Large model-based automatic parking control method and system, and storage medium

Abstract

A large model-based automatic parking control method and system, and a storage medium. The method comprises the following steps: performing data initialization, acquiring and processing a parking lot map, and generating a long-term reference trajectory; on the basis of large model technology, monitoring an environment where an autonomous vehicle is located, and predicting the movement of an obstacle; on the basis of a predicted obstacle movement result, checking whether the autonomous vehicle is safe, if yes, performing obstacle avoidance planning, and if not, checking whether the long-term reference trajectory needs to be updated due to the movement of the obstacle; and if the long-term reference trajectory needs to be updated, optimizing the long-term reference trajectory to obtain a drivable ego-vehicle motion trajectory, and otherwise, tracking the latest reference trajectory in a dynamic environment on the basis of model predictive control. Compared with the prior art, the recognition and decision-making capabilities of autonomous driving technology under complex road conditions are improved, thereby improving the overall safety and automation level of autonomous driving.

Description

An automated parking control method, system, and storage medium based on a large model Technical Field

[0001] This invention relates to the field of automatic parking control, and more particularly to a method, system, and storage medium based on a large model. Background Technology

[0002] Fully automated parking remains challenging in current technologies, especially in dynamic environments with multiple independent agents. This is not only because it involves motion planning within confined spaces, but also because automated vehicles (AVs) must intelligently react to surrounding obstacle vehicles (OVs). Unlike driving on highways or expressways, vehicle movement in parking areas lacks a set of clear rules and depends heavily on the driver's intent and even skill level. This makes predicting the parking environment highly challenging; therefore, achieving a good level of control over the entire automated parking system requires an integrated predictive and planning autonomous parking system.

[0003] Motion prediction is crucial in automated parking because it determines the safety constraints of the planning module, thus determining the feasibility and smoothness of motion planning. Existing automated parking processes mainly suffer from the following technical problems:

[0004] 1) The problem that autonomous driving technology cannot accurately identify and judge in complex road conditions; 2) The problem that vehicle-to-everything (V2X) technology experiences network congestion and data transmission delay when a large number of vehicles are connected; 3) The problem that automatic parking systems cannot accurately identify and judge in narrow parking spaces or complex parking environments. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies, such as the inability of autonomous driving technology to accurately identify and judge complex road conditions, and to provide an automatic parking control method, system, and storage medium based on a large model.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] An automated parking control method based on a large model includes the following steps:

[0008] S1: Initialize the data, acquire and process the parking lot map, and generate a long-term reference trajectory;

[0009] S2: Based on large model technology, monitor the environment in which the autonomous vehicle is located and predict the movement of obstacles;

[0010] S3: Check whether the autonomous vehicle is safe based on the predicted obstacle movement results. If it is, perform obstacle avoidance planning; otherwise, proceed to step S4.

[0011] S4: Check whether the long-term reference trajectory needs to be repaired due to the movement of obstacles. If so, proceed to step S5; otherwise, proceed to step S6.

[0012] S5: Optimize the long-term reference trajectory to obtain a drivable vehicle trajectory; if optimization is not possible, update the parking lot map and regenerate the long-term reference trajectory.

[0013] S6: Model predictive control tracks the latest reference trajectory in a dynamic environment.

[0014] Furthermore, the obstacle movement prediction process includes a short-term motion prediction process and a long-term pattern prediction process;

[0015] The short-term motion prediction process includes determining the corresponding forward propagation estimated state based on the average state of the obstacle, thereby obtaining the short-term motion prediction result of the obstacle;

[0016] The long-term pattern prediction process includes determining the long-term motion pattern of the obstacle based on its historical state, dynamic model, and motion relative to the environment. The historical state and dynamic model of the obstacle are obtained through a short-term motion prediction process. The motion of the obstacle relative to the environment is used to combine with a cost map to determine the motion pattern of the obstacle, forming a safety margin and a safety boundary. Finally, the long-term motion prediction result of the obstacle is determined. The safety margin is used to determine whether the autonomous vehicle is safe, and the safety boundary is used to limit the movement trajectory of the autonomous vehicle.

[0017] Furthermore, the model predictive control process is used to track a reference trajectory given a safety margin and safety boundary by optimizing the base planner to calculate the tracking trajectory.

