A global driving data adaptive vehicle trajectory planning integrated system

Abstract

A vehicle trajectory planning integration system with full-domain driving data adaptation includes: a multi-source heterogeneous data acquisition module, a full-domain data adaptation engine, a dynamic constraint generator, a hierarchical trajectory planner, and an execution feedback correction unit. The multi-source heterogeneous data acquisition module is connected to the vehicle sensor array via an onboard bus interface to acquire raw state data in real time and output it as multi-protocol data packets. The full-domain data adaptation engine is communicatively connected to the multi-source heterogeneous data acquisition module. This invention achieves preliminary cleaning and feature extraction of raw sensing data before it enters the full-domain data adaptation engine through the coordinated operation of the point cloud preprocessing circuit of the lidar acquisition submodule and the velocity vector calculation circuit of the millimeter-wave radar acquisition submodule within the multi-source heterogeneous data acquisition module. This ensures priority transmission of critical control commands while discarding redundant or conflicting data.

Description

Technical Field

[0001] This invention belongs to the field of driving data planning technology, specifically a vehicle trajectory planning integration system that adapts to full-domain driving data. Background Technology

[0002] Full-domain driving data adaptation and vehicle trajectory planning are underlying technologies in the fields of autonomous driving, intelligent road network collaborative management and control, and in-vehicle intelligent driving control. They rely on the coordinated acquisition of multiple driving data streams from in-vehicle sensing devices, roadside sensing terminals, vehicle chassis bus, and cloud-based road network databases. Through data adaptation, trajectory extrapolation, trajectory optimization, and closed-loop correction processes, they output executable vehicle driving trajectories, adapting to driving operations in all scenarios, including highways, urban roads, and suburban secondary roads. However, current driving data planning faces the following technical challenges: First, existing multi-source vehicle perception data lacks a source-level hierarchical preprocessing mechanism, resulting in a disconnect between perception data and vehicle bus data adaptation. On one hand, the raw radar-type perception data is mixed with a large amount of spatial noise and redundant dimensional data, requiring high-capacity computing equipment at the back end to complete secondary noise reduction and decomposition, leading to high data processing latency. On the other hand, the parallel transmission of multiple bus messages from the vehicle's powertrain, chassis, and body lacks a fixed arbitration priority, resulting in mixed transmission of driving control messages, road condition perception messages, and maintenance messages. This easily leads to message conflicts, delays in critical driving commands, and invalid redundant messages occupying transmission bandwidth, making it impossible to improve the quality of perception data sources and achieve hierarchical transmission of bus commands, thus lacking pre-processing capabilities.

[0003] Secondly, existing vehicle trajectory planning uses fixed safety boundary modeling and a fixed trajectory optimization evaluation mechanism, which has extremely poor adaptability to dynamic driving scenarios. Firstly, traditional solutions rely on static road boundaries and fixed vehicle mechanical thresholds to define the trajectory planning search range, failing to link dynamic obstacle movement states and real-time road condition updates to safety constraints. This results in insufficient safety redundancy in dynamic scenarios involving mixed pedestrian and vehicle traffic and changing traffic flow. Secondly, traditional trajectory optimization only performs basic point interpolation without a curvature adaptive control mechanism. The curvature at trajectory inflection points easily exceeds the physical limits of vehicle steering and braking. Furthermore, the trajectory selection weight parameters are fixed, failing to consider multiple indicators such as traffic efficiency, driving smoothness, and obstacle avoidance safety. Moreover, the lack of a closed-loop adaptive correction link for driving status means that trajectory deviations cannot be dynamically and iteratively corrected, resulting in poor stability during trajectory execution. Summary of the Invention

[0004] In order to overcome the shortcomings of the prior art, the present invention provides a vehicle trajectory planning integration system that adapts to all driving data, so as to at least partially solve the above-mentioned technical problems.

[0005] The technical solution adopted in this invention is as follows: This invention proposes a vehicle trajectory planning integration system with full-domain driving data adaptation, comprising: Multi-source heterogeneous data acquisition module, global data adaptation engine, dynamic constraint generator, hierarchical trajectory planner and execution feedback correction unit; The multi-source heterogeneous data acquisition module is connected to the vehicle sensor array through the vehicle bus interface, and is used to acquire raw status data in real time and output it as multi-protocol data packets. The global data adaptation engine is communicatively connected to the multi-source heterogeneous data acquisition module. It contains a unified timestamp alignment component and a protocol parsing and conversion component, which are used to convert multi-protocol data packets into standard spatiotemporal data streams. The dynamic constraint generator is connected to the global data adaptation engine, extracts environmental feature vectors based on the standard spatiotemporal data stream, and calculates the feasible domain boundary at the current moment in combination with the vehicle dynamics model. The hierarchical trajectory planner is connected to the dynamic constraint generator and the execution feedback correction unit respectively, receives the feasible domain boundary and standard spatiotemporal data stream, and generates the target trajectory sequence using a two-level architecture of global coarse planning and local fine planning. The execution feedback correction unit is connected to the hierarchical trajectory planner and is used to receive the target trajectory sequence and compare it with the actual vehicle response data, and generate an error compensation signal to feed back to the dynamic constraint generator or the hierarchical trajectory planner.

[0006] In one embodiment of the present invention, the multi-source heterogeneous data acquisition module includes a lidar acquisition submodule, a millimeter-wave radar acquisition submodule, a high-precision map reading submodule, and a CAN bus data reading submodule; The lidar acquisition submodule is equipped with a point cloud preprocessing circuit, which is used to perform noise reduction and dimensionality reduction processing on the raw point cloud data before it enters the global data adaptation engine. The millimeter-wave radar acquisition submodule is equipped with a velocity vector calculation circuit, which is used to directly output the radial velocity and azimuth information of the target; The high-precision map reading submodule reads road topology data through a wireless communication network or local storage medium, and encapsulates the road topology data into a structured text format; The CAN bus data reading submodule is configured with multi-channel arbitration logic, which can simultaneously monitor multiple bus messages from the vehicle chassis, power system, and body control system, and discard or retain conflicting messages according to preset priorities.

[0007] In one embodiment of the present invention, the global data adaptation engine includes a clock synchronization unit, a spatial coordinate system transformation unit, and a data cleaning unit; The clock synchronization unit has a hardware trigger signal input terminal, which is used to receive external trigger pulses from each sensor and perform microsecond-level alignment of the sampling time of each sensor data based on the external trigger pulses. The spatial coordinate system transformation unit pre-stores the vehicle coordinate system, the world coordinate system, and the sensor mounting pose parameter matrix, which is used to uniformly transform the raw coordinate data output by different sensors into a unified coordinate system with the vehicle's center of mass as the origin. The data cleaning unit is equipped with an outlier detection algorithm to identify and remove outliers caused by sensor malfunctions or environmental interference, and to perform linear interpolation to complete missing time window data.

