Automatic parking auxiliary method, automatic parking auxiliary system and vehicle

Abstract

The invention relates to an automatic parking auxiliary method, an automatic parking auxiliary system and a vehicle, and the method comprises the steps: obtaining sensing data collected by a vehicle, carrying out the first analysis processing of the sensing data, determining the environment type around the vehicle, and determining a weight distribution strategy corresponding to the environment type; second analysis processing is carried out on the sensing data, and the waiting state of the rear vehicle and the pedestrian postures of the surrounding pedestrians are calculated; judging whether the waiting state of the rear vehicle and the pedestrian posture meet preset conditions or not; if it is judged that the waiting state of the rear vehicle and the pedestrian posture meet the preset conditions, a weight distribution strategy is adjusted, and multiple alternative paths are generated based on the adjusted weight distribution strategy; and determining a target path from the plurality of alternative paths, and executing a parking action based on the target path.

Description

Technical Field

[0001] The present application relates to the field of parking, and in particular to an automatic parking assistance method, an automatic parking assistance system, and a vehicle. Background Art

[0002] Automatic Parking Assist (APA) technology, a core function of intelligent driving systems, primarily uses sensors such as ultrasonic radar and cameras to identify parking spaces and integrates vehicle dynamics models with pre-set algorithms to complete parking maneuvers. Its workflow encompasses four steps: environmental perception, parking space detection, path planning, and trajectory tracking. It relies on high-precision electronic maps and vehicle positioning data to achieve centimeter-level control. Most related technologies employ rule-based control strategies, using pre-programmed steering angle and speed parameters to perform standardized parking maneuvers.

[0003] However, this technology has limitations in practical application. Traditional automated parking assist (APA) technology relies on a pre-set path planning algorithm, a single decision-making approach, and lacks the ability to perceive the status of vehicles waiting behind or the posture of surrounding pedestrians. This makes it impossible to dynamically adjust parking strategies, resulting in lengthy parking times and increased traffic congestion in high-density parking lots (such as underground shopping mall parking lots and parking lots in front of open-air shops).

[0004] Regarding the problem that automatic parking assistance methods in related technologies take a long time to park in complex dynamic environments, no effective solution has yet been provided. Summary of the Invention

[0005] Based on this, it is necessary to provide an automatic parking assistance method, an automatic parking assistance system and a vehicle that can reduce parking time in complex dynamic environments to address the above technical problems.

[0006] In a first aspect, the present application provides an automatic parking assistance method, comprising:

[0007] Acquiring sensor data collected by the vehicle, performing a first analysis on the sensor data to determine an environment type around the vehicle, and determining a weight distribution strategy corresponding to the environment type, the weight distribution strategy including a plurality of weight parameters;

[0008] Performing a second analysis and processing on the sensor data to calculate the waiting state of the rear vehicle and the posture of the surrounding pedestrians;

[0009] determining whether the waiting state of the rear vehicle and the posture of the pedestrian satisfy preset conditions; wherein the preset conditions include: the distance between the rear vehicle and the vehicle does not exceed a first distance, the waiting time of the rear vehicle is not less than a first time, and the posture of the pedestrian poses a risk of entering the parking area;

[0010] If it is determined that the waiting state of the rear vehicle and the posture of the pedestrian meet the preset conditions, adjusting the weight distribution strategy and generating multiple alternative paths based on the adjusted weight distribution strategy;

[0011] A target path is determined from the plurality of candidate paths, and a parking maneuver is performed based on the target path.

[0012] In one embodiment, obtaining sensor data collected by the vehicle, performing a first analysis and processing on the sensor data, determining an environment type around the vehicle, and determining a weight distribution strategy corresponding to the environment type include:

[0013] extracting camera data, millimeter-wave radar data, and lidar data from the sensor data;

[0014] determining camera clarity and weather conditions based on the camera data;

[0015] determining a lidar point cloud density based on the lidar data;

[0016] Calculate the proportion of the millimeter-wave radar data in the sensor data, the proportion of millimeter-wave radar data;

[0017] Determine the current environment type based on the camera clarity, the weather conditions, the lidar point cloud density, and the millimeter-wave radar data ratio;

[0018] According to the association relationship between the environment type and the weight distribution strategy, a weight distribution strategy corresponding to the current environment type is determined.

