Automatic driving vehicle path planning method and system combined with high-precision perception

Abstract

The invention discloses an automatic driving vehicle path planning method and system combined with high-precision perception, and relates to the technical field of automatic driving vehicles, and the method comprises the steps: obtaining the multi-mode real-time data of the surrounding environment of a target vehicle through a sensor module; performing fusion processing to generate a high-precision environment perception map; generating a global path and an initial local path by combining the current position and the target position of the target vehicle; dynamically optimizing the initial local path based on model predictive control to obtain a final local path; and controlling the target vehicle according to the global path and the final local path. The technical problem that in the prior art, due to the fact that environment sensing precision is insufficient, path planning is not flexible enough and dynamic obstacle effective prediction is lacked, automatic driving safety and path planning real-time performance are poor is solved, and the technical effects of improving automatic driving safety and enhancing path planning real-time performance and adaptability are achieved.

Description

Technical Field

[0001] This application relates to the technical field of autonomous vehicles, and specifically to an autonomous vehicle path planning method and system that combines high-precision perception. Background Art

[0002] As one of the cores of autonomous driving technology, autonomous vehicle path planning aims to achieve safe and efficient driving of the vehicle from the starting point to the ending point. However, traditional autonomous driving systems often rely on a single sensor or a combination of low-precision sensors, resulting in incomplete and inaccurate environmental perception information. Current path planning methods often rely too much on static map data, lack flexible response to real-time environmental changes, and in complex traffic environments, the prediction of dynamic obstacles is an important challenge for autonomous driving systems. Existing methods often have difficulty accurately predicting the movement trajectories of dynamic obstacles, increasing the collision risk and affecting the safety of autonomous driving.

[0003] Therefore, in the current related technologies, there are technical problems such as insufficient environmental perception accuracy, inflexible path planning, and lack of effective prediction of dynamic obstacles, which in turn lead to poor safety and real-time performance of autonomous vehicle path planning. Summary of the Invention

[0004] This application provides an autonomous vehicle path planning method and system that combines high-precision perception, solves the technical problems in the prior art of insufficient environmental perception accuracy, inflexible path planning, and lack of effective prediction of dynamic obstacles, which in turn lead to poor safety and real-time performance of autonomous vehicle path planning, and achieves the technical effects of improving the safety of autonomous driving, enhancing the real-time performance and adaptability of path planning.

[0005] This application provides an autonomous vehicle path planning method that combines high-precision perception. The method includes: obtaining multi-modal real-time data of the environment around the target vehicle through a sensor module; performing fusion processing on the multi-modal real-time data to generate a high-precision environmental perception map; generating a global path and an initial local path according to the high-precision environmental perception map, in combination with the current position and the target position of the target vehicle; dynamically optimizing the initial local path based on model predictive control to obtain a final local path; and controlling the target vehicle according to the global path and the final local path.

[0006] In a possible implementation, the autonomous vehicle path planning method that combines high-precision perception further performs the following processing: the sensor module includes high-precision sensors and a redundant sensor group, where the high-precision sensors include lidar, cameras, and ultrasonic sensors; performing real-time fault detection on the high-precision sensors, and if a fault exists in the high-precision sensors, switching to the redundant sensor group to trigger the path safety mode.

[0007] In a possible implementation, the path planning method for an autonomous driving vehicle combined with high-precision perception further performs the following processing: collecting V2X traffic flow data based on multiple communication methods, where V2X includes vehicle-to-vehicle, vehicle-to-infrastructure, vehicle-to-pedestrian, and vehicle-to-network. Among them, the V2X traffic flow data includes local map information and traffic flow data; combining the local map information, the traffic flow data, and traffic signal phase information to establish an intersection passing strategy; the target vehicle loads the intersection passing strategy in real time to establish the initial local path.

[0008] In a possible implementation, the path planning method for an autonomous driving vehicle combined with high-precision perception further performs the following processing: monitoring environmental changes in real time, and triggering local path replanning when the collision risk exceeds a threshold, where the local path replanning is used to adjust the driving trajectory of the target vehicle in real time.

[0009] In a possible implementation, the path planning method for an autonomous driving vehicle combined with high-precision perception further performs the following processing: using the global path as a reference trajectory for local path optimization; if the deviation between the initial local path and the global path exceeds the path deviation threshold, dynamically adjusting the local path planning parameters to obtain an optimized local path; predicting the position of the target vehicle based on the current state of the target vehicle combined with the dynamic model; adjusting the optimized local path based on the position of the target vehicle to obtain the final safe path.

[0010] In a possible implementation, the path planning method for an autonomous driving vehicle combined with high-precision perception further performs the following processing: performing spatio-temporal alignment on the real-time data, including mapping the multi-modal real-time data to the BEV space, and performing fusion processing on the multi-modal real-time data and the historical data of the target vehicle to capture dynamic environmental changes; extracting multiple features of the multi-modal real-time data, and performing confidence-weighted fusion on the multiple features; based on the fused features, processing and generating a high-precision environmental perception map, including road boundary, obstacle position, traffic sign, and traffic signal information.