[0008] In one embodiment of the present invention, the dynamic constraint generator includes an environment perception fusion unit, a kinematic constraint calculation unit, and a safety buffer layer construction unit; The environmental perception fusion unit uses the obstacle location information in the standard spatiotemporal data stream to predict the future trajectory of the obstacle using a Kalman filter algorithm. The kinematic constraint calculation unit calculates the vehicle's lateral acceleration limit and longitudinal acceleration / deceleration limit based on the current vehicle speed, steering angle, and estimated road friction coefficient. The safety buffer layer construction unit constructs a dynamic polygonal region that changes over time based on the predicted obstacle trajectory and vehicle kinematic constraints. The dynamic polygonal region serves as the search space boundary of the hierarchical trajectory planner, and the closest distance between the edge of the region and the obstacle is not less than a preset safety threshold.

[0009] In one embodiment of the present invention, the hierarchical trajectory planner includes a global path search submodule, a local trajectory optimization submodule, and a cost function evaluation submodule; The global path search submodule generates a global reference path that meets connectivity requirements based on high-precision map data and the current vehicle location using an algorithm. The local trajectory optimization submodule generates multiple candidate trajectory segments based on the global reference path and the feasible region boundary, using a polynomial curve fitting method. The cost function evaluation submodule assigns weights to each candidate trajectory segment, including path length weight, curvature change weight, comfort weight, and distance to obstacles weight, and selects the optimal target trajectory sequence by weighted summation.

[0010] In one embodiment of the present invention, the local trajectory optimization submodule is provided with a trajectory smoothing filter for performing curvature continuity processing on the generated candidate trajectory segments; The smoothing filter employs a cubic spline interpolation algorithm to ensure the continuity of the position and tangent direction of the candidate trajectory segment at the start point, end point, and intermediate nodes. The smoothing filter is also equipped with a maximum curvature limit condition. When the curvature of a certain trajectory exceeds the vehicle's physical limit, the control point coordinates are automatically adjusted to reduce the peak curvature.

[0011] In one embodiment of the present invention, the execution feedback correction unit includes a state observer, an error calculator, and an adaptive gain regulator; The state observer monitors the vehicle's yaw rate, lateral displacement, and longitudinal velocity in real time and compares them with the expected state corresponding to the target trajectory sequence. The error calculator outputs lateral deviation, longitudinal deviation, and heading angle deviation signals based on the comparison results. The adaptive gain regulator dynamically adjusts the proportional gain and integral gain of the feedback control loop according to the magnitude of the lateral and longitudinal deviations. When the deviation is large, the gain is increased to speed up the response, and when the deviation is small, the gain is decreased to avoid system oscillation.

[0012] In one embodiment of the present invention, it further includes a human-computer interaction interface module and a remote cloud collaboration module; The human-machine interface module includes a display screen driving circuit and a speech synthesis module, which are used to display the current planned trajectory, estimated arrival time and system operating status to the driver. The remote cloud collaboration module connects to the cloud server via a cellular mobile communication network to upload anonymized driving data to the cloud and receive traffic congestion warning information and global traffic flow optimization strategies from the cloud. The cloud server is equipped with a large-scale historical trajectory database, which is used to analyze driving behavior patterns on specific road sections and push the analysis results to the global data adaptation engine to assist it in data cleaning and anomaly detection.

[0013] In one embodiment of the present invention, a power management module and a fault diagnosis module are also included; The power management module is equipped with multiple voltage monitoring circuits, which monitor the vehicle battery voltage, sensor power supply voltage and processor power supply voltage respectively, and issue a low voltage alarm signal when the voltage is lower than a set threshold. The fault diagnosis module periodically checks the operating status of each functional module. When it detects that any module in the multi-source heterogeneous data acquisition module, the global data adaptation engine, or the hierarchical trajectory planner has a communication interruption or data overflow, it immediately switches to the degraded operation mode. The degraded operation mode prioritizes the use of the basic kinematic constraint calculation unit and the simplified local trajectory optimization submodule, retaining only the trajectory planning capabilities for straight-line driving or minimum-risk stopping, until the fault is recovered.

[0014] In one embodiment of the present invention, the system is deployed on an embedded industrial computer or an in-vehicle computing platform, wherein the embedded industrial computer or in-vehicle computing platform integrates a graphics processor and a neural network acceleration chip. The graphics processor is used to process the massive point cloud data and image data output by the multi-source heterogeneous data acquisition module in parallel. The neural network acceleration chip is used to run the deep learning model in the dynamic constraint generator to achieve rapid identification and classification of irregular obstacle behavior in complex traffic scenarios. The embedded industrial control computer or vehicle-mounted computing platform is connected to the vehicle chassis actuator via a high-speed serial bus to achieve millisecond-level trajectory command issuance and action feedback.

[0015] The beneficial effects of the technical solution of this invention are as follows: This invention achieves preliminary cleaning and feature extraction of raw sensing data before it enters the global data adaptation engine through the collaborative operation of the point cloud preprocessing circuit of the lidar acquisition submodule and the velocity vector calculation circuit of the millimeter-wave radar acquisition submodule within the multi-source heterogeneous data acquisition module. The point cloud preprocessing circuit performs noise reduction and dimensionality reduction operations at the data source, retaining effective spatial structure information; the velocity vector calculation circuit directly outputs the radial velocity and azimuth of the target, obtaining dynamic target motion parameters without subsequent complex calculations; and the multi-channel arbitration logic filters multiple bus messages from the chassis, power, and body control systems according to preset priorities, ensuring priority transmission of critical control commands while discarding redundant or conflicting data.

[0016] This invention forms a tight data loop between the standard spatiotemporal data stream generated by the global data adaptation engine and the environmental perception fusion unit in the dynamic constraint generator. The environmental perception fusion unit uses the Kalman filter algorithm to process obstacle position information, predict their future trajectory, and inputs the prediction results into the safety buffer layer construction unit. The safety buffer layer construction unit combines the vehicle lateral acceleration limit value and longitudinal acceleration and deceleration limit value output by the kinematic constraint calculation unit to construct a dynamic polygon region that changes with time in real time. This ensures that the search space of the hierarchical trajectory planner is always within the safety boundary. The linkage mode of the dynamic constraint generator defining the safety boundary and the hierarchical trajectory planner optimizing within the boundary ensures that the generated trajectory conforms to the road topology rules and meets the vehicle physical limits and environmental safety requirements, effectively avoiding the problem of insufficient adaptability of traditional fixed boundary planning methods in dynamic scenarios.

[0017] This invention employs a cubic spline interpolation algorithm with a trajectory smoothing filter to process the curvature continuity of candidate trajectory segments, ensuring the continuity of the positions and tangent directions at the start, end, and intermediate nodes. When the curvature of a trajectory segment exceeds the vehicle's physical limits, the smoothing filter automatically adjusts the control point coordinates to reduce the curvature peak, avoiding unexecutable sharp turn commands. The cost function evaluation submodule receives the smoothed candidate trajectories, assigns weights based on path length, curvature change, comfort, and distance to obstacles, and selects the optimal target trajectory sequence through weighted summation. This results in a final target trajectory sequence that is not only mathematically continuous and differentiable but also provides a smooth and natural driving experience, while simultaneously considering traffic efficiency and safety distance.