[0019] In one embodiment, determining the current environment type based on the camera clarity, the weather conditions, the LiDAR point cloud density, and the millimeter wave radar data ratio includes:

[0020] When the camera clarity is not higher than a preset brightness threshold and the millimeter-wave radar data ratio is not lower than a preset ratio, it is determined that the current environment type is a low-light environment; or,

[0021] When the density of the laser radar point cloud is not less than a preset ratio compared with the system setting value, and the weather condition is precipitation, the current environment type is determined to be a precipitation weather environment; or,

[0022] When the data extracted based on the sensor data do not meet the above-mentioned low-light environment or precipitation weather environment, the current environment type is determined to be a normal environment.

[0023] In one embodiment, performing a second analysis on the sensor data to calculate the waiting state of the rear vehicle and the posture of surrounding pedestrians includes:

[0024] Acquiring historical motion data of the rear vehicle from the sensor data, the historical motion data including vehicle speed and vehicle distance change values;

[0025] Inputting the historical motion data into the LSTM network to predict the waiting time of the rear vehicle;

[0026] determining a waiting state of the rear vehicle according to the vehicle distance change value and the waiting time;

[0027] Obtaining camera data from the sensor data, and performing pedestrian posture estimation based on the camera data;

[0028] If the pedestrian posture estimation result shows that the target pedestrian is moving and the distance between the target pedestrian and the vehicle does not exceed the second distance, it is determined that there is a risk that the pedestrian posture will enter the parking area.

[0029] In one embodiment, if it is determined that the waiting state of the rear vehicle and the posture of the pedestrian meet the preset conditions, adjusting the weight distribution strategy includes:

[0030] Obtaining a waiting time of the rear vehicle, determining an adjustment range of a path length weight according to the waiting time, and adjusting the path length weight based on the adjustment range;

[0031] Adjust the safety distance weight up or down by the first preset ratio from the current value;

[0032] Adjust the parking time weight up or down by a second preset ratio based on the current value;

[0033] Increase the comfort weight by a third preset ratio from the current value.

[0034] In one embodiment, multiple alternative paths are generated based on the adjusted weight distribution strategy, including:

[0035] Creating a state space and an action space respectively; wherein the state space includes environmental parameters, the vehicle state of the vehicle, and parking space parameters; and the action space includes a steering wheel angle sequence corresponding to each of the alternative paths;

[0036] For each of the alternative paths, performing weighted summation on the preset reward items according to the corresponding weight parameters to obtain an immediate reward corresponding to each of the alternative paths;

[0037] Determine the actions consisting of the alternative paths at each moment, and based on the immediate rewards, calculate the long-term expected reward Q that can be obtained by performing the actions under the state at each moment;

[0038] Determine alternative paths for the corresponding actions based on the maximum value of the long-term expected return.

[0039] In one embodiment, determining an action consisting of alternative paths at each moment, and calculating the long-term expected reward Q that can be obtained by executing the action under the state at each moment based on the immediate reward, includes:

[0040] Q(St,At)←Q(St,At)+α×[Rt+γ×maxaQ(St+1,a)-Q(St,At)];

[0041] Rt=W1×L+W2×S+W3×T+W4×J;

[0042] Where t represents the time, St represents the state of the vehicle at time t, At represents the action of the vehicle at time t, Q(St, At) is the long-term expected reward that can be obtained after executing action At in state St, α is the learning rate, 0.1≤α≤0.3, γ is the discount factor, 0.8≤γ≤0.95, and a is the action variable in action At; Rt represents the immediate reward, W1 represents the path length weight, W2 represents the safety distance weight, W3 represents the parking time weight, W4 represents the comfort weight, L represents the path length, S represents the minimum safety distance, T represents the expected parking time, and J represents the comfort evaluation value.

[0043] In one embodiment, determining a target path from a plurality of candidate paths includes:

[0044] Verifying the feasibility of each of the alternative routes and calculating the comfort of each of the alternative routes;

[0045] Among the multiple alternative paths, the alternative path that passes the feasibility verification and meets the comfort index is used as the target path.

[0046] In a second aspect, the present application provides an automatic parking assistance system, comprising: a sensor, a neural network processor, and a domain controller, wherein the sensor and the domain controller are respectively connected to the neural network processor; wherein,

[0047] The sensor is used to generate sensing data;

[0048] The neural network processor is configured to execute the automatic parking assistance method described in the first aspect to determine a target path based on the sensor data;

[0049] The domain controller is configured to execute a parking action based on the target path.

[0050] In a third aspect, the present application provides a vehicle, which is used to execute the automatic parking assistance method described in the first aspect above.

