generating feasible paths in confined spaces, this application embodiment employs a Hybrid A algorithm planning algorithm that combines discrete graph search with a continuous vehicle kinematics model, aiming to plan a safe (collision-free) and feasible (actually achievable by the vehicle) path in complex environments (such as confined spaces).

[0092] The algorithm does not search for simple grid points, but rather searches for continuous states representing the vehicle's position and heading angle. Starting from each state, instead of moving to adjacent grid cells, it simulates and generates a short segment of possible future trajectories using the vehicle's kinematics model; these are called motion primitives, such as going straight, turning left, turning right, or even reversing. 2. Intelligent Search and Guidance: The algorithm uses a cost function to determine the priority of exploration directions. This function consists of two parts: the actual cost, which evaluates the paths already taken, such as path length, turning and reversing penalties, encouraging the generation of shorter, smoother paths; and the heuristic cost, which estimates the optimal cost from the current state to the goal, guiding the search direction, avoiding blind exploration, and improving efficiency.

[0093] For each generated candidate path segment, rigorous collision detection is performed. The detection target is not the original environment, but an inflated virtual environment model (i.e., with increased safety margin) to ensure that the final planned path maintains a safe distance from real obstacles.

[0094] The initially searched path is composed of many short motion primitives, which is feasible but may not be smooth enough. Therefore, a post-processing optimization step is needed to smooth the path and make it more suitable for the comfort requirements of vehicle driving.

[0095] In summary, the generated paths strictly adhere to the vehicle's motion capabilities, ensuring the vehicle can actually execute them. It excels at escaping difficult situations in narrow and complex environments and finding feasible paths. Through interaction with a virtual safety environment, it ensures that the planning results have inherent safety redundancy.

[0096] Based on this, the planning algorithm for graph search is a hybrid A algorithm. The heuristic cost function of the planning algorithm is calculated based on the following parameters: the lateral deviation between the expansion point and the memory path, the Euclidean distance between the expansion point and the target point, and the difference between the heading angle of the expansion point and the heading angle of the target point.

[0097] The heuristic cost function serves to guide the search direction.

[0098] The lateral deviation between the expanded point and the remembered path measures the horizontal distance between the currently explored point and the remembered path. Guiding the search to be close to the remembered path tends to generate a new path that matches the remembered path as closely as possible. This ensures that vehicle behavior does not deviate too much from the remembered path, improving comfort and predictability. Furthermore, it avoids blindly searching in useless areas far from the remembered path, focusing the search area on a reasonable region surrounding the remembered path.

[0099] The Euclidean distance between the expansion point and the target point is the straight-line distance between them, which is fundamental to ensuring that a path can eventually be found. It essentially gives the algorithm a basic sense of direction, guiding the search process towards the target point. Under certain conditions, it ensures that a solution is eventually found.

[0100] The difference between the heading angle at the expansion point and the heading angle at the target point measures the angular difference between the vehicle's orientation at the expansion point and its expected orientation at the target point. This guides the search process by considering not only where the vehicle should go but also which direction it should be heading. Furthermore, during the search process, the algorithm prioritizes movement methods that allow the vehicle's posture to gradually approach the target posture, resulting in a smoother path that better matches the vehicle's motion characteristics and reduces unnecessary wheel spins.

[0101] Furthermore, the three parameters can be assigned different weights and then summed to form a total heuristic cost.

[0102] In summary, the above heuristic functions enable the efficient planning of a high-quality path that closely resembles the memory path, is safe and smooth (with continuous curvature and reasonable posture), and has a clear objective within a narrow and complex space.

[0103] In one example, if no suitable feasible path is found, the extended search mode is adaptively activated, which can plan the escape route in advance, give the driver advance notice, and improve the driver's sense of security.

[0104] First, a graph search-based planning algorithm is used to perform bidirectional path search in both forward and backward directions.

[0105] Among them, bidirectional search refers to a forward search from the vehicle's current position toward the target point, and a backward search from the target point toward the vehicle's current position.