[0018] This invention utilizes a state observer within a feedback correction unit to monitor yaw rate, lateral displacement, and longitudinal velocity in real time. An error calculator outputs lateral deviation, longitudinal deviation, and heading angle deviation signals. An adaptive gain regulator dynamically adjusts the proportional and integral gain based on the magnitude of the deviation. When the deviation is large, the gain is increased to accelerate the response speed; when the deviation is small, the gain is decreased to avoid system oscillation. The generated error compensation signal is fed back to a dynamic constraint generator or a hierarchical trajectory planner to correct the feasible region boundary or target trajectory sequence for the next cycle.

[0019] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0020] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a module framework diagram of the vehicle trajectory planning and integration system for full-domain driving data adaptation proposed in an embodiment of the present invention; Figure 2 This is a first functional framework diagram of the vehicle trajectory planning and integration system for full-domain driving data adaptation proposed in an embodiment of the present invention. Figure 3 This is a second functional framework diagram of the vehicle trajectory planning integration system for full-domain driving data adaptation proposed in an embodiment of the present invention; Figure 4 This is a third functional framework diagram of the vehicle trajectory planning integration system for full-domain driving data adaptation proposed in this embodiment of the invention. Detailed Implementation

[0021] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0022] The following description, with reference to the accompanying drawings, illustrates an embodiment of the present invention: a vehicle trajectory planning integration system for adapting to full-domain driving data.

[0023] like Figures 1 to 4 As shown, this embodiment of the invention provides a vehicle trajectory planning integration system with full-domain driving data adaptation, including: a multi-source heterogeneous data acquisition module, a full-domain data adaptation engine, a dynamic constraint generator, a hierarchical trajectory planner, and an execution feedback correction unit; The multi-source heterogeneous data acquisition module is connected to the vehicle sensor array via the vehicle bus interface to acquire raw status data in real time and output it as multi-protocol data packets; The global data adaptation engine communicates with the multi-source heterogeneous data acquisition module and contains a unified timestamp alignment component and a protocol parsing and conversion component to convert multi-protocol data packets into standard spatiotemporal data streams. The dynamic constraint generator is connected to the global data adaptation engine, extracts environmental feature vectors based on standard spatiotemporal data streams, and calculates the feasible domain boundary at the current moment by combining the vehicle dynamics model. The hierarchical trajectory planner is connected to the dynamic constraint generator and the execution feedback correction unit respectively. It receives feasible domain boundary and standard spatiotemporal data stream, and generates target trajectory sequence using a two-level architecture of global coarse planning and local fine planning. The execution feedback correction unit is connected to the hierarchical trajectory planner and is used to receive the target trajectory sequence and compare it with the actual vehicle response data to generate an error compensation signal that is fed back to the dynamic constraint generator or the hierarchical trajectory planner.

[0024] In practical applications, this invention's multi-source heterogeneous data acquisition module obtains raw state data from the vehicle's sensor array. Internally, it integrates multiple protocol parsing logics, enabling simultaneous processing of high-frequency data streams from LiDAR, millimeter-wave radar, cameras, and the CAN bus. The data is packaged into multi-protocol data packets with original timestamps. These packets then enter a global data adaptation engine, which uses a unified timestamp alignment component to perform microsecond-level calibration of the physical acquisition time of each data stream, eliminating time deviations caused by transmission delays or differences in sensor response times. Next, the protocol parsing and conversion component converts raw data from different manufacturers and using different protocols into a unified spatiotemporal data stream. This process not only includes coordinate system normalization but also maps point clouds, image features, and vehicle kinematic parameters to the same reference frame, ensuring that data received by subsequent modules is strictly synchronized in time and space.

[0025] Furthermore, the hierarchical trajectory planner receives feasible domain boundaries from the dynamic constraint generator and standard spatiotemporal data streams from the global data adaptation engine, employing a two-tiered architecture of global coarse planning and local fine planning. In the global coarse planning stage, the planner generates a global reference path that meets connectivity requirements based on high-precision map data and the current vehicle position, using an improved path search algorithm. This path primarily focuses on macroscopic reachability and efficiency, ignoring microscopic instantaneous dynamic details. In the local fine planning stage, the planner uses the global reference path as a framework, combined with the real-time feasible domain boundaries provided by the dynamic constraint generator, and generates multiple candidate trajectory segments using a polynomial curve fitting method. These candidate trajectory segments must not only avoid obstacles but also meet multiple performance indicators such as smoothness, comfort, and energy consumption. The cost function evaluation submodule quantifies and scores each candidate trajectory segment, with weights allocated based on path length, rate of curvature change, passenger comfort, and safe distance from obstacles, ultimately selecting the optimal target trajectory sequence.

[0026] Specifically, the state observer within the unit monitors the vehicle's yaw rate, lateral displacement, and longitudinal velocity in real time, comparing them with the desired state corresponding to the target trajectory sequence. The error calculator outputs lateral deviation, longitudinal deviation, and heading angle deviation signals based on the comparison results. These signals are fed into an adaptive gain regulator, which dynamically adjusts the proportional and integral gain of the feedback control loop according to the magnitude of the deviation. When the deviation is large, the gain is increased to accelerate the response speed and quickly correct the vehicle deviation; when the deviation is small, the gain is decreased to avoid system oscillation and ensure smooth driving. The generated error compensation signal is directly fed back to the dynamic constraint generator or hierarchical trajectory planner, enabling the feasible domain boundary or candidate trajectory for the next cycle to be corrected based on the current actual operating conditions, forming a complete closed loop of "perception-decision-control-feedback".

[0027] Furthermore, the multi-source heterogeneous data acquisition module supports the access of various sensor types and can be flexibly combined according to different vehicle configurations. The protocol parsing and conversion component of the full-domain data adaptation engine supports multiple communication protocols and is compatible with data formats of both new and old vehicle models. The vehicle dynamics model of the dynamic constraint generator can be parameter-calibrated according to different vehicle types to adapt to the driving characteristics of different models such as cars and trucks. The cost function weights of the hierarchical trajectory planner can be customized according to different driving styles to meet personalized driving needs. The adaptive gain regulator of the execution feedback correction unit can automatically adjust the control strategy according to road conditions and weather conditions, improving the system's adaptability in complex environments.