[0018] Furthermore, the obstacle avoidance planning is executed when the movement of an obstacle violates the corresponding safety margin, by exploring the forward drivable environment to find the optimal target.

[0019] Furthermore, when the reference trajectory needs to be repaired due to the movement of obstacles, if the area ahead is not feasible, the autonomous vehicle is controlled to stop on the reference trajectory and wait for the obstacle to disappear; if the waiting time exceeds a preset waiting threshold, the reference trajectory is updated to guide the autonomous vehicle to bypass the obstacle and return to the original path.

[0020] Furthermore, in step S2, big data and deep learning technologies are used to monitor the environmental information around the vehicle.

[0021] Furthermore, the method enables real-time interaction and information sharing between autonomous vehicles and their surrounding environment through vehicle-to-everything (V2X) communication between autonomous vehicles, roadside terminals, and the cloud.

[0022] Furthermore, the vehicle-to-everything (V2X) collaborative communication adopts a decentralized network structure, with data communication and processing performed through multi-node parallel processing and load balancing technology.

[0023] The present invention also provides an automatic parking control system based on a large model, including a memory and a processor, wherein the memory stores a computer program, and the processor calls the computer program to execute the steps of the method described above.

[0024] The present invention also provides a computer-readable storage medium on which a computer program is stored, the computer program being executed by a processor using the method described above.

[0025] Compared with the prior art, the present invention has the following advantages:

[0026] (1) This invention adopts an automatic parking cooperative technology solution based on a large model, which can more accurately identify and judge the environmental information around the vehicle under complex road conditions through big data and deep learning technology; further realize the prediction of the short-term and long-term motion of obstacles; based on the prediction results, it first detects whether the autonomous vehicle violates the safety margin through motion planning; if not, it checks whether the initial long-term reference trajectory needs to be corrected due to the motion of obstacles; if any of these situations occur, the reference trajectory will be updated; finally, a collision-free motion is planned based on the model prediction controller to track the latest long-term motion trajectory in the dynamic environment; overall, it improves the safety and automation level of autonomous driving.

[0027] (2) Optimize vehicle network performance: This invention, through a large-scale model-based automatic parking cooperative technology solution, can effectively solve the problems of network congestion and data transmission delay in the case of large-scale vehicle access by using distributed computing and edge computing technologies, thereby optimizing the performance of vehicle network and improving the efficiency of information sharing and exchange.

[0028] (3) Improve the convenience of automatic parking: This invention uses a large-model-based automatic parking cooperative technology solution to generate a better parking path through computer algorithms in narrow parking spaces or complex parking environments, thereby improving the convenience of automatic parking and reducing the complexity and difficulty of driver operation. Attached Figure Description

[0029] Figure 1 is a flowchart illustrating an automatic parking control method based on a large model provided in an embodiment of the present invention;

[0030] Figure 2 is a schematic diagram of the framework of an automatic parking control method based on a large model provided in an embodiment of the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0032] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0033] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0034] Example 1

[0035] A typical example of automated parking planning and control is described below:

[0036] Motion planning for automated parking in intelligent vehicles is a major challenge in automated parking scenarios. General motion planning algorithms cannot be directly applied to parking in areas with obstacles because this requires rapid replanning of complex driving maneuvers. Furthermore, motion planning algorithms specifically designed for automated parking either fail to combine short-term and long-term planning or cannot incorporate online path repair into new obstacles in a dynamic environment. Scene-aware planners implementing multiple strategies may be computationally efficient, resulting in a higher replanning rate, which is necessary to ensure safety. Overall, the planning and control module for automated parking mainly comprises four parts: obstacle and vehicle trajectory prediction, drivable area selection, local trajectory planning, and vehicle control.

[0037] In a typical distributed software model for parking path planning, prediction can utilize an Interacting Multiple Model (IMM) filter. Inputs include the vehicle's position, the heading angle, velocity, acceleration, and yaw rate of obstacle vehicles. The output intention has two types: going straight and avoiding oncoming vehicles. The "drivable area" selection employs a potential field method. Obstacle prediction trajectories, road edges, and the poses of static obstacles can all serve as inputs to the drivable boundary, constructing a potential energy function to assess hazard and drivability. "Local path planning" can use the Restricted Tracking Time (RRT) algorithm, employing two techniques to improve speed: dividing the drivable area into seeing surfaces and randomly sampling on these surfaces; and using the reciprocal of the aforementioned potential energy function as the probability density function for sampling to reduce the number of sampling points. Finding the minimum-cost path among the sampling points is achieved using dynamic programming (DP). The output trajectory can be smoothed using a piecewise polynomial (B-spline).