[0106] In highly complex environments, one-way search may fail by getting stuck in a local minimum, while two-way search can increase the probability of finding a feasible path.

[0107] When at least multiple feasible path solutions are found, an optimal feasible path is selected from the multiple feasible path solutions based on a preset optimization evaluation index.

[0108] It should be noted that the search will not stop upon finding the first feasible path, but will continue iteratively until at least a predetermined number of distinct feasible path solutions are found, at which point the search will exit. For example, at least three feasible path solutions.

[0109] Since the feasible path solution at this time usually includes the escape path that requires reversing, the preset optimization evaluation indicators can include the length of the reversing path, the curvature change of the reversing path, the curvature change of the forward path, and the distance between the feasible path and the obstacle or road boundary.

[0110] For example, the shorter the path, the better. The smoother the path, the less steering wheel movement is required, resulting in a more comfortable ride. The greater the distance between the path and obstacles, the better.

[0111] The optimal method can be to perform a weighted summation of the scores of multiple evaluation indicators, calculate the total score of each feasible path, and select the path with the highest score as the optimal feasible path.

[0112] If no feasible path solution is found, generate a parking trajectory with a speed of 0 along the previously planned tracking path.

[0113] If neither bidirectional search can find a safe path to the target point, the vehicle stops attempting to escape the predicament or reach the target point and instead sets the speed at the end of the previously planned tracking path to 0. The vehicle will then decelerate. If no feasible path solution is found in the next planning attempt, it means no tracking path has been planned, and the same deceleration operation will be performed along the same tracking path.

[0114] In summary, by comparing multiple solutions, and considering that the path to escape a difficult situation requires an optimal balance between safety, comfort, and efficiency, a globally superior path can be selected in advance. When the system realizes that it cannot proceed safely, it brings the vehicle to a safe stop, preventing a collision or further danger, thus improving the system's reliability.

[0115] In one example, if the optimal feasible path involves reversing, segmented planning can improve the one-time pass rate in narrow spaces.

[0116] First, when the optimal feasible path includes a reversing operation, the endpoint of the reversing path is set as the waypoint.

[0117] The optimal feasible path may include a reversing operation. For example, to avoid obstacles or adjust parking, it's necessary to reverse a certain distance before moving forward. Therefore, the point where the reversing operation ends and the vehicle is ready to switch to forward operation is marked as a waypoint. This waypoint becomes a new, temporary task objective.

[0118] Then, a graph search-based planning algorithm is used to generate a first segmented path from the vehicle's current location to the waypoint, and a second segmented path from the waypoint to the target point.

[0119] Among them, a forward path search can be performed based on a graph search planning algorithm. The graph search algorithm can be a hybrid A algorithm, and the heuristic cost function is calculated based on the following parameters: the lateral deviation between the expansion point and the optimal feasible path, the Euclidean distance between the expansion point and the target point, and the difference between the heading angle of the expansion point and the heading angle of the target point.

[0120] Therefore, since the problem to be solved by each path segment is simpler, it is easier to find a very smooth, safe and efficient segmented path, thereby improving the quality of each path segment.

[0121] Finally, the first segmented path and the second segmented path are concatenated to replace the optimal feasible path.

[0122] The second segment path is spliced ​​together with the first segment path, forming a completely new path from the vehicle's current starting point to the target point.

[0123] In summary, the new path obtained through segmentation, optimization, and splicing exhibits higher overall quality (e.g., smoothness, safety, and execution efficiency) than the optimal feasible path obtained through a single, direct search. Because each segment of the new path is superior, and the connections between segments are achieved through reverse and forward movement, the vehicle can complete the entire maneuver very smoothly, significantly improving the success rate of navigating narrow spaces or curves in a single pass.

[0124] For example, when a vehicle is on a right-angle curve, the curvature of the remembered path is greater than the vehicle's maximum curvature.