[0011] In a possible implementation, the path planning method for an autonomous driving vehicle combined with high-precision perception further performs the following processing: evaluating the confidence of the multiple features; based on the confidence, constructing a feature map of the target vehicle environmental data; using the feature map for weighted calculation to establish the fused features.

[0012] In a possible implementation, the path planning method for an autonomous driving vehicle combined with high-precision perception further performs the following processing: the high-precision environmental perception map further includes a predicted trajectory of dynamic obstacles, where the predicted trajectory of dynamic obstacles is established through an LSTM network.

[0013] In a possible implementation, the method for path planning of an autonomous driving vehicle combined with high-precision perception further performs the following processing: obtaining historical motion trajectory data of dynamic obstacles collected by the high-precision sensor; performing normalization processing on the historical motion trajectory data to obtain a standardized data sequence; training the LSTM network according to the standardized data sequence to establish a dynamic obstacle prediction model; performing dynamic obstacle prediction based on the dynamic obstacle prediction model to establish a dynamic obstacle prediction trajectory, and updating it in real time to the high-precision environmental perception map.

[0014] This application also provides a path planning system for an autonomous driving vehicle combined with high-precision perception, including: a multi-modal real-time data acquisition module for acquiring multi-modal real-time data of the environment around the target vehicle through a sensor module; a high-precision environmental perception map generation module for performing fusion processing on the multi-modal real-time data to generate a high-precision environmental perception map; a path generation module for generating a global path and an initial local path according to the high-precision environmental perception map, in combination with the current position and the target position of the target vehicle; a final local path obtaining module for dynamically optimizing the initial local path based on model predictive control to obtain a final local path; and a target vehicle control module for controlling the target vehicle according to the global path and the final local path.

[0015] It is intended to use the method and system for path planning of an autonomous driving vehicle combined with high-precision perception proposed in this application to acquire multi-modal real-time data of the environment around the target vehicle through a sensor module; perform fusion processing to generate a high-precision environmental perception map; generate a global path and an initial local path in combination with the current position and the target position of the target vehicle; dynamically optimize the initial local path based on model predictive control to obtain a final local path; and control the target vehicle according to the global path and the final local path. This solves the technical problems in the prior art, such as insufficient environmental perception accuracy, inflexible path planning, and lack of effective prediction of dynamic obstacles, which lead to poor safety and real-time performance of autonomous driving path planning, and achieves the technical effects of improving the safety of autonomous driving, enhancing the real-time performance and adaptability of path planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments of the present disclosure will be briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the operations before or below do not necessarily need to be executed precisely in sequence. On the contrary, according to the need, they can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several operations can be removed from these processes.

[0017] Figure 1Schematic flowchart of the autonomous driving vehicle path planning method combined with high-precision perception provided by the embodiments of the present application.

[0018] Figure 2 Schematic structural diagram of the autonomous driving vehicle path planning system combined with high-precision perception provided by the embodiments of the present application.

[0019] Explanation of reference numerals: Multi-modal real-time data acquisition module 10, high-precision environment perception map generation module 20, path generation module 30, final local path acquisition module 40, target vehicle control module 50. Detailed implementation manners

[0020] The above description is only an overview of the technical solution of the present application. In order to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the following specifically describes the specific implementation manners of the present application.

[0021] In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations of the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present application.

[0022] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. The terms "first" and "second" involved are only used to distinguish similar objects and do not represent a specific order for the objects. The terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.

[0023] The embodiments of the present application provide an autonomous driving vehicle path planning method combined with high-precision perception, as Figure 1 shown, the method includes:

[0024] Step S100, acquiring multi-modal real-time data of the surrounding environment of the target vehicle through a sensor module.

[0025] Step S100 further includes step S110. The sensor module includes a high-precision sensor and a redundant sensor group. Among them, the high-precision sensor includes a lidar, a camera, and an ultrasonic sensor; step S120, perform real-time fault detection on the high-precision sensor. If a fault exists in the high-precision sensor, switch to the redundant sensor group and trigger the path safety mode.

[0026] Preferably, the sensor module includes a high-precision sensor and a redundant sensor group. Among them, the high-precision sensor includes a lidar, a camera, and an ultrasonic sensor, which are used to capture high-precision information about the vehicle's surrounding environment, that is, different types of sensors are used to capture different types of information about the vehicle's surrounding environment in real time. Specifically, a lidar (LiDAR) measures distances by emitting laser light and receiving reflected signals, generating 3D point cloud data and providing high-precision environmental depth information; the camera captures information about the texture, color, and shape of the environment, and combines image processing techniques to achieve object detection, tracking, and recognition; the ultrasonic sensor is commonly used for near-field object detection, such as parking assistance and detecting pedestrians, and measures the distance to nearby obstacles by emitting ultrasonic waves and receiving reflected signals; the redundant sensor group serves as a backup for the high-precision sensor. When a fault occurs in the high-precision sensor, the redundant sensor group will take over the sensing task to ensure that the vehicle can continue to drive safely.