[0038] As shown in Figure 1, this embodiment provides an automatic parking control method based on a large model, including the following steps:

[0039] S1: Initialize the data, acquire and process the parking lot map, and generate a long-term reference trajectory;

[0040] S2: Based on large model technology, monitor the environment in which the autonomous vehicle is located and predict the movement of obstacles;

[0041] Specifically, big data and deep learning technologies are used to monitor the environmental information around the vehicle;

[0042] The obstacle movement prediction process includes short-term motion prediction and long-term pattern prediction.

[0043] The short-term motion prediction process includes determining the corresponding forward propagation estimated state based on the average state of the obstacle, thereby obtaining the short-term motion prediction result of the obstacle;

[0044] The long-term pattern prediction process includes determining the long-term motion pattern of obstacles based on their historical state, dynamic model, and motion relative to the environment. The historical state and dynamic model of obstacles are obtained through the short-term motion prediction process. The motion of obstacles relative to the environment is used to combine with the cost map to determine the motion pattern of obstacles, forming a safety margin and safety boundary. Finally, the long-term motion prediction result of obstacles is determined. The safety margin is used to determine whether the autonomous vehicle is safe, and the safety boundary is used to limit the movement trajectory of the autonomous vehicle.

[0045] S3: Check whether the autonomous vehicle is safe based on the predicted obstacle movement results. If it is, perform obstacle avoidance planning; otherwise, proceed to step S4.

[0046] Specifically, obstacle avoidance planning is used when the movement of an obstacle violates the corresponding safety margin, and the optimal target is found by exploring the forward drivable environment;

[0047] S4: Check if the long-term reference trajectory needs to be repaired due to the movement of obstacles. If so, proceed to step S5; otherwise, proceed to step S6.

[0048] Specifically, when the reference trajectory needs to be repaired due to the movement of obstacles, if the area ahead is infeasible, the autonomous vehicle is controlled to stop on the reference trajectory and wait for the obstacle to disappear; if the waiting time exceeds a preset waiting threshold, the reference trajectory is updated to guide the autonomous vehicle to bypass the obstacle and return to the original path.

[0049] S5: Optimize the long-term reference trajectory to obtain a drivable vehicle trajectory; if optimization is not possible, update the parking lot map and regenerate the long-term reference trajectory.

[0050] S6: Model predictive control tracks the latest reference trajectory in a dynamic environment;

[0051] Model predictive control is used to track a reference trajectory given safety margins and boundaries by optimizing a base planner to compute the tracking trajectory.

[0052] This scheme optimizes effective information based on a fundamental parking algorithm. First, a bidirectional A-search guided tree (BIAGT) is used to generate long-term motion references. Then, a strategic motion planner based on the results of a hybrid environment predictor implements three strategies: a model predictive control (MPC)-based safety controller for trajectory tracking if the reference remains valid in the environment; a search-based collision avoidance path planning system for quickly finding collision avoidance paths in emergency situations; and an optimized repairable path planning system when the reference fails. Based on these operations, a model-based hybrid environment predictor can plan the aforementioned predicted short-term and long-term motion patterns. Thus, a strategic motion planner is developed that can effectively plan under different conditions.

[0053] The specific process is as follows:

[0054] For system-level strategies in automated parking, current automated parking systems in static environments employ Interactive Multi-Model (IMM) for prediction and a sampling-based planning method. This method first predicts obstacle information and then selects the navigation strategy for the autonomous vehicle. For intelligent vehicles, this involves navigating the parking lot without actually parking. Conversely, by adopting a more comprehensive integrated parking system approach, both short-term and long-term environmental predictions can be made, which can then be used for autonomous parking strategic planning.

[0055] A. Definition of Motion Planning

[0056] The planning problem considering vehicle dynamics can be expressed as: X'=f(X)+g(X;u); (1)

[0057] Where X = [x; y; θ]T represents the two-dimensional coordinates and the vehicle heading, and u = [δ; v]T is the control input including longitudinal speed and steering angle.