[0028] In one specific implementation, the multi-source heterogeneous data acquisition module includes a lidar acquisition submodule, a millimeter-wave radar acquisition submodule, a high-precision map reading submodule, and a CAN bus data reading submodule. The lidar acquisition submodule is equipped with a point cloud preprocessing circuit for denoising and dimensionality reduction of the raw point cloud data before it enters the global data adaptation engine. The millimeter-wave radar acquisition submodule is equipped with a velocity vector calculation circuit for directly outputting the radial velocity and azimuth information of the target. The high-precision map reading submodule reads road topology data through a wireless communication network or local storage medium and encapsulates the road topology data into a structured text format. The CAN bus data reading submodule is equipped with multi-channel arbitration logic, which can simultaneously monitor multiple bus messages from the vehicle chassis, powertrain, and body control systems, and discard or retain conflicting messages according to preset priorities.

[0029] In specific applications, the point cloud preprocessing circuit integrated within the lidar acquisition submodule performs denoising and dimensionality reduction processing on the raw point cloud data before it enters the global data adaptation engine. The circuit uses hardware logic to directly filter out invalid noise points caused by rain, fog, or road surface reflections. Simultaneously, it dynamically adjusts the point cloud density based on distance, preserving key obstacle contour features while eliminating redundant background points, effectively reducing the bandwidth pressure on subsequent data transmission and ensuring that the data stream entering the adaptation engine maintains a high signal-to-noise ratio and manageable volume. The velocity vector calculation circuit configured in the millimeter-wave radar acquisition submodule outputs the target's radial velocity and azimuth information. This circuit performs Doppler frequency shift analysis and angle calculation at the radar signal receiving end, converting the analog signal into digital motion vector data without relying on secondary calculations by backend algorithms. This allows the system to capture the motion trend of high-speed moving objects in real time, providing timely velocity constraints for subsequent trajectory planning.

[0030] Furthermore, the high-precision map reading submodule acquires road topology data through wireless communication networks or local storage media, and encapsulates the extracted road alignment, lane attributes, and traffic sign information into a structured text format. The module supports breakpoint resume and incremental update mechanisms to ensure that the vehicle terminal always holds the latest version of road geometry information. Even in areas with signal obstruction, such as under overpasses or in tunnels, it can maintain positioning accuracy by relying on locally cached data, providing reliable prior knowledge support for global path search. The CAN bus data reading submodule is equipped with multi-channel arbitration logic, capable of simultaneously monitoring multiple bus messages from the vehicle chassis, powertrain, and body control systems. It discards or retains conflicting messages according to preset priorities. The logic determines data validity based on timestamps and message ID weights. When multiple subsystems send the same control command simultaneously, high-priority messages from the chassis control unit are prioritized for retention, preventing the loss or misjudgment of key motion parameters due to excessive bus load, and ensuring the integrity and timeliness of vehicle status data.

[0031] Specifically, the reduced data volume of the point cloud preprocessing circuit alleviates the computational burden on the adaptation engine, the motion parameters output by the velocity vector solution circuit reduce the algorithm inference layers, the high-precision map data in structured text format facilitates rapid retrieval and matching, and the high-reliability CAN messages filtered by multi-channel arbitration logic improve the accuracy of state estimation. This achieves seamless connection from the sensor physical layer to the application logic layer, ensuring that the system can quickly acquire high-quality environmental perception and vehicle state data under complex and ever-changing road conditions, providing a solid data foundation for the dynamic constraint generator to construct feasible domain boundaries.

[0032] Specifically, data interaction between submodules follows a time synchronization protocol. The point cloud data frame header of the LiDAR carries a high-precision time stamp, the velocity vector data of the millimeter-wave radar includes a sampling timestamp, the high-precision map reading submodule records the data version update time, and the CAN bus data reading submodule automatically corrects message transmission delays during arbitration. All data undergoes preliminary time reference calibration before entering the global data adaptation engine. The point cloud preprocessing circuit retains the spatial coordinates of key feature points during dimensionality reduction, the radial velocity and azimuth angle output by the velocity vector calculation circuit are directly mapped to the vehicle coordinate system, the structured text of the high-precision map includes lane centerline curvature and slope information, and the chassis data after CAN bus arbitration includes steering angle, throttle opening, and braking pressure. Together, these data constitute a multi-dimensional spatiotemporal description matrix, eliminating the limitations of a single sensor perspective.

[0033] In one specific implementation, the global data adaptation engine includes a clock synchronization unit, a spatial coordinate system transformation unit, and a data cleaning unit; The clock synchronization unit has a hardware trigger signal input terminal to receive external trigger pulses from each sensor and to perform microsecond-level alignment of the sampling time of each sensor data based on the external trigger pulses. The spatial coordinate system transformation unit pre-stores the vehicle coordinate system, world coordinate system, and sensor mounting pose parameter matrix to uniformly transform the raw coordinate data output by different sensors into a unified coordinate system with the vehicle's center of mass as the origin. The data cleaning unit is equipped with an outlier detection algorithm to identify and remove outliers caused by sensor failures or environmental interference, and to perform linear interpolation to complete the missing time window data.

[0034] In specific applications, the clock synchronization unit of this invention directly connects to the physical trigger ports of external sensors such as LiDAR, millimeter-wave radar, and vision camera through a hardware trigger signal input terminal. It uses high-speed FPGA logic circuits to capture the edges of external trigger pulses emitted by each sensor. When the pulse signal arrives, the clock synchronization unit immediately locks the high-precision count value of the current system master clock and marks the timestamp in the header metadata of the corresponding data packet, ensuring that sensor data from different sampling frequencies and transmission delay characteristics are aligned on the time axis.

[0035] The time-aligned data stream then enters the spatial coordinate system transformation unit. This unit pre-stores the vehicle's centroid coordinate system, the world geographic coordinate system, and the mounting pose parameter matrices of various sensors. The spatial coordinate system transformation unit reads the real-time acquired raw coordinate data and, combined with the pre-stored mounting pose parameters, performs rigid body transformation operations. For the point cloud data output by the LiDAR, the transformation unit rotates and translates it from the radar's local coordinate system to a unified coordinate system with the vehicle's centroid as the origin. Similarly, the target distance and angle information output by the millimeter-wave radar is transformed to three-dimensional Cartesian coordinates in the unified coordinate system. The center point of the target bounding box identified by the camera is also mapped to the same reference system, ensuring that all perceived data is comparable and fusionable within the same spatial framework.

[0036] The data stream, after achieving spatiotemporal alignment and coordinate unification, ultimately flows into the data cleaning unit. This unit is equipped with an outlier detection algorithm based on statistical distribution and physical constraints. The data cleaning unit traverses the data points in the unified coordinate system, calculates the density relationship between each data point and its surrounding neighbors, and compares it with historical movement trends to identify outliers caused by sensor malfunctions, rain / fog interference, or electromagnetic noise. Simultaneously, to address the issue of missing time window data due to communication packet loss or momentary occlusion by some sensors, the data cleaning unit initiates a linear interpolation completion strategy. This strategy fills in the missing values ​​through linear extrapolation based on the valid data sequences before and after the missing time points, maintaining the continuity and integrity of the data stream.