[0027] Preferably, the in-vehicle computer of the autonomous vehicle continuously monitors the high-precision sensor to detect whether its working state is normal, that is, perform quality analysis on the sensor data, such as checking the integrity, continuity, and consistency of the data, etc. If it is found that there is a fault in the high-precision sensor (such as abnormal data, signal loss, etc.), it will immediately switch to the redundant sensor group to ensure the continuity of the vehicle's sensing ability, and at the same time trigger the path safety mode. Among them, the path safety mode is an emergency measure to reduce the vehicle's driving speed and increase the safety margin to prevent potential dangers caused by sensing failures. Through the combined use of the high-precision sensor and the redundant sensor group, as well as the implementation of real-time fault detection and redundant switching mechanisms, it can be ensured that the vehicle can maintain accurate sensing ability and a safe driving state in various environments.

[0028] Step S200, perform fusion processing on the multi-modal real-time data to generate a high-precision environmental perception map.

[0029] Preferably, the raw data from each sensor is cleaned and calibrated, including extrinsic calibration and intrinsic optimization, to ensure the consistency, accuracy, and integrity of the data. Useful feature information is extracted from each type of sensor data. For example, image features such as corners and edges are extracted from camera data, and point cloud features are extracted from lidar data. Then, feature matching is performed, that is, features from different sensors or different time points are associated. Next, advanced data fusion algorithms (such as Kalman filtering, particle filtering, etc.) are applied to integrate multi-modal data, including flexibly adjusting weights and fusion strategies according to the characteristics of the sensors and the application scenarios to achieve the optimal fusion effect. Based on the fused multi-modal data, a high-precision environmental perception map is constructed, which not only contains static information such as road structures and obstacle positions but also may contain dynamic information such as vehicles and pedestrians, providing a detailed environmental model for autonomous vehicles and helping the vehicles make accurate path planning and obstacle avoidance decisions. Multi-modal data fusion can make full use of the advantages of different sensors, thus significantly improving the perception accuracy. At the same time, in complex or harsh environments (such as at night, in rainy days, foggy days, etc.), it ensures that the vehicle can maintain accurate perception ability in various environments. In addition, the high-precision environmental perception map is also used for advanced autonomous driving functions (such as automatic parking, autonomous driving navigation, etc.), improving the perception accuracy, flexibility, and robustness of the vehicle.

[0030] Step S200 further includes step S210 of performing spatio-temporal alignment on the real-time data, including mapping the multi-modal real-time data to the BEV space and performing fusion processing on the multi-modal real-time data and the historical data of the target vehicle to capture dynamic environmental changes; step S220 of extracting multiple features of the multi-modal real-time data and performing confidence-weighted fusion on the multiple features; step S230 of generating a high-precision environmental perception map based on the fused features, including road boundaries, obstacle positions, traffic sign, and traffic signal information.

[0031] Preferably, spatio-temporal alignment is performed on the real-time data, that is, the multi-modal real-time data is mapped to the BEV space. Among them, the BEV (Bird's Eye View) space is a representation method that projects the three-dimensional world onto a two-dimensional plane and is commonly used for environmental perception and path planning in autonomous driving. Mapping the multi-modal real-time data (such as lidar point clouds, camera images, ultrasonic sensor data, etc.) to the BEV space can enable the data of different sensors to be identified and processed in a unified coordinate system. Then, the historical data of the target vehicle is obtained, and fusion processing is performed on the multi-modal real-time data and the historical data of the target vehicle to capture dynamic environmental changes. Among them, the historical data may include information such as the past driving trajectory, speed, and acceleration of the vehicle. By fusing the real-time data and the historical data, the current environmental state and the movement trend of dynamic obstacles can be understood more accurately.

[0032] Preferably, useful feature information is extracted from multi-modal real-time data, which may include information such as the density, shape, texture, color, and speed of the point cloud, to more deeply understand the state of the environment and the attributes of dynamic obstacles. Since there may be errors and uncertainties in different sensors and data sources, it is necessary to perform confidence-weighted fusion on the multiple extracted features. Specifically, it involves evaluating the confidence of each feature and weighting the features according to the confidence. Through weighted fusion, the true state of the environment can be more accurately reflected, and the accuracy and reliability of the environmental perception map can be improved. Based on the fused features, a high-precision environmental perception map can be generated, which not only contains static information such as road boundaries and obstacle positions, but may also contain dynamic information such as traffic signs, traffic signals, and predicted trajectories of dynamic obstacles. Specifically, by processing data such as lidar point clouds and camera images, the positions of road boundaries and obstacles can be identified; camera image data can be used to identify information such as traffic signs and traffic signals. Through machine learning algorithms such as the LSTM network, the future trajectories of dynamic obstacles can be predicted and integrated into the high-precision environmental perception map, thereby ensuring the safe driving of autonomous vehicles.