[0058] Collision-free configuration space It is the configuration set where there is no intersection between the vehicle and the obstacle. The motion planning problem considered in this paper can be defined as finding a feasible trajectory Pt given an initial configuration X0 free, a target configuration Xf Cfree and formula (1), where (1) starts from 0 and ends at Xf, and simultaneously satisfies (1) and (2) lies in the collision-free configuration space Cfree. The common bicycle model is used to represent the vehicle motion. The discrete-time model obtained by Euler discretization is as follows:

[0059] (2) Where Ts is the sampling time.

[0060] B. System Architecture

[0061] Figure 2 illustrates the architecture of the proposed automated parking system. The hybrid environment predictor and the strategic motion planner are the two main components. During runtime, the central control first processes the parking lot map Mmap and generates an initial long-term trajectory Pref. Here, a guided tree search (BIAGT) is used to plan and generate a trajectory that allows the automated vehicle to accurately reach the target, a crucial feature for parking in confined spaces.

[0062] The hybrid environment predictor monitors the environment and predicts obstacle movement. Based on the predictions, the strategic motion planner checks if the autonomous vehicle has violated safety margins. If not, it checks if the initial long-term predicted trajectory (Pref) needs correction due to obstacle movement. If any of these occur, the Pref is updated. Finally, a collision-free motion is planned based on the Model Predictive Controller (MPC) to track the latest long-term motion trajectory (Pref) in the dynamic environment. If the correction planner cannot successfully optimize the drivable vehicle trajectory, it requests the central domain controller to update the map and regenerate the reference trajectory.

[0063] C. Hybrid Environment Predictor

[0064] Hybrid environment predictors comprise three main components: obstacle motion estimation, motion prediction, and pattern estimation. In particular, accurate short-term motion prediction enables autonomous vehicles to plan and react safely to obstacles; a major challenge in short-term prediction is estimating the obstacle's steering angle. Typically, an extended Kalman filter (EKF) can be used to reconstruct the obstacle's velocity, followed by steering estimation using an adaptive observer. Long-term planning / pattern prediction ensures that autonomous vehicles can plan reasonable driving paths more efficiently and smoothly during operation. While long-term motion is often difficult to predict accurately, it can be observed that drivers generally follow certain routes due to driving habits (e.g., in China, cars should drive on the right). Furthermore, the vehicle's motion throughout the parking / exit process can be captured through several "patterns" (e.g., maneuvering into and out of narrow spaces and cruising in aisles). Based on these two priors, cost maps can be used to capture these routes, combined with short-term predictions to determine obstacle patterns and make long-term predictions.

[0065] The following section will explain its main characteristics.

[0066] 1) Cascade motion estimation:

[0067] Motion estimation reconstructs the state X from measurements of (x; y) based on a unicycle model. Such processing is insufficient for parking processes involving frequent changes in direction of movement and steering. To accurately predict the short-term motion of obstacles, it is advantageous to reconstruct the control input u. This can be viewed as an unknown input estimation problem or a problem of estimating the state of the obstacle vehicle OV by augmenting the system state with the control input. Assume that the obstacle motion evolves according to model (2). An augmented model of OV is obtained by assuming that the control input (δ; v) is a piecewise constant, and the augmented state [x; y; θ; δ; v]T is estimated. Given a nonlinear augmented model, it is natural to apply a sophisticated nonlinear state estimator (e.g., an extended Kalman filter EKF or a particle filter) for state estimation. Accurately estimating the steering angle by adjusting the EKF is not straightforward, partly due to the presence of the term v tan(δ) involving multiplication of unmeasured states. While particle filters are good, their computational burden can hinder practical operation.

[0068] 2) Short-term motion prediction:

[0069] Short-term motion prediction prioritizes computational efficiency and real-time performance. We assume that the short-term motion of the obstacle vehicle (OV) is entirely captured by the average value and covariance of its state Xcc. The state Xcc, k is estimated through forward propagation, and a short-term prediction XH is obtained; k = [X1, kT, ..., XH, kT]T is used for the next H steps in the time range. Similarly, forward propagation is performed according to the forward prediction formula of the Extended Kalman Filter (EKF) to obtain the covariance matrix PmH, k = {Pmk+1, ..., Pmk+H}. This information will aid in long-term prediction and is used to determine the safety margin for each future time step.