[0051] The above-mentioned automatic parking assistance method, automatic parking assistance system and vehicle determine the current environment type by fusing multiple sensor data, and preliminarily determine the weight distribution strategy based on the correlation between the current environment type and weight parameters to avoid system paralysis caused by failure of a single sensor. Then, by detecting the waiting status of the rear vehicle and the posture of the surrounding pedestrians, it is determined whether the current environment is in a high-density traffic state. If so, the preliminary weight distribution strategy is adjusted to obtain the final weight distribution strategy. Based on the final weight distribution strategy, multiple alternative paths are output, and then the optimal path is evaluated from the multiple alternative paths to guide parking, which reduces the parking time in complex dynamic environments, reduces the risk of traffic congestion while also taking into account safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A schematic diagram of the hardware structure of an automatic parking assistance system in one embodiment;

[0053] Figure 2 1 is a flow chart of an automatic parking assistance method according to an embodiment;

[0054] Figure 3 1 is a flow chart of an automatic parking assistance method according to an embodiment;

[0055] Figure 4 A schematic flow chart of a method for generating an alternative path in one embodiment;

[0056] Figure 5 FIG. 1 is a schematic diagram of the software structure of an automatic parking assistance system in one embodiment. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0058] Unless otherwise defined, technical or scientific terms used in this application shall have the ordinary meanings as understood by persons of ordinary skill in the art to which this application belongs. The terms "a," "an," "the," "these," and similar expressions in this application do not denote limitations on quantity and may be singular or plural. The terms "comprise," "include," "have," and any variations thereof, as used in this application, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device comprising a series of steps or modules (units) is not limited to the listed steps or modules (units) but may include unlisted steps or modules (units) or other steps or modules (units) inherent to the process, method, product, or device. The terms "connected," "connected," "coupled," and similar expressions used in this application are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. As used in this application, "plurality" means two or more. "And / or" describes an association between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone; A and B exist simultaneously; or B exists alone. Generally, the character " / " indicates that the objects in the preceding and following relationship are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0059] In one embodiment, a vehicle is provided, which is capable of executing an automatic parking assistance method. The flowchart of the automatic parking assistance method can be found in the following Figure 3 、 Figure 4 and Figure 5 In this embodiment, the vehicle includes a vehicle body and an automatic parking assistance system, and the automatic parking assistance system is installed in the vehicle body. Figure 1 This is a hardware structure diagram of the automatic parking assistance system. Figure 1 As shown, the system includes: a sensor 11, a neural network processor 12 and a domain controller 13, and the sensor 11 and the domain controller 13 are connected to the neural network processor 12 respectively.

[0060] Sensor 11 is used to generate sensor data. Sensor 11 may include millimeter-wave radar, lidar, and cameras, which can be used to build a dynamic environment perception network. The millimeter-wave radar can be used to detect the speed of surrounding vehicles, the lidar can be used to obtain parking space geometry, and the camera can be used to capture the trajectories of surrounding pedestrians. The neural network processor 12 is used to execute an automatic parking assistance method based on the sensor data to determine a target path. The domain controller 13 is used to execute parking actions based on the target path.

[0061] A heterogeneous computing architecture (a neural network processor 12 handles real-time decision-making, while a domain controller 13 handles path planning) reduces computing power requirements. This architecture generates and evaluates three alternative routes per second, meeting the real-time requirements of high-density parking lots (e.g., parking times under 60 seconds). Furthermore, lightweight model deployment supports on-board edge computing, reducing reliance on cloud computing.

[0062] The automatic parking assistance method provided in the embodiment of the present application can be applied to Figure 2 In the application environment shown. The automatic parking assistance system 1 communicates with the server 2 through the network. The server 2 can be implemented as an independent server or a server cluster composed of multiple servers. The data storage system 3 can store data that the server needs to process. The data storage system 3 can be integrated on the server 1, or placed on the cloud or other network servers. The automatic parking assistance system 1 can also include an on-board electronic terminal for communicating with the server 2, such as obtaining parking space data from the server 2, or downloading a trained machine learning model from the server 2 and deploying it to the neural network processor 12. The on-board electronic terminal can be a T-Box (Telematics Box, remote information processor) or an IVI host (In-Vehicle Infotainment, in-vehicle infotainment system).