[0037] In one specific implementation, the dynamic constraint generator includes an environment perception fusion unit, a kinematic constraint calculation unit, and a safety buffer layer construction unit; The environmental perception fusion unit uses obstacle location information from standard spatiotemporal data streams and a Kalman filter algorithm to predict the future trajectory of obstacles. The kinematic constraint calculation unit calculates the vehicle's lateral acceleration limit and longitudinal acceleration / deceleration limit based on the current vehicle speed, steering angle, and road friction coefficient. The safety buffer layer construction unit constructs a dynamic polygonal region that changes over time based on the predicted obstacle trajectory and vehicle kinematic constraints. The dynamic polygonal region serves as the search space boundary for the hierarchical trajectory planner, and the closest distance between the edge of the region and the nearest obstacle is not less than a preset safety threshold.

[0038] In a specific application of this invention, the environmental perception fusion unit receives a standard spatiotemporal data stream output from the global data adaptation engine, extracts the obstacle position coordinates, velocity vectors, and category attribute information contained therein, and uses a Kalman filter algorithm to recursively estimate and predict the motion state of multiple targets. The unit establishes an obstacle motion model, smooths the position noise observed at the current moment, and calculates the obstacle's motion path within several future time steps based on historical trajectory trends. The output is a predicted trajectory sequence with confidence intervals, which not only eliminates fluctuations caused by sensor measurement errors but also predicts the lane-changing intentions or acceleration / deceleration behaviors of surrounding vehicles in advance, providing a forward-looking environmental situation input for subsequent constraint calculations.

[0039] Furthermore, the kinematic constraint calculation unit simultaneously acquires the vehicle's current operating state parameters, including instantaneous vehicle speed, front wheel steering angle, and road friction coefficient estimated based on tire slip ratio. These parameters are input into the vehicle dynamics model for solution. Based on the Ackermann steering geometry and longitudinal dynamic equations, the unit calculates the vehicle's lateral acceleration and longitudinal acceleration / deceleration limits in real time under the current operating conditions. The lateral acceleration limit is constrained by the balance between tire grip and centrifugal force. When the vehicle speed is high or the road surface is slippery, the limit value is automatically lowered to prevent the vehicle from skidding and losing control. The longitudinal acceleration / deceleration limits are constrained by the engine torque output capability and braking system efficiency to ensure that the vehicle does not exceed the physical response range of the powertrain system.

[0040] The safety buffer layer construction unit integrates the obstacle prediction trajectory output by the environmental perception fusion unit with the vehicle motion limits generated by the kinematic constraint calculation unit, and performs dynamic synthesis calculations of the spatial region. The unit expands outwards by a preset safety threshold distance from the predicted obstacle trajectory, forming a repulsion zone around the obstacle. Simultaneously, it combines the vehicle's own kinematic constraints to define the vehicle's drivable activity range. Based on this, a dynamic polygonal region that evolves over time is constructed. The vertex coordinates of the polygon are updated in real time as the obstacle moves and the vehicle's state changes, and its edges always maintain a distance from the nearest obstacle that is no less than a preset safety threshold.

[0041] Furthermore, the acceleration limit values ​​provided by the kinematic constraint calculation unit are used to correct the shape of the safety buffer layer. When the vehicle is in a high-speed sharp turn, the reduction in the lateral acceleration limit causes the safety buffer layer to shrink in the lateral direction, forcing the planner to choose a gentler detour route. The dynamic polygonal region output by the safety buffer layer construction unit is directly passed to the hierarchical trajectory planner through the interface as a hard constraint condition for its global coarse planning and local fine planning. Every candidate trajectory generated by the planner must be completely inside the polygon, and any trajectory that crosses the boundary is automatically eliminated.

[0042] In one specific implementation, the hierarchical trajectory planner includes a global path search submodule, a local trajectory optimization submodule, and a cost function evaluation submodule; The global path search submodule generates a global reference path that meets connectivity requirements based on high-precision map data and the current vehicle position. The local trajectory optimization submodule generates multiple candidate trajectory segments based on the global reference path and the feasible region boundary using a polynomial curve fitting method. The cost function evaluation submodule assigns weights to each candidate trajectory segment, including path length weight, curvature change weight, comfort weight, and distance to obstacles weight, and selects the optimal target trajectory sequence by weighted summation.

[0043] The local trajectory optimization submodule has a trajectory smoothing filter inside, which is used to process the curvature continuity of the generated candidate trajectory segments. The smoothing filter uses a cubic spline interpolation algorithm to ensure that the position and tangent direction of the candidate trajectory segments are continuous at the start, end and intermediate nodes. The smoothing filter is also configured with a maximum curvature limit condition. When the curvature of a certain trajectory exceeds the vehicle's physical limit, the coordinates of the control points are automatically adjusted to reduce the curvature peak.

[0044] In a specific application of this invention, the global path search submodule receives high-precision map data and the current vehicle position coordinates from the global data adaptation engine. It then uses an improved path search algorithm to construct a global reference path that meets connectivity requirements within the road topology network. The submodule divides the road into discrete nodes, with the starting point as the root node and the ending point as the target node. It traverses all feasible lane segments, prioritizing path sequences with high traffic efficiency and compliance with traffic rules, generating a medium-to-long-term reference trajectory containing only lane centerline information. Based on the global reference path, the local trajectory optimization submodule, combined with the feasible region boundary output by the dynamic constraint generator, uses a polynomial curve fitting method to generate multiple candidate trajectory segments. The submodule takes the current vehicle state as the starting point and the future target point on the global reference path as the ending point. Within the spatial range defined by the feasible domain boundary, it constructs a variety of motion trajectories using polynomial functions of different orders. The smoothing filter uses a cubic spline interpolation algorithm to ensure that the position and tangent direction of the candidate trajectory segments are continuous at the starting point, ending point and intermediate nodes, eliminating sharp corners and abrupt changes at the trajectory connection points, enabling the vehicle to transition smoothly and avoiding steering shocks or passenger discomfort caused by trajectory discontinuity.

[0045] Furthermore, when the curvature of a certain trajectory exceeds the vehicle's physical limits, the filter automatically adjusts the control point coordinates to reduce the peak curvature, ensuring that the generated trajectory is within the vehicle's steering system's capabilities. While maintaining the overall shape of the trajectory, the filter fine-tunes the control point positions to make the curvature distribution more uniform, avoiding the risk of loss of vehicle control due to localized cornering. The smoothed candidate trajectory segments are then fed into the cost function evaluation submodule. This module assigns multi-dimensional weights to each trajectory, including path length weight, curvature change weight, comfort weight, and obstacle distance weight. The path length weight measures the length of the trajectory, favoring shorter paths to improve efficiency; the curvature change weight focuses on the smoothness of the trajectory, reducing frequent steering wheel movements; the comfort weight is based on the rate of change of longitudinal and lateral acceleration, reducing discomfort caused by rapid acceleration and deceleration; and the obstacle distance weight ensures that the trajectory maintains a sufficient safe distance from surrounding obstacles to prevent collision risks.