[0070] 3) Long-term pattern prediction:

[0071] Long-term motion analysis focuses on the motion of the obstacle vehicle (OV) over a period of time, which depends on its historical state, dynamic model, and motion relative to the environment. The first two factors are captured to some extent by short-term motion prediction. To utilize motion relative to the environment, a cost map can be introduced (as shown in the figure below, a hybrid predictor predicts the short-term OV trajectory (green line) and combines it with pattern prediction to generate safety margins and safety boundaries at h=1 and h=H) to capture the possible long-term motion of the obstacle vehicle OV.

[0072] Additionally, a route planner can be used to construct a cost graph path Mroute, where the cost graph contains possible routes taken by obstacle vehicles (OVs).

[0073] Where Xmk;i is the i-th waypoint of the route in the pattern; and W1 and W2 are weight matrices. The function f(mk;Xcc;k) is proportional to the magnitude of the OV steering angle and the deviation between the OV heading angle and the final heading angle of the route. Finally, normalization is performed to obtain b(mk), and the value of mk with the highest confidence is m^k. In addition, vehicles in the parking lot typically operate in two modes, "manual mode" and "cruise mode". Vehicles in manual mode may frequently change steering and deviate from the route in the cost diagram (the black dashed line in the above figure) to park or leave narrow parking spaces. Vehicles in cruise mode have smaller or more stable steering angles and typically follow one of the routes. This mode is used when a vehicle first enters the parking lot and approaches a parking space or when a vehicle leaves a parking space and exits the parking lot.

[0074] 4) Safety margin and safety limits:

[0075] Because the motion of obstacle vehicles (OVs) in maneuvering mode is difficult to predict, the predictor needs to generate a safety margin for the vehicle's motion (a convex boundary of the OV's historical posture, shown by the orange line in the figure above). This results in a more conservative planning model, ensuring that the vehicle stays away from unpredictable obstacle vehicles (OVs). Note that, considering the kinematic model and route information, safety margins and safety boundaries can also be applied to other moving obstacles, such as pedestrians or motorcycles.

[0076] 5) Strategy Motion Planner

[0077] Using the reference trajectory (Pref), the main modules of the motion strategy planner include an MPC-based safety controller and two support modules: a collision avoidance trajectory planner and a repair trajectory planner. Both planners are activated if the vehicle's current position and the reference trajectory become invalid due to the vehicle's motion.

[0078] A. MPC-based Safety Controller: The safety controller tracks a reference trajectory (Pref) given a safety margin and safety limits. Using the Pref, an optimized base planner is employed to compute the tracking trajectory within the MPC framework. Assuming Xref;k is a segment of the Pref, a time step k is set for tracking. External references k are selected and pruned to ensure that the required safety margins in all modes or the safety boundaries in "maneuver" mode are not violated. The entire trajectory tracking problem can be formulated as a non-convex optimization problem, which can be solved using a nonlinear programming solver (e.g., IPOPT), with the reference path as a warm start, and an average solution time of approximately 0.06 seconds.

[0079] B. Obstacle Avoidance Planning: When the vehicle remains on the original reference trajectory, it may be considered unsafe, requiring the initiation of an obstacle avoidance planning process. This situation may occur when an obstacle (OV) approaches the vehicle, its motion significantly different from the previous prediction—potentially violating safety margins and posing a safety threat. Therefore, the vehicle needs to find a way to avoid the obstacle. Reverse movement is generally not recommended because the vehicle lacks a safe target. Instead, exploring the forward drivable environment to find the optimal target is preferable; therefore, a search-based obstacle avoidance planner is proposed, which can explore the space and quickly find an obstacle avoidance trajectory.

[0080] 6) Track Repair Planner:

[0081] The movement of an obstacle vehicle (OV) may invalidate the vehicle's reference trajectory. When the area ahead is infeasible, the safety controller will command the vehicle to stop on the reference trajectory. Unless a new reference trajectory is received, the safety controller will stop the vehicle and wait for the obstacle OV to disappear—if the OV remains stationary for an extended period, the trajectory planning is considered unreasonable or inefficient. In this case, the reference trajectory (Pref) should be updated so that the safety planner can guide the vehicle to bypass the OV and merge back into the original path. Note that the repaired trajectory is usually in the same trajectory set as the original trajectory, allowing for the direct generation of an executable trajectory based on optimizing the current trajectory to address safety hazards. To quickly obtain a repaired path, trajectory planning needs to be repaired in two-dimensional space, i.e., the trajectory is set as Xrepair = [x; y]. T The constraints are then modified accordingly. While the resulting path avoids collisions, it cannot guarantee consistent accuracy and could potentially lead to the autonomous vehicle hitting obstacles. Therefore, only if it passes the kinematic feasibility check is it recognized as a validated path and accepted as the corrected trajectory. If the correction fails, the central control unit is notified to take over the correction task.