[0063] In one embodiment, Figure 3 A flow chart of an automatic parking assistance method is provided, wherein the method is run on Figure 1 or Figure 2 Taking the automatic parking assistance system in the example as an example, the following steps are included:

[0064] Step S101 , acquiring sensor data collected by the vehicle, performing a first analysis and processing on the sensor data, determining the type of environment around the vehicle, and determining a weight distribution strategy corresponding to the environment type, wherein the weight distribution strategy includes a plurality of weight parameters.

[0065] Sensors in a vehicle may include millimeter-wave radars, lidars, and cameras, and the sensing data includes millimeter-wave radar data (such as the speed of surrounding vehicles), lidar data (such as parking space geometry parameters), and camera data (such as the trajectories of surrounding pedestrians).

[0066] The automated parking assist system (hereinafter referred to as the "system") performs the initial analysis and processing of sensor data by acquiring sensor data, extracting camera data, millimeter-wave radar data, and lidar data. It then extracts feature information from each sensor data and matches this information with pre-defined environment types to determine the current environment type. Feature information includes camera clarity, weather conditions, lidar point cloud density, and millimeter-wave radar data percentage. Pre-defined environment types include low-light, precipitation, and normal. This step integrates multiple sensors, enhancing the comprehensiveness of decision-making dimensions.

[0067] In some embodiments, the system extracts camera data, millimeter-wave radar data, and lidar data from the sensor data; determines camera clarity and weather conditions based on the camera data; determines lidar point cloud density based on the lidar data; calculates the proportion of millimeter-wave radar data in the sensor data and the proportion of millimeter-wave radar data; determines the current environment type based on camera clarity, weather conditions, lidar point cloud density, and millimeter-wave radar data; and determines a weight allocation strategy corresponding to the current environment type based on the correlation between the environment type and the weight allocation strategy.

[0068] When the camera clarity is no higher than the preset brightness threshold (for example, 50 lux) and the proportion of millimeter-wave radar data is no lower than the preset proportion (for example, 50%), the current environment type is determined to be a low-light environment.

[0069] Alternatively, when the density of the lidar point cloud is not less than a preset ratio (e.g., 30%) compared to the system setting value, and the weather condition is precipitation (e.g., rain, snow, or hail), the current environment type is determined to be a precipitation weather environment.

[0070] Alternatively, when the data extracted based on the sensor data do not satisfy the above-mentioned low-light environment or precipitation weather environment, it is determined that the current environment type is a normal environment.

[0071] The weight parameters of the weight allocation strategy include at least two of the following: path length weight, safety distance weight, parking time weight, and comfort weight. For example, the mapping relationship between environment types and weight parameters (i.e., the weight allocation strategy) is shown in the following table:

[0072]

[0073] Step S102 , performing a second analysis on the sensor data to calculate the waiting state of the rear vehicle and the posture of the surrounding pedestrians.

[0074] When the system performs the second analysis and processing on the sensor data, the following steps are included:

[0075] (1) Obtain the historical motion data of the rear vehicle from the sensor data, which includes the vehicle speed and the change in distance between the vehicles. Input the historical motion data into the LSTM (Long Short-Term Memory) network to predict the waiting time of the rear vehicle (preferably, the error is <0.5 seconds). Determine the waiting state of the rear vehicle based on the change in distance between the vehicles and the waiting time.

[0076] (2) Obtain camera data from the sensor data, and perform pedestrian posture estimation based on the camera data and the OpenPose algorithm; if the pedestrian posture estimation result shows that the target pedestrian is moving and the distance between the target pedestrian and the vehicle does not exceed the second distance, it is determined that the pedestrian posture has the risk of entering the parking area.

[0077] The system improves its perception of rear vehicles and surrounding pedestrians by performing a secondary analysis and processing of sensor data. The system combines the waiting status of rear vehicles and the posture of surrounding pedestrians to determine whether the current environment requires adjustment of the weight distribution strategy, making the decision dimension more comprehensive.

[0078] Step S103, determining whether the waiting state of the rear vehicle and the posture of the pedestrian meet preset conditions; wherein the preset conditions include: the distance between the rear vehicle and the own vehicle does not exceed a first distance, the waiting time of the rear vehicle is not less than a first time, and the posture of the pedestrian poses a risk of entering the parking area.

[0079] The first distance may be 2 to 3 meters, and the first time may be 10 to 15 seconds. For example, if the distance between the rear vehicle and the vehicle is detected to be no more than 2 meters for 10 seconds, and the pedestrian's posture poses a risk of entering the parking area, it can be determined that the waiting state of the rear vehicle and the pedestrian's posture meet the preset conditions.