[0125] The system determines that the curvature of the remembered path exceeds the limit and activates the driving-restricted planning strategy. Hybrid A / B algorithm search: The starting point is the vehicle's current position. The algorithm tries various combinations of motion primitives, and the quality of feasible paths found in the forward search is evaluated (e.g., maximum curvature, distance to obstacles). If the requirements are met, the path is adopted; otherwise, a bidirectional search is performed. When finding the optimal solution bidirectionally, if this optimal path contains obvious reversing points, two high-quality, smooth paths can be generated separately to the waypoint and to the target point, and then spliced ​​together.

[0126] In one example, before the vehicle can autonomously drive to the target parking space, it needs to stop and then switch to parking, or it needs to drive past the parking space and then switch to parking. The trajectory of the memory parking cruise segment and the trajectory of the parking segment cannot be well connected, which affects the user experience.

[0127] Figure 3 A flowchart illustrating another path planning method in a scenario provided by an embodiment of this application is shown below. Based on this, when the decision result is to activate the mooring connection planning strategy, the method includes the following steps: S301: When the decision result is to activate the cruise connection planning strategy, determine whether the distance between the end point of the cruise memory path and the target storage location is greater than the preset distance threshold.

[0128] It should be noted that the vehicle has not yet reached the end of the cruise memory path. However, since the vehicle's planned tracking path is usually close to or the same as the cruise memory path, the end of the cruise memory path is used for comparison.

[0129] The activation conditions for the cruise-and-parking connection planning strategy include: sensing the target parking space and not completing the memory path of the cruise segment. That is, the vehicle has not yet reached the target parking space, so intervention can be carried out in advance when the vehicle is approaching the parking space but has not yet arrived, avoiding replanning after parking.

[0130] The target storage location is typically a rectangular area with four corner points (vertices). Therefore, the distance between the end point of the cruise memory path and the target storage location can be calculated as follows: The direct distance between the endpoint of the cruise memory path and the nearest vertex of the target storage location. That is, the left or right vertex of the storage location entrance.

[0131] Alternatively, it can be the straight-line distance from the end of the cruise memory path to the center point of the target storage location entrance.

[0132] S302: When the distance exceeds a preset threshold, sampling is performed on the cruise memory path to find the target path point, so that the first trajectory driving direction of the parking path starting from the target path point meets the direction consistency requirement with the trajectory driving direction of the vehicle at the target path point.

[0133] If the distance exceeds the preset threshold, it means that the end point of the original planned cruise memory path is in a spacious and safe location, which can serve as an ideal transition zone.

[0134] On the cruise memory path, dynamic sampling is used to find an ideal parking starting point (e.g., a path point 5 meters away from the target parking space). This is iterated through multiple adjustments to the sampling distance (e.g., starting with closer points and then moving further away) until a path point that satisfies the directional consistency requirement is found. When the vehicle reaches this path point on the cruise memory path, the direction of the car's front during parking is naturally consistent with the final direction of the cruise.

[0135] The sampling distance is usually an empirical value, such as 5 meters. The sampling distance can be calculated as follows: the straight-line distance between the sampling path point and the nearest vertex of the target storage location; or the straight-line distance from the sampling path point to the center point of the target storage location's entrance.

[0136] It should be noted that the requirement for directional consistency can be that the first trajectory direction of the parking path is the same as the trajectory direction of the vehicle at the target path point, or that the difference between the heading angles at the two points is less than a preset difference threshold.

[0137] S303: Determine the target path point as the starting point of the parking path and plan the parking path for the vehicle.

[0138] Among them, a parking planning strategy can be activated, and a parking algorithm can be used to plan the parking path from the starting point of the parking path to the target parking space.

[0139] S304: Combine the cruise memory path or the cruise tracking path planned with the target waypoint as the endpoint with the parking path.

[0140] In this example, the cruise tracking path planned with the target waypoint as the endpoint may be the same as or different from the cruise memory path. Therefore, when they are the same, it's equivalent to connecting the cruise memory path and the parking path into a continuous, seamless path; when they are different, it's equivalent to connecting the cruise tracking path and the parking path into a continuous, seamless path—the stitching is not simply a matter of connecting the beginning and end. After stitching, smooth sampling is performed to ensure a uniform distribution of waypoints and continuous curvature changes. This eliminates potential abrupt changes at connection points, enabling the vehicle control system to execute very smooth steering and speed control.