[0082] This embodiment also provides an automatic parking control system based on a large model, including a memory and a processor. The memory stores a computer program, and the processor calls the computer program to execute the steps of the method described above.

[0083] This embodiment also provides a computer-readable storage medium storing a computer program, which is executed by a processor using the method described above.

[0084] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0085] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A large model-based automatic parking control method, characterized by, Includes the following steps: S1: Initialize the data, acquire and process the parking lot map, and generate a long-term reference trajectory; S2: Based on large model technology, monitor the environment in which the autonomous vehicle is located and predict the movement of obstacles; S3: Check whether the autonomous vehicle is safe based on the predicted obstacle movement results. If it is, perform obstacle avoidance planning; otherwise, proceed to step S4. S4: Check whether the long-term reference trajectory needs to be repaired due to the movement of obstacles. If so, proceed to step S5. Otherwise, proceed to step S6; S5: Optimize the long-term reference trajectory to obtain a drivable vehicle trajectory; if optimization is not possible, update the parking lot map and regenerate the long-term reference trajectory. S6: Model predictive control tracks the latest reference trajectory in a dynamic environment.

2. The automatic parking control method based on a large model according to claim 1, characterized in that, The obstacle movement prediction process includes a short-term motion prediction process and a long-term pattern prediction process; The short-term motion prediction process includes determining the corresponding forward propagation estimated state based on the average state of the obstacle, thereby obtaining the short-term motion prediction result of the obstacle; The long-term pattern prediction process includes determining the long-term motion pattern of the obstacle based on its historical state, dynamic model, and motion relative to the environment. The historical state and dynamic model of the obstacle are obtained through a short-term motion prediction process. The motion of the obstacle relative to the environment is used to combine with a cost map to determine the motion pattern of the obstacle, forming a safety margin and a safety boundary. Finally, the long-term motion prediction result of the obstacle is determined. The safety margin is used to determine whether the autonomous vehicle is safe, and the safety boundary is used to limit the movement trajectory of the autonomous vehicle.

3. The automatic parking control method based on a large model according to claim 2, characterized in that, The model predictive control process is used to track a reference trajectory given a safety margin and safety boundary, and calculates the tracking trajectory by optimizing the base planner.

4. The automatic parking control method based on a large model according to claim 2, characterized in that, The obstacle avoidance planning is executed when the movement of an obstacle violates the corresponding safety margin, and the optimal target is found by exploring the forward drivable environment.

5. The automatic parking control method based on a large model according to claim 2, characterized in that, When the reference trajectory needs to be repaired due to the movement of obstacles, if the area ahead is not feasible, the autonomous vehicle is controlled to stop on the reference trajectory and wait for the obstacle to disappear; if the waiting time exceeds a preset waiting threshold, the reference trajectory is updated to guide the autonomous vehicle to bypass the obstacle and return to the original path.

6. The automatic parking control method based on a large model according to claim 1, characterized in that, In step S2, big data and deep learning technologies are used to monitor the environmental information around the vehicle.

7. The automatic parking control method based on a large model according to claim 1, characterized in that, The method enables real-time interaction and information sharing between autonomous vehicles and their surrounding environment through vehicle-to-everything (V2X) communication between autonomous vehicles, roadside terminals, and the cloud.

8. The automatic parking control method based on a large model according to claim 7, characterized in that, The vehicle-to-everything (V2X) collaborative communication adopts a decentralized network structure, with data communication and processing performed through multi-node parallel processing and load balancing technology.

9. A large model-based automatic parking control system, characterized by, It includes a memory and a processor, the memory storing a computer program, the processor invoking the computer program to perform the steps of the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which is executed by a processor according to any one of claims 1 to 8.

Images