[0080] It should be noted that if this embodiment only considers the waiting state of the following vehicle, ignores the posture of surrounding pedestrians, and simply determines whether a single dynamic factor meets the preset conditions, and then adjusts the weight allocation strategy if it does, it will easily lead to safety risks. Therefore, this embodiment combines the perception of the following vehicle and surrounding pedestrians, comprehensively considering these two dynamic factors to adjust the weight allocation strategy, thus balancing parking efficiency and safety.

[0081] In step S104 , if it is determined that the waiting state of the rear vehicle and the posture of the pedestrian meet the preset conditions, the weight distribution strategy is adjusted, and a plurality of alternative paths are generated based on the adjusted weight distribution strategy.

[0082] When it is determined that the waiting state of the rear vehicle and the posture of the pedestrian meet the preset conditions, it means that the current scene is a high-density traffic scene. On the one hand, there are rear vehicles waiting, and on the other hand, there is a risk of pedestrians entering the parking area.

[0083] When adjusting the weight distribution strategy, adjustments can be made based on the current environment type, and the adjustment components can be superimposed on each weight parameter to adjust each weight parameter separately, making the adjustment more refined.

[0084] When generating multiple alternative paths based on the adjusted weight distribution strategy, the traditional S-shaped trajectory can be used as the baseline path to generate a standard path; a path that reduces the number of turns can be designed, allowing a lateral acceleration of 0.8g (requiring seatbelt pretensioning) to generate a fast path for quick parking; a 5cm buffer zone can also be added around the path to generate a safe path suitable for special weather environments such as rain and snow to ensure parking safety.

[0085] Step S105 : determining a target path from the multiple candidate paths, and performing a parking action based on the target path.

[0086] The system verifies the feasibility of each alternative path and calculates the comfort of each alternative path; among multiple alternative paths, the alternative path that passes the feasibility verification and meets the comfort index will be used as the target path.

[0087] (1) Feasibility verification can be achieved through the following methods:

[0088] Obtain the steering angle range, lateral acceleration, and wheel slip rate corresponding to each alternative path; for each alternative path, determine whether the above parameters corresponding to the alternative path fall within a preset range. If so, the alternative path passes the feasibility verification; otherwise, it is determined to have failed. For example, the following vehicle parameters can be verified based on the vehicle dynamics model:

[0089] Steering angle range: -45°~+45°;

[0090] Lateral acceleration: ≤0.8g (fast path) or ≤0.5g;

[0091] Wheel slip rate: ≤15%;

[0092] If any of the parameters in the alternative path exceeds the above range, it is determined that the alternative path fails the feasibility verification.

[0093] (2) Comfort calculation can be achieved through the following methods:

[0094] Continuously derivate the alternative path x(t) and calculate the jerk (acceleration rate) value on the alternative path. The calculation formula is as follows:

[0095] speed:

[0096] Acceleration:

[0097] jerk:

[0098] Where j(t) represents jerk, the rate of change of acceleration, t represents time, and x(t) represents the position of the ego vehicle at time t. For example, the comfort index can be set to 2 m / s³. When jerk is less than 2 m / s³, parking comfort is ensured.

[0099] In an automatic parking assist system, the neural network processor will determine the target path through the above steps and input it into the domain controller. The domain controller will convert the selected target path into specific control instructions (such as steering wheel angle, throttle / brake percentage, etc.) to control the vehicle to perform parking actions.

[0100] In the above steps S101 to S105, the current environment type (low-light scene, precipitation weather scene, normal scene) is determined by fusing multiple sensor data, and a weight allocation strategy is preliminarily determined based on the correlation between the environment type and the weight parameters to avoid system paralysis caused by the failure of a single sensor. Then, by detecting the waiting state of the rear vehicle and the posture of the surrounding pedestrians, it is determined whether the current environment is in a high-density traffic state. If so, the preliminary weight allocation strategy is adjusted to obtain a final weight allocation strategy. Based on the final weight allocation strategy, multiple alternative paths are output, and the optimal path is evaluated from the multiple alternative paths to guide parking. This reduces the parking time in complex dynamic environments, reduces the risk of traffic congestion, and also takes safety into account.