[0033] Further, step S220 further includes step S221, evaluating the confidence of the multiple features; step S222, constructing a feature map of the target vehicle's environmental data based on the confidence; step S223, using the feature map to perform weighted calculation to establish the fused features.

[0034] Preferably, confidence evaluation is used to determine the reliability of each feature. In autonomous driving, due to factors such as sensor noise, environmental changes, and occlusion, the accuracy of different features will vary. Evaluating the confidence of each feature allows appropriate weights to be given in subsequent processing, including but not limited to using models such as Bayesian networks and support vector machines to predict the confidence of features. Then, according to the calculated confidence of the features, a feature map of the target vehicle's environmental data is constructed. Specifically, the original sensor data is cleaned, calibrated, and synchronized, and useful features such as distance, speed, and direction are extracted from the preprocessed data. The confidence of each feature is mapped to the corresponding position on the feature map. Here, the feature map is a data structure used to represent the features and their confidences at different positions in the environment. The feature map usually divides the environment into grids or pixels and assigns one or more feature values and their confidences to each grid or pixel. Weighted calculation is performed using the feature map, that is, each feature value is weighted according to its confidence and fused to establish the fused features, generating a more accurate and reliable representation of the environment to ensure the accuracy of the path planning of autonomous vehicles.

[0035] Further, step S200 further includes step S240. The high-precision environmental perception map further includes a predicted trajectory of dynamic obstacles, where the predicted trajectory of dynamic obstacles is established through an LSTM network.

[0036] Preferably, multi-modal real-time data is collected from a sensor module (such as lidar, camera, etc.), including information such as the position, speed, and acceleration of dynamic obstacles. These data are preprocessed to extract features that can reflect the motion laws and trends of dynamic obstacles, and historical trajectory data is used as a training set to train the LSTM network. During the training process, the network will learn the motion patterns of dynamic obstacles and attempt to predict their future motion trajectories. Some optimization algorithms (such as Adagrad, Adam, etc.) may be used to update the weight parameters of the network to improve the prediction accuracy; the information of dynamic obstacles at the current moment (such as position, speed, etc.) is input into the trained LSTM network, and the predicted trajectory at a future moment is output. By predicting the future trajectory of dynamic obstacles, the autonomous driving vehicle can make an obstacle avoidance decision in advance, thereby avoiding potential collision risks.

[0037] Further, step S240 further includes step S241, obtaining historical motion trajectory data of dynamic obstacles collected by the high-precision sensor; step S242, normalizing the historical motion trajectory data to obtain a standardized data sequence; step S243, training the LSTM network according to the standardized data sequence to establish a dynamic obstacle prediction model; step S244, performing dynamic obstacle prediction based on the dynamic obstacle prediction model, establishing a predicted trajectory of dynamic obstacles, and updating it to the high-precision environmental perception map in real time.

[0038] Preferably, historical motion trajectory data of dynamic obstacles collected by high-precision sensors are obtained, including motion parameters such as their position, speed, and acceleration, which are usually presented in the form of a time series and record the motion trajectories of dynamic obstacles over a past period of time. Since there may be differences in the measurement ranges and accuracies of different sensors and data sources, directly using the original data for model training may lead to poor model performance. Therefore, the historical motion trajectory data are normalized to convert data in different ranges into the same scale, obtaining a standardized data sequence, enabling the model to better learn the internal laws of the data during the training process. The LSTM network is trained according to the standardized data sequence to obtain a dynamic obstacle prediction model, including defining the network structure, setting the loss function and optimization algorithm, and inputting the training data into the network for iterative training. By continuously adjusting the network parameters, the prediction error of the network on the training data gradually decreases. Among them, the LSTM (Long Short-Term Memory) network is a special recurrent neural network that can effectively handle long-term dependencies in time series data. In autonomous driving, the LSTM network is often used to predict the future motion trajectories of dynamic obstacles.

[0039] Preferably, the dynamic obstacle prediction model is used for real-time dynamic obstacle prediction, that is, the motion trajectory data of dynamic obstacles at the current moment and before are input into the prediction model to obtain the predicted trajectories of dynamic obstacles in a future period of time. At the same time, considering the current motion state, road environment, traffic rules, etc. of the dynamic obstacles, a predicted trajectory of dynamic obstacles that conforms to the actual situation is generated and updated to the high-precision environment perception map. For example, the predicted trajectory is fused with information such as static obstacles and road boundaries in the map to generate a comprehensive environment representation containing dynamic and static information, thereby realizing effective prediction and avoidance of dynamic obstacles, planning a safe and efficient driving path, and improving the safety of autonomous driving vehicles.