[0141] Based on this, the spliced ​​path is smoothed to ensure that the intervals between path points meet the requirements of the cruise and parking segments. The intervals between path points can be calculated as follows: take the median of the required intervals for the cruise and parking segments.

[0142] It should be noted that speed planning needs to be done in segments before speed smoothing.

[0143] In summary, the initial distance threshold judgment effectively prevents incorrect planning in dangerous or narrow environments, improving the overall safety of the system. Before the vehicle reaches the parking space, by ensuring directional consistency and path smoothness, a complete path is planned in advance, smoothly, and safely to seamlessly transition from the current cruising state to the parking state. This achieves a seamless transition between cruising and parking, greatly improving comfort, reducing the number of adjustments required for parking, shortening overall parking time, and improving parking efficiency.

[0144] In one example, sampling is performed on the cruise memory path to locate the target waypoint, including the following steps: Step 1: Using the first sampling distance from the target storage location as the initial sampling value, sample the first path point on the cruise memory path.

[0145] Step 2: Determine whether the first trajectory direction of the parking path starting from the first path point meets the direction consistency requirement with the trajectory direction of the vehicle at the first path point.

[0146] Step 3: If the condition is not met, resample the second path point at a second sampling distance that is greater than the first sampling distance.

[0147] In general, multiple samples are taken first from the direction closest to the vehicle and farthest from the target storage location.

[0148] Step 4: Repeat the direction determination until a target path point that meets the direction consistency requirement is found or the number of samplings exceeds the preset threshold.

[0149] That is, repeat steps 1-3 above until the directional consistency requirement is met or the number of samplings exceeds the preset threshold, at which point the judgment stops.

[0150] It should be noted that when the number of samplings exceeds a preset threshold, the vehicle will be controlled to continue driving along the current cruise tracking path.

[0151] In one example, if the distance is less than or equal to a preset distance threshold, it indicates that the original path's endpoint may be too close to the storage location. In this case, the system will trigger a more advanced planning algorithm (such as polynomial programming) to regenerate a safe cruise tracking path.

[0152] Based on this, when the distance between the end point of the cruise memory path and the target storage location is less than or equal to a preset distance threshold, the following steps are included: Plan the first cruise tracking path from the vehicle's current position to the path point at the initial sampling distance, sample along the first cruise tracking path, and find the target path point.

[0153] Among these, polynomial programming can be used to determine the first cruise tracking path.

[0154] Polynomial programming is one of the core algorithms for generating smooth, continuous paths, playing a crucial role, especially in planning a second new cruise path or the first cruise tracking path. Its basic idea is to parameterize the path using a polynomial function and solve for a smooth trajectory that conforms to vehicle kinematic constraints by setting boundary conditions. That is, to generate a smooth path connecting the vehicle's current position and the destination. This method ensures that the path is not only continuous but also conforms to the vehicle's motion characteristics by accurately matching the vehicle states at the start and end points.

[0155] Cubic polynomials can guarantee the continuity of the path and its first-order continuity, meaning the tangent direction is continuous. This is suitable for scenarios requiring only the position and heading angle constraints of the start and end points, such as simple lane changes. However, its second derivative of curvature changes linearly, which may lead to curvature discontinuities, requiring instantaneous adjustments to the vehicle's steering wheel, resulting in a less smooth path.

[0156] Fifth-order polynomials are one of the most commonly used choices in autonomous driving. They can simultaneously constrain the positions of the start and end points, the heading angle (first derivative), and the curvature (second derivative). This means that the planned path is not only directionally continuous, but its curvature also changes continuously, resulting in a smooth steering wheel rotation velocity and greatly improving ride comfort.