[0101] In some embodiments, if it is determined that the waiting state of the rear vehicle and the posture of the pedestrian meet the preset conditions, the weight distribution strategy is adjusted, which can be achieved by the following method:

[0102] Obtain the waiting time of the vehicle behind, determine the adjustment range of the path length weight based on the waiting time, and adjust the path length weight based on the adjustment range; adjust the safety distance weight up or down by a first preset ratio from the current value; adjust the parking time weight up or down by a second preset ratio from the current value; and adjust the comfort weight up or down by a third preset ratio from the current value.

[0103] The first, second, and third preset ratios can be 40%, 50%, or 60%. For every 5 seconds the waiting time of the following vehicle increases, the path length weight increases linearly by 0.05 (upper limit 0.5) to prioritize shorter paths and reduce queuing.

[0104] In some embodiments, Figure 4 A flowchart of an alternative path generation method is provided, such as Figure 4 As shown, generating multiple alternative paths based on the adjusted weight distribution strategy includes the following steps:

[0105] Step S201 , creating a state space and an action space respectively; wherein the state space includes environmental parameters, the vehicle state of the vehicle and parking space parameters, and the action space includes a steering wheel angle sequence corresponding to each alternative path.

[0106] Create a state space S = {E, V, P}, where E represents environmental parameters (millimeter-wave radar data, lidar data, camera data), V represents the vehicle state (speed, steering angle, dynamic parameters), and P represents parking space parameters (length, width, and angle between the vehicle and the parking space).

[0107] Define the action space A = {a1, a2, a3}, where a1 is the steering wheel angle sequence corresponding to the standard path, a2 is the steering wheel angle sequence corresponding to the fast path, and a3 is the steering wheel angle sequence corresponding to the safe path.

[0108] Step S202 : For each candidate path, a weighted sum is performed on the preset reward items according to the corresponding weight parameters to obtain an immediate reward corresponding to each candidate path.

[0109] In this embodiment, the weighted parameters of the weighted allocation strategy include path length weight, safety distance weight, parking time weight, and comfort weight. The corresponding reward items include path length, minimum safety distance, expected parking time, and comfort evaluation value. The instant reward calculation formula for each alternative path is as follows:

[0110] Rt=W1×L+W2×S+W3×T+W4×J

[0111] Among them, R t represents the immediate reward, W1 represents the path length weight, W2 represents the safety distance weight, W3 represents the parking time weight, W4 represents the comfort weight, L represents the path length reward (unit: meter), S represents the minimum safety distance reward (unit: meter), T represents the estimated parking time reward (unit: second), and J represents the comfort reward (unit: m / s³). The weight parameters of each alternative path are as follows:

[0112] The standard path takes W1=0.25, W2=0.35, W3=0.2, W4=0.2;

[0113] The fast path takes W1=0.4, W2=0.2, W3=0.3, W4=0.1;

[0114] The safe path takes W1=0.2, W2=0.5, W3=0.15, and W4=0.15.

[0115] Step S203: Determine the actions consisting of the alternative paths at each moment, and calculate the long-term expected reward Q that can be obtained by executing the actions under the state at each moment based on the immediate reward.

[0116] The formula for calculating the long-term expected return Q is as follows:

[0117] Q(S t , A t )←Q(S t , A t )+α×[R t +γ×max a Q(S t+1 , a)-Q(S t , A t )]

[0118] Rt=W1×L+W2×S+W3×T+W4×J

[0119] Among them, t represents the time, S t Represents the state of the vehicle at time t, A t represents the action of the vehicle at time t, Q(S t , A t ) is in state S t Next, perform action A t The long-term expected return that can be obtained after α is the learning rate, 0.1≤α≤0.3, γ is the discount factor, 0.8≤γ≤0.95, and a is the action A t The action variable in R t represents the immediate reward, W1 represents the path length weight, W2 represents the safety distance weight, W3 represents the parking time weight, W4 represents the comfort weight, L represents the path length, S represents the minimum safety distance, T represents the expected parking time, and J represents the comfort evaluation value.

[0120] Step S204: determining an alternative path for the corresponding action based on the maximum value of the long-term expected return.

[0121] Based on the real-time updated Q value, determine the corresponding action A t The standard path, fast path, and safe path in .

[0122] For example, in high-density traffic scenarios, shorter alternative paths offer higher rewards. For example, when generating fast paths, by increasing W3 (parking time weight), the algorithm prioritizes paths with fewer turns and allows lateral accelerations up to 0.8g.