[0040] Step S300, according to the high-precision environment perception map, combined with the current position and target position of the target vehicle, generate a global path and an initial local path.

[0041] Preferably, the high-precision environmental perception map provides the vehicle with detailed surrounding environmental information, including road structure, obstacle positions, traffic signs and signals, etc. Based on the high-precision environmental perception map, combined with the current position and the target position (i.e., the destination) of the target vehicle, a global path and an initial local path can be generated. Among them, the global path refers to the overall driving route from the current position of the target vehicle to the target position (destination), which is usually planned at the macroscopic level of the high-precision environmental perception map, without considering specific road details and dynamic obstacles. Specifically, the current position of the target vehicle is matched with the high-precision environmental perception map to determine the exact position of the vehicle on the map. Then, a path search algorithm (such as the Dijkstra algorithm, etc.) is used to search for the optimal path from the current position to the target position on the map, considering road connectivity, driving distance, traffic rules, etc. Finally, the searched path is optimized to remove unnecessary detours and redundant points, generating a smoother and more efficient global path.

[0042] Preferably, the initial local path refers to the short-term driving route generated by the target vehicle at the start of driving, based on the global path and the current environmental information. It is usually planned at the microscopic level of the high-precision environmental perception map and needs to consider specific road details and dynamic obstacles. Specifically, high-precision sensors are used to continuously sense the surrounding environmental information, including the positions and speeds of dynamic obstacles, etc. Based on the global path and the current environmental information, a local path planning algorithm (such as the dynamic window method, model predictive control, etc.) is used to generate the initial local path, and factors such as the vehicle's dynamics constraints, road curvature, and obstacle avoidance are considered to generate and smooth the initial local path, removing possible mutation points and jitters to ensure the smoothness and comfort of the vehicle's driving.

[0043] Preferably, the global path provides the vehicle with the overall driving direction and target, while the initial local path refines and adjusts the global path according to the current environmental information. During the actual driving process, the autonomous driving system will continuously update the local path according to the high-precision environmental perception map and real-time sensor data to ensure that the vehicle can safely and efficiently reach the destination. At the same time, if the global path changes (such as due to traffic congestion, road closure, etc.), the autonomous driving system will also re-plan the global path and adjust the local path accordingly, thus ensuring the safe and efficient driving of the autonomous driving vehicle.

[0044] Step S400, dynamically optimize the initial local path based on model predictive control to obtain the final local path.

[0045] Step S400 further includes step S410 of collecting V2X traffic flow data based on multiple communication methods, where V2X includes vehicle-to-vehicle, vehicle-to-infrastructure, vehicle-to-pedestrian, and vehicle-to-network. Among them, the V2X traffic flow data includes local map information and traffic flow data; step S420 of combining the local map information, the traffic flow data, and traffic signal phase information to establish an intersection passing strategy; step S430 of the target vehicle loading the intersection passing strategy in real time to establish the initial local path.

[0046] Preferably, V2X communication includes information interaction between vehicle-to-vehicle (V2V), vehicle-to-infrastructure (V2I), vehicle-to-pedestrian (V2P), and vehicle-to-network (V2N). The information interaction between vehicle-to-vehicle (V2V), vehicle-to-infrastructure (V2I), vehicle-to-pedestrian (V2P), and vehicle-to-network (V2N), that is, the V2X traffic flow data includes local map information and traffic flow data. Among them, the local map information includes intersection, road section, lane information, road structure, traffic signs, and signal positions, etc. The traffic flow data includes the number of vehicles, average speed, and congestion level. Through multiple communication methods such as V2V, V2I, V2P, and V2N, these traffic flow data can be comprehensively collected. For example, a vehicle can share its location, speed, and driving intention information with other vehicles through a wireless communication module. At the same time, the vehicle can also communicate with roadside infrastructure (such as intelligent traffic lights, roadside sensors, etc.) to obtain the real-time traffic conditions at the intersection.