[0157] Based on this, the starting point (current vehicle position, direction, curvature, and speed) and ending point (expected position, direction, curvature, and speed of the target safe starting point) are clearly defined. Utilizing the property that a fifth-order polynomial can simultaneously satisfy position, direction (first derivative), and curvature (second derivative) constraints, the states of the starting and ending points are transformed into mathematical equations, constructing a system of linear equations. This system of equations is solved using efficient numerical computation methods to obtain a uniquely determined, smooth mathematical path. The generated path is then subjected to collision detection against a virtual environment model that has been inflated (i.e., a safety margin has been added). If the path is unsafe, the ending point state is fine-tuned and the path is replanned until a smooth and safe feasible path is found.

[0158] The path generated by the fifth-order polynomial thus possesses continuously varying curvature, avoiding abrupt steering wheel inputs and resulting in an extremely comfortable driving experience. Solving linear equation systems is a mature and rapid technique in numerical computation, suitable for the real-time planning requirements of autonomous driving. By changing boundary conditions, various path shapes such as straight lines, arcs, and S-curves can be easily planned, making it highly suitable for scenarios requiring precise control of the endpoint's pose, such as cruise transitions. Verification with a virtual environment that has reserved safety space fundamentally ensures path safety.

[0159] When a target waypoint is available, the target waypoint is determined as the starting point of the parking path, the parking path of the vehicle is planned, and the cruise memory path is spliced ​​with the parking path.

[0160] In summary, the system proactively guides the user to plan the vehicle's cruise memory path from locations unsuitable for direct parking. The ultimate goal remains finding a smooth starting point that maintains directional consistency. This means that regardless of the process, the final parking experience for the user remains seamless.

[0161] In one example, when the distance between the end of the cruise memory path and the target storage location is less than or equal to a preset distance threshold, the following steps are included: First, plan a second cruise tracking path from the vehicle's current position to a safe starting point, located in front of the target parking space and maintaining a safe distance from it. This second cruise tracking path can be planned using polynomial programming.

[0162] In this example, the process of determining the safe starting point can be as follows: Sampling is performed in front of the target storage location to find a safe point. This safe point can plan a preset number of berthing entry paths or berthing exit paths. For example, the trajectory of two berths can be adjusted during berthing, and the trajectory of one berth can be adjusted during berthing.

[0163] Then, using the safe starting point as the starting or ending point of the parking path, a parking path is planned and spliced ​​with the second cruise tracking path.

[0164] In summary, we first adopt a reverse approach, starting not from the vehicle's position but from the target parking action to find an ideal safe starting point. A safe starting point must meet a key condition: from that point, a preset number of parking paths for the current vehicle can be successfully planned. Sampling is performed in the area in front of the target parking space, and a parking planning algorithm is used to verify whether each sampled point can plan a trajectory that meets the required number of parking attempts. The point where successful planning is found is the safe starting point. This allows for a smooth connection between the cruising and parking segments, reducing the number of parking attempts, improving parking efficiency, and enhancing the user experience.

[0165] It should be noted that the execution entity of the above path planning method can be the vehicle's own onboard computing platform (or a dedicated parking controller), or a cloud server that performs the Hybrid A algorithm calculation and then sends the generated tracking path to the vehicle for execution. Alternatively, a vehicle-cloud collaborative system can be used, with both the onboard computing platform and the cloud server serving as the execution entities. For example, the cloud performs coarse-grained global path planning, while the vehicle performs fine-grained, real-time local path replanning using the Hybrid A algorithm based on the real-time local environment.

[0166] Based on the same idea, some embodiments of this application also provide devices and non-volatile computer storage media corresponding to the above methods.

[0167] Figure 4A schematic diagram of a path planning device for memory parking provided in this application embodiment includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform a path planning method for memory parking as described above.

[0168] Some embodiments of this application provide a non-volatile computer storage medium for path planning of memory parking, storing computer-executable instructions capable of executing any of the above-described path planning methods for memory parking.

[0169] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0170] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0171] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0172] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0173] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0174] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0175] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0176] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0177] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0178] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0179] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the technical principles of this application should fall within the protection scope of this application.

Claims

1. A path planning method for memory-based parking, characterized in that, The method includes: Acquire environmental perception information of the vehicle during the memory parking process; Based on the vehicle's current location, extract potential hazards within a predetermined range from environmental perception information. These potential hazards include at least one of road boundaries and obstacles. The boundary of the hazard source is offset in a direction that increases the margin for safe driving, thereby adding noise to the environmental perception information; Based on the environmental perception information from noise processing, the vehicle performs path planning.