[0123] In steps S201 to S204, a Q-learning reinforcement learning framework is employed to generate multiple alternative paths. A dynamic reward function is then used to balance path length, safe distance, parking time, and comfort. This mechanism updates the Q-value table in real time, responding to dynamic environmental changes such as the speed of surrounding vehicles and pedestrian trajectories (e.g., a vehicle suddenly accelerating from behind or a pedestrian entering the parking area). This mechanism mitigates the collision risks associated with delayed path planning in traditional systems. Furthermore, through differentiated path design (e.g., a fast path allows lateral acceleration up to 0.8g, while a safe path adds a 5cm buffer zone), parking time is reduced by 30%-40% while ensuring a collision probability of less than 0.1%, effectively addressing the imbalance between safety and efficiency during parking.

[0124] In one embodiment, Figure 5 A software structure diagram of an automatic parking assistance system is provided. Figure 5 As shown in the figure, the automatic parking assistance system includes the perception layer, decision layer and execution layer. The layers are described as follows:

[0125] The perception layer is responsible for collecting information about the vehicle's surroundings and its own status. Millimeter-wave radar detects the speed of surrounding vehicles; the YOLOv5 object detection algorithm uses cameras to capture pedestrian trajectories; and lidar acquires parking space geometry. The perception layer also collects parameters such as the vehicle's speed, steering angle, yaw rate, and sideslip angle. The perception layer transmits this data to the decision layer.

[0126] Decision-making layer: Based on the reinforcement learning framework, the Q-learning model is used to process the perception layer data, complete dynamic environment assessment, multi-path generation and decision-making mechanism optimization, and generate corresponding control instructions.

[0127] The execution layer receives control commands from the decision layer, controls the vehicle's steering wheel angle, accelerator / brake, and executes specific parking maneuvers. The execution results are then fed back to the perception layer, forming a closed-loop control loop. Dynamic changes in the external environment will also continuously affect the data collected by the perception layer.

[0128] This embodiment coordinates the perception, decision-making, and execution layers. The decision-making layer uses a reinforcement learning (Q-learning) algorithm to evaluate path feasibility in real time and generate three alternative parking paths (standard path, fast path, and safe path). It also introduces a traffic pressure perception mechanism. When a vehicle waiting behind is detected, it dynamically adjusts the path weight (for example, shortening the buffer zone to 5cm and allowing lateral acceleration of up to 0.8g) to switch to "Quick Entry Mode." While ensuring a collision probability of less than 0.1%, this reduces parking time by over 30%, significantly improving traffic efficiency and user experience in scenarios such as high-density parking lots and shared mobility.

[0129] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0130] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), data processing logic devices based on quantum computing, and the like.

[0131] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0132] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. An automatic parking assistance method, characterized in that: include: Acquiring sensor data collected by the vehicle, performing a first analysis on the sensor data to determine an environment type around the vehicle, and determining a weight distribution strategy corresponding to the environment type, the weight distribution strategy including a plurality of weight parameters; Performing a second analysis and processing on the sensor data to calculate the waiting state of the rear vehicle and the posture of the surrounding pedestrians; determining whether the waiting state of the rear vehicle and the posture of the pedestrian satisfy preset conditions; wherein the preset conditions include: the distance between the rear vehicle and the vehicle does not exceed a first distance, the waiting time of the rear vehicle is not less than a first time, and the posture of the pedestrian poses a risk of entering the parking area; If it is determined that the waiting state of the rear vehicle and the posture of the pedestrian meet the preset conditions, adjusting the weight distribution strategy and generating multiple alternative paths based on the adjusted weight distribution strategy; A target path is determined from the plurality of candidate paths, and a parking maneuver is performed based on the target path.

2. The automatic parking assistance method according to claim 1, characterized in that: Acquiring sensor data collected by the vehicle, performing a first analysis and processing on the sensor data, determining an environment type around the vehicle, and determining a weight distribution strategy corresponding to the environment type, including: extracting camera data, millimeter-wave radar data, and lidar data from the sensor data; determining camera clarity and weather conditions based on the camera data; determining a lidar point cloud density based on the lidar data; Calculate the proportion of the millimeter-wave radar data in the sensor data, the proportion of millimeter-wave radar data; Determine the current environment type based on the camera clarity, the weather conditions, the lidar point cloud density, and the millimeter-wave radar data ratio; According to the association relationship between the environment type and the weight distribution strategy, a weight distribution strategy corresponding to the current environment type is determined.