[0047] Preferably, after collecting V2X traffic flow data, it is necessary to combine local map information, traffic flow data, and traffic signal phase information to establish an intersection passing strategy. Specifically, the local map information, traffic flow data, and traffic signal phase information are fused to form fused traffic data. Then, based on the fused data, an intersection passing strategy is formulated, which may include the order and speed limit of autonomous vehicles passing through the intersection. According to the changes in real-time traffic conditions, the intersection passing strategy is dynamically adjusted and optimized. For example, during traffic congestion, the signal timing plan can be adjusted to relieve congestion. The target vehicle receives and loads the intersection passing strategy in real time through its on-vehicle communication system to guide the vehicle's driving behavior at the intersection, including when to accelerate, decelerate, turn, or stop. Finally, based on the loaded intersection passing strategy, the target vehicle can establish an initial local path, that is, the optimal driving route from the current position to the other side of the intersection under the current traffic conditions, considering factors such as vehicle dynamics constraints, road curvature, obstacle avoidance, and traffic signals. During actual driving, the target vehicle will dynamically adjust and optimize the initial local path according to the continuous update of real-time sensor data and the intersection passing strategy. For example, when an obstacle is detected ahead, the vehicle will timely adjust its driving trajectory to avoid the obstacle; when the traffic signal changes, the vehicle will correspondingly adjust its driving speed and direction.

[0048] Furthermore, step S400 also includes continuously monitoring environmental changes and triggering local path replanning when the collision risk exceeds a threshold, where the local path replanning is used to adjust the driving trajectory of the target vehicle in real time.

[0049] Preferably, during vehicle driving, when the sensing system detects changes in the surrounding environment (such as the sudden appearance or movement of other traffic participants, changes in road obstacles, etc.), and after the safety assessment module determines that the collision risk exceeds a preset threshold, the vehicle will immediately activate a local path adjustment mechanism. The local path replanning mainly focuses on the local area near the current position of the vehicle and re-plans the driving trajectory of the vehicle in this local area according to the real-time data provided by the sensor to avoid obstacles or reduce the collision risk, thereby more effectively avoiding potential dangerous situations and improving the safety of vehicle driving.

[0050] Furthermore, step S400 also includes step S440 of using the global path as a reference trajectory for local path optimization; step S450 of dynamically adjusting the local path planning parameters to obtain an optimized local path if the deviation between the initial local path and the global path exceeds the path deviation threshold; step S460 of predicting the position of the target vehicle based on the current state of the target vehicle combined with the dynamics model; and step S470 of adjusting the optimized local path in combination with the position of the target vehicle to obtain the final safe path.

[0051] Preferably, the global path is the overall planned route of the autonomous vehicle from the starting point to the ending point, providing the vehicle with a macroscopic driving direction and goal. Taking the global path as the reference trajectory for local path optimization, when planning the local path, the similarity with the global path will be maintained as much as possible, while considering real-time environmental changes and vehicle dynamics constraints. Specifically, due to the real-time update of the environmental changes and vehicle states, the initial local path may deviate from the global path to a certain extent. A path deviation threshold is set to maintain driving consistency and efficiency. When it is detected that the deviation between the initial local path and the global path exceeds this threshold, the local path planning parameters are dynamically adjusted, including changing the driving direction, speed or acceleration of the vehicle, etc., to obtain an optimized local path; the dynamic model is used to predict the current state of the target vehicle, predicting information such as the position and speed of the vehicle in the future for a period of time, to more accurately evaluate the relative relationship between the vehicle and the surrounding environment, so as to more precisely plan the local path; finally, the optimized local path is adjusted in combination with the position of the target vehicle, including changing the driving trajectory of the vehicle, adjusting the vehicle speed or taking avoidance measures, etc., to ensure that the vehicle can drive along the safest and most efficient path, and then obtain the final safe path, that is, the final local path, to ensure the safety and flexibility of autonomous driving.

[0052] Step S500, controlling the target vehicle according to the global path and the final local path.

[0053] Preferably, by combining the global path and the final local path, the autonomous driving system can achieve precise control of the target vehicle. Specifically, according to the information of the final local path, the steering and speed of the vehicle are controlled to make it drive along the planned path; during driving, if an obstacle or other traffic participants (such as pedestrians) are encountered, the autonomous driving system makes an avoidance decision in a timely manner according to the information of the final local path to ensure the safe driving of the vehicle; according to information such as road conditions, traffic signals and the speeds of surrounding vehicles, the autonomous driving system adjusts the speed of the vehicle to maintain coordination with the surrounding environment; thus ensuring the accuracy of path planning and efficient and safe control, and further enhancing the safety of the vehicle.

[0054] In the above text, reference is made to Figure 1 The method for path planning of an autonomous vehicle combined with high-precision perception according to an embodiment of the present invention is described in detail. Next, reference will be made to Figure 2 Describe the path planning system of an autonomous vehicle combined with high-precision perception according to an embodiment of the present invention.

[0055] An autonomous driving vehicle path planning system combined with high-precision perception according to an embodiment of the present invention is used to solve the technical problems existing in the prior art, such as insufficient environmental perception accuracy, inflexible path planning, and lack of effective prediction of dynamic obstacles, which lead to poor safety and real-time performance of autonomous driving path planning. It achieves the technical effects of improving the safety of autonomous driving, enhancing the real-time performance and adaptability of path planning. As Figure 2 shown, the autonomous driving vehicle path planning system combined with high-precision perception includes: a multi-modal real-time data acquisition module 10, a high-precision environmental perception map generation module 20, a path generation module 30, a final local path acquisition module 40, and a target vehicle control module 50.