[0046] In congested road sections, the weight of distance to obstacles is increased to prioritize safety; in unobstructed road sections, the weight of path length is increased to pursue traffic efficiency. Ultimately, the optimal target trajectory sequence with the highest score is selected. The sequence satisfies both the guidance requirements of the global path and the dynamic constraints of the local environment, while also possessing excellent smoothness and safety. The macro-level guidance provided by the global path search submodule and the micro-level trajectory generated by the local trajectory optimization submodule complement each other, solving the problem that a single planning level cannot simultaneously consider global accessibility and local flexibility.

[0047] In one specific implementation, the feedback correction unit includes a state observer, an error calculator, and an adaptive gain regulator; The state observer monitors the vehicle's yaw rate, lateral displacement, and longitudinal velocity in real time and compares them with the expected state corresponding to the target trajectory sequence. The error calculator outputs lateral deviation, longitudinal deviation, and heading angle deviation signals based on the comparison results. The adaptive gain regulator dynamically adjusts the proportional gain and integral gain of the feedback control loop according to the magnitude of the lateral and longitudinal deviations. When the deviation is large, the gain is increased to speed up the response speed, and when the deviation is small, the gain is decreased to avoid system oscillation.

[0048] In practical applications, the state observer of this invention collects key dynamic parameters of the vehicle, such as yaw rate, lateral displacement, and longitudinal velocity, in real time through an onboard sensor network. These physical quantities are mapped to the vehicle coordinate system and synchronously compared with the desired state corresponding to the target trajectory sequence output by the hierarchical trajectory planner. This process is completed on a microsecond-level timescale, continuously outputting the difference between the vehicle's actual motion state and its ideal motion state at the current moment, providing the raw data foundation for subsequent error analysis.

[0049] Furthermore, the error calculator receives actual state data and target desired state data from the state observer, performs point-by-point subtraction, and generates error signals in three dimensions: lateral deviation, longitudinal deviation, and heading angle deviation. Lateral deviation reflects the vertical distance between the vehicle's centerline and the target trajectory's centerline; longitudinal deviation characterizes the vehicle's relative position lag or lead along the trajectory direction; and heading angle deviation reflects the angle between the vehicle's heading and the trajectory tangent. These three error signals constitute a complete vector describing the degree of vehicle deviation, reflecting the performance state of the current control loop. The error calculator also includes a dead-zone handling module. When the deviations in all three dimensions are less than a preset small threshold, it is determined that the vehicle has entered the steady-state tracking region. At this point, error signal updates are paused to prevent frequent jittering of control commands caused by sensor zero drift or minor road surface undulations.

[0050] Specifically, when the lateral or longitudinal deviation is large, it indicates that the vehicle is deviating significantly from its target trajectory. In this case, the adaptive gain regulator automatically increases the proportional gain to improve the response speed of the control torque, while simultaneously increasing the integral gain to accelerate the elimination of accumulated errors, prompting the vehicle to quickly return to the target trajectory. As the deviation gradually decreases and enters the steady-state region, the regulator gradually reduces the proportional and integral gains to weaken the control effect and avoid system overshoot and oscillation caused by excessive gain.

[0051] The state observer first completes data acquisition and preprocessing, and then transmits the cleaned actual state data to the error calculator. After the error calculator completes the deviation calculation, it inputs the three-dimensional error signal into the adaptive gain regulator. The adaptive gain regulator calculates the current optimal gain combination based on the deviation amplitude by looking up a table or function, and feeds back the generated control parameters to the underlying actuator.

[0052] In the region where the absolute value of the deviation is greater than the first threshold, a large-slope linear growth mode is adopted to ensure rapid convergence; when the absolute value of the deviation is between the first and second thresholds, a smooth-transition nonlinear decay mode is adopted to avoid shocks caused by sudden gain changes; in the region where the absolute value of the deviation is less than the second threshold, a small-slope fine-tuning mode is adopted to maintain the steady state of the system.

[0053] In one specific implementation, it also includes a human-computer interaction interface module and a remote cloud collaboration module; The human-machine interface module includes a display screen driver circuit and a speech synthesis module, which are used to display the current planned trajectory, estimated arrival time and system operating status to the driver; the remote cloud collaboration module connects to the cloud server through a cellular mobile communication network, which is used to upload anonymized driving data to the cloud and receive traffic congestion warning information and global traffic flow optimization strategies issued by the cloud; the cloud server is equipped with a large-scale historical trajectory database, which is used to analyze driving behavior patterns on specific road sections and push the analysis results to the full-domain data adaptation engine to assist it in data cleaning and anomaly detection.

[0054] In specific applications, the display driving circuit integrated within the human-machine interface module of this invention transforms the target trajectory sequence, estimated arrival time, and system operating status generated by the hierarchical trajectory planner into a visual graphical interface, which is then projected onto the vehicle's central control screen in real time. The display driving circuit employs a high refresh rate signal output protocol to ensure that the dynamic trajectory line operates without ghosting or delay in high-speed motion scenarios, allowing the driver to intuitively grasp the vehicle's driving intentions. The voice synthesis module converts key warning information and system status into natural language broadcasts, which are then played through the vehicle's speakers, providing auditory assistance to the driver. Especially in environments where visual attention is distracted or in low-light conditions at night, voice broadcasts can effectively compensate for the lack of visual information and improve the reliability of human-vehicle interaction.

[0055] Furthermore, the remote cloud collaboration module establishes a secure connection with the cloud server via a cellular mobile communication network, enabling bidirectional flow of vehicle-side data and cloud computing power. The module collects anonymized driving data during vehicle operation, including trajectory features, driving behavior patterns, and raw sensor data fragments. After de-identification and compression encoding, this data is uploaded to the cloud server. The cloud server is equipped with a large-scale historical trajectory database, storing massive amounts of driving records of passing vehicles on the same road sections, at different times, and under various weather conditions. Through in-depth mining and analysis of this historical data, the cloud server can identify typical driving behavior patterns on specific road sections, such as traffic patterns in areas with frequent emergency braking, congested intersections, or curves with special curvatures.

[0056] The global traffic flow optimization strategy is generated based on the macroscopic distribution characteristics of regional traffic flow. It guides individual vehicles to integrate into the rhythm of the overall traffic flow while maintaining individual safety, reducing the chain reaction caused by aggressive driving by individual vehicles. After receiving the historical behavior analysis results pushed by the cloud, the global data adaptation engine introduces them as prior knowledge into the data preprocessing process. When abnormal fluctuations that do not conform to the historical typical patterns appear in the real-time sensor data, the engine quickly identifies and removes noise interference based on the behavior model provided by the cloud, retaining the true environmental characteristics, thereby improving data quality and perception reliability.