3. The automatic parking assistance method according to claim 2, characterized in that: Determine the current environment type based on the camera clarity, the weather conditions, the LiDAR point cloud density, and the millimeter-wave radar data ratio, including: When the camera clarity is not higher than a preset brightness threshold and the millimeter-wave radar data ratio is not lower than a preset ratio, it is determined that the current environment type is a low-light environment; or, When the density of the laser radar point cloud is not less than a preset ratio compared with the system setting value, and the weather condition is precipitation, the current environment type is determined to be a precipitation weather environment; or, When the data extracted based on the sensor data do not meet the above-mentioned low-light environment or precipitation weather environment, the current environment type is determined to be a normal environment.

4. The automatic parking assistance method according to claim 1, characterized in that: Performing a second analysis and processing on the sensor data to calculate the waiting state of the rear vehicle and the posture of the surrounding pedestrians, including: Acquiring historical motion data of the rear vehicle from the sensor data, the historical motion data including vehicle speed and vehicle distance change values; Inputting the historical motion data into the LSTM network to predict the waiting time of the rear vehicle; determining a waiting state of the rear vehicle according to the vehicle distance change value and the waiting time; Obtaining camera data from the sensor data, and performing pedestrian posture estimation based on the camera data; If the pedestrian posture estimation result shows that the target pedestrian is moving and the distance between the target pedestrian and the vehicle does not exceed the second distance, it is determined that there is a risk that the pedestrian posture will enter the parking area.

5. The automatic parking assistance method according to claim 1, characterized in that: If it is determined that the waiting state of the rear vehicle and the posture of the pedestrian meet the preset conditions, adjusting the weight distribution strategy includes: Obtaining a waiting time of the rear vehicle, determining an adjustment range of a path length weight according to the waiting time, and adjusting the path length weight based on the adjustment range; Adjust the safety distance weight up or down by the first preset ratio from the current value; Adjust the parking time weight up or down by a second preset ratio based on the current value; Increase the comfort weight by a third preset ratio from the current value.

6. The automatic parking assistance method according to claim 1, characterized in that: Generate multiple alternative paths based on the adjusted weight distribution strategy, including: Creating a state space and an action space respectively; wherein the state space includes environmental parameters, the vehicle state of the vehicle, and parking space parameters; and the action space includes a steering wheel angle sequence corresponding to each of the alternative paths; For each of the alternative paths, performing weighted summation on the preset reward items according to the corresponding weight parameters to obtain an immediate reward corresponding to each of the alternative paths; Determine the actions consisting of the alternative paths at each moment, and based on the immediate rewards, calculate the long-term expected reward Q that can be obtained by performing the actions under the state at each moment; Determine alternative paths for the corresponding actions based on the maximum value of the long-term expected return.

7. The automatic parking assistance method according to claim 6, characterized in that: Determine the actions consisting of the alternative paths at each moment, and based on the immediate rewards, calculate the long-term expected reward Q that can be obtained by executing the actions under the state at each moment, including: Q(S t ,A t )←Q(S t ,A t )+α×[R t +γ×max a Q(S t+1 ,a)-Q(S t ,A t )]; Rt=W1×L+W2×S+W3×T+W4×J; Among them, t represents the time, S t Represents the state of the vehicle at time t, A t represents the action of the vehicle at time t, Q(S t , A t ) is in state S t Next, perform action A t The long-term expected return that can be obtained after α is the learning rate, 0.1≤α≤0.3, γ is the discount factor, 0.8≤γ≤0.95, and a is the action A t The action variable in R t represents the immediate reward, W1 represents the path length weight, W2 represents the safety distance weight, W3 represents the parking time weight, W4 represents the comfort weight, L represents the path length, S represents the minimum safety distance, T represents the expected parking time, and J represents the comfort evaluation value.

8. The automatic parking assistance method according to claim 1, characterized in that: Determining a target path from the plurality of candidate paths includes: Verifying the feasibility of each of the alternative routes and calculating the comfort of each of the alternative routes; Among the multiple alternative paths, the alternative path that passes the feasibility verification and meets the comfort index is used as the target path.

9. An automatic parking assistance system, characterized in that: include: A sensor, a neural network processor and a domain controller, wherein the sensor and the domain controller are respectively connected to the neural network processor; wherein, The sensor is used to generate sensing data; The neural network processor is used to execute the automatic parking assistance method according to any one of claims 1 to 8 to determine a target path based on the sensor data; The domain controller is configured to execute a parking action based on the target path.

10. A vehicle, characterized in that: The vehicle is used to execute the automatic parking assistance method according to any one of claims 1 to 8.

Images