[0056] The multi-modal real-time data acquisition module 10 is used to acquire multi-modal real-time data of the environment around the target vehicle through a sensor module; the high-precision environmental perception map generation module 20 is used to perform fusion processing on the multi-modal real-time data to generate a high-precision environmental perception map; the path generation module 30 is used to generate a global path and an initial local path according to the high-precision environmental perception map, in combination with the current position and the target position of the target vehicle; the final local path acquisition module 40 is used to dynamically optimize the initial local path based on model predictive control to obtain a final local path; the target vehicle control module 50 is used to control the target vehicle according to the global path and the final local path.

[0057] Next, the specific configuration of the multi-modal real-time data acquisition module 10 will be described in detail. The multi-modal real-time data acquisition module 10 further includes: acquiring multi-modal real-time data of the environment around the target vehicle through a sensor module; performing fusion processing on the multi-modal real-time data to generate a high-precision environmental perception map; generating a global path and an initial local path according to the high-precision environmental perception map, in combination with the current position and the target position of the target vehicle; dynamically optimizing the initial local path based on model predictive control to obtain a final local path; controlling the target vehicle according to the global path and the final local path.

[0058] Next, the specific configuration of the final local path acquisition module 40 will be described in detail. The final local path acquisition module 40 further includes: collecting V2X traffic flow data based on multiple communication methods, where V2X includes vehicle-to-vehicle, vehicle-to-infrastructure, vehicle-to-pedestrian, and vehicle-to-network, and the V2X traffic flow data includes local map information and traffic flow data; establishing an intersection passing strategy by combining the local map information, the traffic flow data, and traffic signal phase information; the target vehicle loads the intersection passing strategy in real time to establish the initial local path.

[0059] Next, the specific configuration of the final local path acquisition module 40 will be further described in detail. The final local path acquisition module 40 further includes: real-time monitoring of environmental changes, triggering local path replanning when the collision risk exceeds a threshold, where the local path replanning is used to adjust the driving trajectory of the target vehicle in real time.

[0060] Next, the specific configuration of the final local path acquisition module 40 will be further described in detail. The final local path acquisition module 40 further includes: using the global path as a reference trajectory for local path optimization; if the deviation of the initial local path from the global path exceeds a path deviation threshold, dynamically adjusting local path planning parameters to obtain an optimized local path; predicting the position of the target vehicle according to the current state of the target vehicle in combination with a dynamic model; adjusting the optimized local path in combination with the position of the target vehicle to obtain a final safe path.

[0061] Next, the specific configuration of the high-precision environmental perception map generation module 20 will be described in detail. The high-precision environmental perception map generation module 20 further includes: performing spatio-temporal alignment on the real-time data, including mapping the multi-modal real-time data to the BEV space and performing fusion processing on the multi-modal real-time data and the historical data of the target vehicle to capture dynamic environmental changes; extracting multiple features of the multi-modal real-time data and performing confidence-weighted fusion on the multiple features; based on the fused features, processing and generating a high-precision environmental perception map containing road boundary, obstacle position, traffic sign, and traffic signal information.

[0062] Next, the specific configuration of the high-precision environmental perception map generation module 20 will be further described in detail. The high-precision environmental perception map generation module 20 further includes: evaluating the confidence of the multiple features; based on the confidence, constructing a feature map of the environmental data of the target vehicle; using the feature map for weighted calculation to establish the fused features.

[0063] Next, the specific configuration of the high-precision environmental perception map generation module 20 will be further described in detail. The high-precision environmental perception map generation module 20 further includes: the high-precision environmental perception map further includes a predicted trajectory of dynamic obstacles, where the predicted trajectory of dynamic obstacles is established through an LSTM network.

[0064] Next, the specific configuration of the high-precision environmental perception map generation module 20 will be further described in detail. The high-precision environmental perception map generation module 20 further includes: obtaining historical motion trajectory data of dynamic obstacles collected by the high-precision sensors; performing normalization processing on the historical motion trajectory data to obtain a standardized data sequence; training the LSTM network according to the standardized data sequence to establish a dynamic obstacle prediction model; performing dynamic obstacle prediction based on the dynamic obstacle prediction model to establish a dynamic obstacle prediction trajectory, and updating it in real time to the high-precision environmental perception map.

[0065] The automatic driving vehicle path planning system combined with high-precision perception provided by the embodiments of the present invention can execute the automatic driving vehicle path planning method combined with high-precision perception provided by any embodiment of the present invention, and has corresponding functional modules and beneficial effects for executing the method.

[0066] Although various references to certain modules in the system according to the embodiments of the present application are made, any number of different modules can be used and run on the user terminal and / or the server. The included individual units and modules are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0067] The above specific embodiments do not constitute a limitation to the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present application should be included within the protection scope of the present application.