[0057] In one specific implementation, it also includes a power management module and a fault diagnosis module; The power management module is equipped with multiple voltage monitoring circuits to monitor the vehicle battery voltage, sensor power supply voltage, and processor power supply voltage, and issues a low-voltage alarm signal when the voltage is lower than a set threshold. The fault diagnosis module periodically self-checks the operating status of each functional module. When it detects a communication interruption or data overflow in any of the multi-source heterogeneous data acquisition module, the global data adaptation engine, or the hierarchical trajectory planner, it immediately switches to a degraded operation mode. In the degraded operation mode, the basic kinematic constraint calculation unit and the simplified local trajectory optimization sub-module are enabled first, and only the trajectory planning capabilities for straight driving or minimum risk parking are retained until the fault is recovered.

[0058] In specific applications, when the vehicle battery voltage drops to a critical point, the circuit triggers a low-voltage alarm signal to notify the upper-level system to limit high-power computing tasks. When the sensor power supply voltage is abnormal, the data validity of the corresponding sensor is immediately marked as low priority to prevent data drift caused by unstable power supply. When the processor power supply voltage fluctuates, voltage regulation protection logic is activated to prevent the computing unit from generating logic errors or crashing due to insufficient voltage. During system startup and operation, the fault diagnosis module executes a periodic self-test program, scanning the communication link status and data buffer level of the multi-source heterogeneous data acquisition module, the global data adaptation engine, and the hierarchical trajectory planner functional units. The module confirms the connectivity between modules through a heartbeat detection mechanism, identifies packet loss or garbled characters during transmission using a data packet checksum algorithm, and monitors the memory overflow flag to determine if the data stream exceeds processing capacity. Once any module is detected to have a communication interruption, data overflow, or computation timeout, the fault diagnosis module immediately interrupts the current normal control flow and triggers a switch command for a degraded operation mode.

[0059] In this mode, the kinematic constraint calculation unit estimates the vehicle's limit motion boundaries based solely on the current vehicle speed and steering angle, outputting basic longitudinal speed limits and lateral curvature limits. The simplified local trajectory optimization submodule, based on the above basic constraints and combined with the vehicle's current position and the geometric features of the road ahead, generates a straight-line driving trajectory or a minimum-risk parking trajectory. The straight-line driving trajectory is used to maintain the vehicle's stable forward movement within the lane and is suitable for scenarios where the perception system partially fails but the vehicle still has basic positioning capabilities. The minimum-risk parking trajectory is used to guide the vehicle to decelerate smoothly and pull over to the side of the road when it is no longer safe to drive, thus avoiding secondary accidents.

[0060] When the voltage is detected to be below a set threshold for a certain period of time, the fault diagnosis module determines that there is a power system fault and then initiates the corresponding degradation strategy. Conversely, after switching to degradation operation mode, the fault diagnosis module sends a low-power command to the power management module to shut down unnecessary peripherals and background processes, extend the lifespan of the remaining battery power, and ensure that the vehicle can safely stop or leave a dangerous area using limited power resources.

[0061] In one specific implementation, the system is deployed on an embedded industrial computer or vehicle-mounted computing platform, which integrates a graphics processor and a neural network acceleration chip. The graphics processor is used to process massive point cloud data and image data output by multi-source heterogeneous data acquisition modules in parallel. The neural network acceleration chip is used to run deep learning models in the dynamic constraint generator to achieve rapid identification and classification of irregular obstacle behaviors in complex traffic scenarios. The embedded industrial computer or vehicle-mounted computing platform is connected to the vehicle chassis actuators via a high-speed serial bus to achieve millisecond-level trajectory command issuance and action feedback.

[0062] In specific applications, the graphics processing unit (GPU) of this invention connects to the output of a multi-source heterogeneous data acquisition module via a high-bandwidth memory interface, undertaking the real-time parsing of point cloud data and image data. The point cloud data contains three-dimensional spatial coordinate information of the vehicle's surrounding environment, while the image data provides texture, color, and semantic features. Both data are massive in volume and vary significantly in their original states. The GPU utilizes its massively parallel computing core to perform voxelization segmentation and downsampling on the point cloud data, while simultaneously performing denoising, enhancement, and feature extraction operations on the image data.

[0063] Furthermore, the dynamic constraint generator needs to identify irregular obstacles in complex traffic scenarios, such as pedestrians crossing the road, vehicles changing lanes or cutting in, overturned goods, or abnormal road protrusions. A neural network acceleration chip loads a recognition model trained on a dataset, performs deep analysis of the environmental features preprocessed by the graphics processor, and outputs obstacle category labels, motion vectors, and future trajectory predictions. The model not only identifies static objects but can also infer the intentions of dynamic objects based on historical frame sequences, such as determining whether a pedestrian is about to cross the road or a vehicle is about to change lanes, enabling it to plan safe and efficient avoidance paths in complex traffic flows.

[0064] Furthermore, the embedded industrial control computer or onboard computing platform serves as the hub for data processing and command distribution. It establishes a physical connection with the vehicle chassis actuators via a high-speed serial bus. The bus protocol supports high-throughput and low-latency communication, ensuring that control commands generated by trajectory planning can be sent to the steering, braking, and drive systems at millisecond speeds. The optimal target trajectory sequence output by the hierarchical trajectory planner is converted into specific control parameters, including target steering angle, target acceleration, and target deceleration, and transmitted in real-time to the chassis controller via the high-speed serial bus. Upon receiving the commands, the chassis actuators immediately adjust wheel angles, motor torque, and brake pressure to drive the vehicle along the predetermined trajectory. Simultaneously, the status feedback signals from the chassis actuators, such as actual vehicle speed, actual steering angle, and wheel speed difference, are transmitted back to the computing platform via the same bus for use by the state observer and error calculator.

[0065] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0066] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A vehicle trajectory planning and integration system with full-domain driving data adaptation, characterized in that, include: Multi-source heterogeneous data acquisition module, global data adaptation engine, dynamic constraint generator, hierarchical trajectory planner and execution feedback correction unit; The multi-source heterogeneous data acquisition module is connected to the vehicle sensor array through the vehicle bus interface, and is used to acquire raw status data in real time and output it as multi-protocol data packets. The global data adaptation engine is communicatively connected to the multi-source heterogeneous data acquisition module. It contains a unified timestamp alignment component and a protocol parsing and conversion component, which are used to convert multi-protocol data packets into standard spatiotemporal data streams. The dynamic constraint generator is connected to the global data adaptation engine, extracts environmental feature vectors based on the standard spatiotemporal data stream, and calculates the feasible domain boundary at the current moment in combination with the vehicle dynamics model. The hierarchical trajectory planner is connected to the dynamic constraint generator and the execution feedback correction unit respectively, receives the feasible domain boundary and standard spatiotemporal data stream, and generates the target trajectory sequence using a two-level architecture of global coarse planning and local fine planning. The execution feedback correction unit is connected to the hierarchical trajectory planner and is used to receive the target trajectory sequence and compare it with the actual vehicle response data, and generate an error compensation signal to feed back to the dynamic constraint generator or the hierarchical trajectory planner.