2. The method according to claim 1, characterized in that, The process of shifting the hazard source information in a direction that increases the safety margin specifically includes: When the hazard source is the road boundary, the road boundary will be shifted towards the center of the road; When the hazard source is an obstacle, shift the obstacle boundary toward the side with the larger drivable area.

3. The method according to claim 1, characterized in that, The step of performing path planning for the vehicle based on environmental perception information obtained from noise processing specifically includes: The vehicle makes behavioral decisions based on multi-source information during the memory parking process; the multi-source information includes the environmental perception information from the noise processing. When the decision result is to activate the driving-restricted planning strategy, a graph search-based planning algorithm performs a forward path search; When at least one feasible path solution is found, evaluate whether the feasible path meets the preset path quality requirements. If the path quality requirements are met, path planning is performed on the vehicle based on the feasible path and the remembered path.

4. The method according to claim 3, characterized in that, The planning algorithm for graph search is a hybrid A algorithm, and the heuristic cost function of the planning algorithm is calculated based on the following parameters: Lateral deviation between the extension point and the memory path, Euclidean distance between the extension point and the target point, and difference between the heading angle of the extension point and the heading angle of the target point.

5. The method according to claim 3, characterized in that, The method further includes: When no feasible path solution is found or the path quality requirements are not met, the graph search-based planning algorithm performs a bidirectional path search in both forward and backward directions. When at least multiple feasible path solutions are found, an optimal feasible path is selected from the multiple feasible path solutions based on a preset optimization evaluation index; If no feasible path solution is found, generate a parking trajectory with a speed of 0 along the previously planned tracking path.

6. The method according to claim 5, characterized in that, The method further includes: When the optimal feasible path includes a reversing operation, the endpoint of the reversing path is set as the waypoint; A graph search-based planning algorithm generates a first segmented path from the vehicle's current location to the waypoint, and a second segmented path from the waypoint to the target point. The first segmented path and the second segmented path are concatenated to replace the optimal feasible path.

7. The method according to claim 3, characterized in that, The method further includes: When the decision result is to activate the cruise connection planning strategy, it is determined whether the distance between the end point of the cruise memory path and the target storage location is greater than the preset distance threshold. When the distance exceeds a preset threshold, sampling is performed on the cruise memory path to find the target path point, so that the first trajectory driving direction of the parking path starting from the target path point meets the direction consistency requirement with the trajectory driving direction of the vehicle at the target path point. The target path point is determined as the starting point of the parking path, and the parking path of the vehicle is planned. The cruise memory path or the cruise tracking path planned with the target waypoint as the endpoint is spliced ​​with the parking path.

8. The method according to claim 7, characterized in that, The step of sampling along the cruise memory path to find the target path point specifically includes: Using the first sampling distance from the target storage location as the initial sampling value, sample the first path point on the cruise memory path; Determine whether the first trajectory direction of the parking path starting from the first path point meets the direction consistency requirement with the trajectory direction of the vehicle at the first path point. If the condition is not met, the second path point is resampled at a second sampling distance greater than the first sampling distance; Repeat the direction determination until a target path point that meets the direction consistency requirement is found or the number of samplings exceeds a preset threshold.

9. The method according to claim 7, characterized in that, The method further includes: When the distance is less than or equal to a preset distance threshold, a first cruise tracking path is planned from the current position of the vehicle to the path point at the initial sampling distance. Sampling is performed on the first cruise tracking path to find the target path point. or, Plan a second cruise tracking path from the vehicle's current position to a safe starting point, wherein the safe starting point is located in front of the target storage location and maintains a safe distance from the storage location; Using the aforementioned safe starting point as the starting or ending point of the parking path, a parking path is planned and then spliced ​​with the second cruise tracking path.

10. A path planning device for memory parking, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform a path planning method for memory parking as described in any one of claims 1-9.

Images