Claims

1. A path planning method for an autonomous driving vehicle combined with high-precision perception, characterized in that: The method comprises: Acquire multi-modal real-time data of the target vehicle's surroundings through sensor modules; Performing fusion processing on the multimodal real-time data to generate a high-precision environment perception map; Generate a global path and an initial local path based on the high-precision environment perception map and in combination with the current position and the target position of the target vehicle; Dynamically optimize the initial local path based on model predictive control to obtain a final local path; The target vehicle is controlled according to the global path and the final local path.

2. The method for path planning of an autonomous driving vehicle combined with high-precision perception according to claim 1, characterized in that: The sensor module includes a high-precision sensor and a redundant sensor group, wherein the high-precision sensor includes a laser radar, a camera and an ultrasonic sensor; Perform real-time fault detection on the high-precision sensor. If the high-precision sensor fails, switch to the redundant sensor group to trigger the path safety mode.

3. The method for path planning of an autonomous driving vehicle combined with high-precision perception according to claim 1, characterized in that: The method of dynamically optimizing the global path and the initial local path based on model predictive control includes: Collect V2X traffic flow data based on multiple communication modes, the V2X including vehicle-to-vehicle, vehicle-to-infrastructure, vehicle-to-pedestrian and vehicle-to-network, wherein the V2X traffic flow data includes local map information and traffic flow data; Establishing a traffic strategy for an intersection by combining the local map information, the traffic flow data and the traffic signal phase information; The target vehicle loads the intersection traffic strategy in real time and establishes the initial local path.

4. The method for autonomous driving vehicle path planning combined with high-precision perception according to claim 3, characterized in that: It also includes real-time monitoring of environmental changes and triggering local path replanning when the collision risk exceeds a threshold, wherein the local path replanning is used to adjust the driving trajectory of the target vehicle in real time.

5. The method for autonomous driving vehicle path planning combined with high-precision perception according to claim 3, characterized in that: Establishing the final local path includes: Using the global path as a reference trajectory for local path optimization; If the deviation between the initial local path and the global path exceeds a path deviation threshold, dynamically adjusting local path planning parameters to obtain an optimized local path; Predict the target vehicle's position based on the current state of the target vehicle combined with the dynamics model; The optimized local path is adjusted in combination with the target vehicle position to obtain a final safe path.

6. The method for path planning of an autonomous driving vehicle combined with high-precision perception according to claim 1, characterized in that: The real-time data is fused and processed to generate a high-precision environment perception map, including: Performing spatiotemporal alignment on the real-time data, including mapping the multimodal real-time data to the BEV space, and fusing the multimodal real-time data with the target vehicle historical data to capture dynamic environmental changes; Extracting multiple features of the multimodal real-time data, and performing confidence-weighted fusion on the multiple features; Based on the fused features, a high-precision environment perception map is generated, which contains road boundaries, obstacle locations, traffic signs and traffic signal information.

7. The method for autonomous driving vehicle path planning combined with high-precision perception according to claim 6, characterized in that: Also includes: evaluating the confidence of the plurality of features; Based on the confidence level, construct a feature map of the target vehicle environment data; The feature map is used to perform weighted calculation to establish fused features.

8. The method for autonomous driving vehicle path planning combined with high-precision perception according to claim 1, characterized in that: The high-precision environment perception map also includes a dynamic obstacle prediction trajectory, wherein the dynamic obstacle prediction trajectory is established through an LSTM network.

9. The method for autonomous driving vehicle path planning combined with high-precision perception according to claim 8, characterized in that: include: Acquire historical motion trajectory data of dynamic obstacles collected by the high-precision sensor; Normalizing the historical motion trajectory data to obtain a standardized data sequence; Training the LSTM network according to the standardized data sequence training to establish a dynamic obstacle prediction model; Dynamic obstacle prediction is performed based on the dynamic obstacle prediction model, a dynamic obstacle prediction trajectory is established, and a high-precision environment perception map is updated in real time.

10. The autonomous driving vehicle path planning system combined with high-precision perception is characterized by: The system is used to implement the path planning method for an autonomous driving vehicle combined with high-precision perception according to any one of claims 1 to 9, and the system comprises: A multi-modal real-time data acquisition module is used to acquire multi-modal real-time data of the surrounding environment of the target vehicle through a sensor module; A high-precision environment perception map generation module, used for fusing the multi-modal real-time data to generate a high-precision environment perception map; A path generation module, used to generate a global path and an initial local path according to the high-precision environment perception map and in combination with the current position and target position of the target vehicle; A final local path acquisition module, used for dynamically optimizing the initial local path based on model predictive control to obtain a final local path; The target vehicle control module is used to control the target vehicle according to the global path and the final local path.

Images