2. The vehicle trajectory planning and integration system for full-domain driving data adaptation according to claim 1, characterized in that, The multi-source heterogeneous data acquisition module includes a lidar acquisition submodule, a millimeter-wave radar acquisition submodule, a high-precision map reading submodule, and a CAN bus data reading submodule. The lidar acquisition submodule is equipped with a point cloud preprocessing circuit, which is used to perform noise reduction and dimensionality reduction processing on the raw point cloud data before it enters the global data adaptation engine. The millimeter-wave radar acquisition submodule is equipped with a velocity vector calculation circuit, which is used to directly output the radial velocity and azimuth information of the target; The high-precision map reading submodule reads road topology data through a wireless communication network or local storage medium, and encapsulates the road topology data into a structured text format; The CAN bus data reading submodule is configured with multi-channel arbitration logic, which can simultaneously monitor multiple bus messages from the vehicle chassis, power system, and body control system, and discard or retain conflicting messages according to preset priorities.

3. The vehicle trajectory planning and integration system for full-domain driving data adaptation according to claim 1, characterized in that, The global data adaptation engine includes a clock synchronization unit, a spatial coordinate system transformation unit, and a data cleaning unit. The clock synchronization unit has a hardware trigger signal input terminal, which is used to receive external trigger pulses from each sensor and perform microsecond-level alignment of the sampling time of each sensor data based on the external trigger pulses. The spatial coordinate system transformation unit pre-stores the vehicle coordinate system, world coordinate system, and sensor installation pose parameter matrix, which is used to uniformly transform the raw coordinate data output by different sensors into a unified coordinate system with the vehicle's center of mass as the origin. The data cleaning unit is equipped with an outlier detection algorithm to identify and remove outliers caused by sensor malfunctions or environmental interference, and to perform linear interpolation to complete missing time window data.

4. The vehicle trajectory planning and integration system for full-domain driving data adaptation according to claim 1, characterized in that, The dynamic constraint generator includes an environment perception fusion unit, a kinematic constraint calculation unit, and a safety buffer layer construction unit; The environmental perception fusion unit uses the obstacle location information in the standard spatiotemporal data stream to predict the future trajectory of the obstacle using a Kalman filter algorithm. The kinematic constraint calculation unit calculates the vehicle's lateral acceleration limit and longitudinal acceleration / deceleration limit based on the current vehicle speed, steering angle, and estimated road friction coefficient. The safety buffer layer construction unit constructs a dynamic polygonal region that changes over time based on the predicted obstacle trajectory and vehicle kinematic constraints. The dynamic polygonal region serves as the search space boundary of the hierarchical trajectory planner, and the closest distance between the edge of the region and the obstacle is not less than a preset safety threshold.

5. The vehicle trajectory planning and integration system for full-domain driving data adaptation according to claim 1, characterized in that, The hierarchical trajectory planner includes a global path search submodule, a local trajectory optimization submodule, and a cost function evaluation submodule; The global path search submodule generates a global reference path that meets connectivity requirements based on high-precision map data and the current vehicle location using an algorithm. The local trajectory optimization submodule generates multiple candidate trajectory segments based on the global reference path and the feasible region boundary, using a polynomial curve fitting method. The cost function evaluation submodule assigns weights to each candidate trajectory segment, including path length weight, curvature change weight, comfort weight, and distance to obstacles weight, and selects the optimal target trajectory sequence by weighted summation.

6. The vehicle trajectory planning and integration system for full-domain driving data adaptation according to claim 5, characterized in that, The local trajectory optimization submodule is equipped with a trajectory smoothing filter, which is used to process the curvature continuity of the generated candidate trajectory segments. The smoothing filter employs a cubic spline interpolation algorithm to ensure the continuity of the position and tangent direction of the candidate trajectory segment at the start point, end point, and intermediate nodes. The smoothing filter is also equipped with a maximum curvature limit condition. When the curvature of a certain trajectory exceeds the vehicle's physical limit, the control point coordinates are automatically adjusted to reduce the peak curvature.

7. The vehicle trajectory planning and integration system for full-domain driving data adaptation according to claim 1, characterized in that, The execution feedback correction unit includes a state observer, an error calculator, and an adaptive gain regulator; The state observer monitors the vehicle's yaw rate, lateral displacement, and longitudinal velocity in real time and compares them with the expected state corresponding to the target trajectory sequence. The error calculator outputs lateral deviation, longitudinal deviation, and heading angle deviation signals based on the comparison results. The adaptive gain regulator dynamically adjusts the proportional gain and integral gain of the feedback control loop according to the magnitude of the lateral and longitudinal deviations. When the deviation is large, the gain is increased to speed up the response, and when the deviation is small, the gain is decreased to avoid system oscillation.

8. The vehicle trajectory planning and integration system for full-domain driving data adaptation according to claim 1, characterized in that, It also includes a human-computer interaction interface module and a remote cloud collaboration module; The human-machine interface module includes a display screen driving circuit and a speech synthesis module, which are used to display the current planned trajectory, estimated arrival time and system operating status to the driver. The remote cloud collaboration module connects to the cloud server via a cellular mobile communication network to upload anonymized driving data to the cloud and receive traffic congestion warning information and global traffic flow optimization strategies from the cloud. The cloud server is equipped with a large-scale historical trajectory database, which is used to analyze driving behavior patterns on specific road sections and push the analysis results to the global data adaptation engine to assist it in data cleaning and anomaly detection.

9. The vehicle trajectory planning and integration system for full-domain driving data adaptation according to claim 1, characterized in that, It also includes a power management module and a fault diagnosis module; The power management module is equipped with multiple voltage monitoring circuits, which monitor the vehicle battery voltage, sensor power supply voltage and processor power supply voltage respectively, and issue a low voltage alarm signal when the voltage is lower than a set threshold. The fault diagnosis module periodically checks the operating status of each functional module. When it detects that any module in the multi-source heterogeneous data acquisition module, the global data adaptation engine, or the hierarchical trajectory planner has a communication interruption or data overflow, it immediately switches to the degraded operation mode. The degraded operation mode prioritizes the use of the basic kinematic constraint calculation unit and the simplified local trajectory optimization submodule, retaining only the trajectory planning capabilities for straight-line driving or minimum-risk stopping, until the fault is recovered.

10. The vehicle trajectory planning and integration system for full-domain driving data adaptation according to claim 1, characterized in that, The system is deployed on an embedded industrial computer or an in-vehicle computing platform, which integrates a graphics processor and a neural network acceleration chip. The graphics processor is used to process the massive point cloud data and image data output by the multi-source heterogeneous data acquisition module in parallel. The neural network acceleration chip is used to run the deep learning model in the dynamic constraint generator to achieve rapid identification and classification of irregular obstacle behavior in complex traffic scenarios. The embedded industrial control computer or vehicle-mounted computing platform is connected to the vehicle chassis actuator via a high-speed serial bus to achieve millisecond-level trajectory command issuance and action feedback.

Images