Personalized man-machine co-driving vehicle lane changing trajectory planning method in complex environment

Abstract

The invention discloses a personalized man-machine co-driving vehicle lane changing trajectory planning method in a complex environment, and belongs to the technical field of intelligent vehicle control. The method comprises the following steps: generating an initial lane changing reference trajectory according to a benchmark personalized cost function; extracting a corresponding theoretical control sequence based on the initial lane changing reference trajectory; performing deviation identification on the input control signal and the theoretical control sequence; resolving a target parameter set capable of minimizing the difference between the theoretical recommendation trajectory and the observation sequence; and performing online correction on the reference personalized cost function by using the target parameter set. According to the method, the deviation between the driver operation and the system planning track is recognized in real time, and the implied driving preference parameters are reversely solved to correct the decision model on line, so that the problem that a personalized driving model is solidified and cannot adapt to the dynamic change state and instant intention of the driver in real time in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent vehicle control technology, and in particular to a method for planning lane-changing trajectories for personalized human-machine co-driving vehicles in complex environments. Background Technology

[0002] With the development of intelligent driving technology, human-machine co-driving has become an important direction for improving driving safety and comfort. In this mode, the key challenge is how to enable the vehicle system to understand and adapt to the personalized preferences and real-time intentions of different drivers, thereby achieving natural and harmonious collaborative control.

[0003] Existing personalized driver assistance systems typically collect and analyze drivers' historical driving data to construct mathematical models reflecting their long-term average driving style (e.g., models based on clustering or parametric style classification), and then generate lane-changing decisions and trajectories accordingly. These systems achieve a degree of personalization by learning drivers' common habits.

[0004] Current technologies primarily rely on summarizing historical average characteristics, resulting in a fixed personalized driving model that cannot adapt to the driver's dynamically changing state and immediate intentions in real time. Because a driver's physiological state, attention level, and driving goals in specific scenarios (such as emergency avoidance or pursuing efficiency) fluctuate in real time, the fixed historical model often deviates from the driver's current expectations. This disconnect directly leads to frequent conflicts between the personalized trajectory planned by the system and the driver's immediate operational intentions, not only reducing the driving experience but also damaging the driver's trust in the system. In the long run, this will cause users to abandon personalized functions, making it difficult for the related technological investment to generate practical benefits. Therefore, there is an urgent need for a personalized planning method that can dynamically perceive and synchronize with the driver's real-time intentions. Summary of the Invention

[0005] This application provides a personalized human-machine co-driving vehicle lane-changing trajectory planning method in complex environments, which solves the problem in the prior art that the personalized driving model is fixed and cannot adapt to the dynamic changes in the driver's state and immediate intentions in real time. It enables the vehicle system to understand and synchronize with the driver's current expectations in real time, just like a tacit co-driver, and generate a lane-changing trajectory that always fits the driver's immediate operating intentions.

[0006] This application provides a personalized lane-changing trajectory planning method for human-machine co-driving vehicles in complex environments, including: acquiring the driver's historical driving data and constructing a baseline personalized cost function, and generating an initial lane-changing reference trajectory based on the baseline personalized cost function; The system collects driver input control signals during vehicle operation in real time and extracts the corresponding theoretical control sequence based on the initial lane-changing reference trajectory. Deviation between the input control signal and the theoretical control sequence is identified. When unplanned behavior is determined to exist, the input control signal is used as the observation sequence to input into the reverse reasoning process. Inverse optimization is performed within the parameter space of the baseline personalized cost function to solve for the target parameter set that minimizes the difference between the theoretically recommended trajectory and the observed sequence; The baseline personalized cost function is corrected online using the target parameter set, and a replanning lane-changing trajectory that matches the driver's real-time intentions is generated by combining the current environmental perception data.

[0007] Furthermore, the steps for obtaining the driver's historical driving data and constructing a baseline personalized cost function include: The vehicle-mounted sensor cluster collects lane-changing behavior data of the target driver within a preset historical period. A baseline personalized cost function is constructed, which is composed of a weighted sum of a safety subfunction, a comfort subfunction, and an efficiency subfunction. The safety sub-function quantifies the collision risk between the vehicle and the road boundary and surrounding vehicles by establishing a potential field model. The comfort sub-function evaluates driving smoothness by calculating the rate of change of the curvature of the trajectory and the longitudinal impact. The efficiency sub-function measures traffic efficiency by calculating the deviation between lane change completion time and expected vehicle speed. Through statistical analysis algorithms, the driver's historical average preference weights for safety, comfort, and efficiency are extracted from the lane-changing behavior data. These historical average preference weights are used as the initial weight vector of the benchmark personalized cost function, thereby constructing a decision benchmark that reflects the driver's long-term driving style.

[0008] Furthermore, the steps of real-time acquisition of driver input control signals during vehicle operation and extraction of the corresponding theoretical control sequence based on the initial lane-changing reference trajectory include: The physical quantities that the driver acts on the vehicle's actuators are continuously read at fixed time steps to form input control signals; Simultaneously acquire the vehicle's current yaw rate, longitudinal velocity, and lateral acceleration motion state parameters to construct a vehicle dynamics model; The initial lane-changing reference trajectory is input into the vehicle dynamics model, and feedforward control simulation calculations are performed to derive the theoretical control sequence. The acquired input control signals and the calculated theoretical control sequence are strictly aligned on the time axis to form a dual-channel data stream containing both measured and theoretical values.

[0009] Furthermore, the steps for identifying the deviation between the input control signal and the theoretical control sequence include: A sliding time window is set, and within each sampling period, a subset of the input control signals and the corresponding subset of the theoretical control sequence within the sliding time window are extracted; Calculate the spatiotemporal trajectory distance between the subset of input control signals and the subset of theoretical control sequences; A dynamic deviation threshold based on the variance of the driver's historical operations is introduced, and the formula for the dynamic deviation threshold is expressed as: ; In the formula, Indicates the dynamic deviation threshold. This represents the expected value of the control deviation under historical normal driving conditions. The standard deviation represents the historical bias. This represents the sensitivity adjustment coefficient; When the calculated spatiotemporal trajectory distance exceeds the dynamic deviation threshold in multiple consecutive sampling periods, it is determined that the driver's current driving expectation has deviated from the historical average style, i.e., it is identified as an unplanned behavior.

[0010] Furthermore, the steps of performing inverse optimization within the parameter space of the baseline personalized cost function to calculate the target parameter set that minimizes the difference between the theoretically recommended trajectory and the observed sequence include: The parameter vector to be optimized in the baseline personalized cost function is defined as follows: This includes weighting coefficients for safety, comfort, and efficiency. Construct an inverse optimization objective function to measure the fit between the predictive control sequence generated under a specific parameter vector and the observed sequence actually executed by the driver; The Bayesian inference method is used to find a set of optimal solutions in the pre-defined parameter vector search space, such that the inverse optimization objective function under the solution achieves the minimum value.

[0011] Furthermore, the specific mathematical implementation process of the reverse optimization is as follows: For each candidate parameter vector within the search space The trajectory generation operator is called to calculate the corresponding simulated trajectory curve. ; The simulated trajectory curve is obtained through an inverse dynamics model. Converted into analog control sequence ; Calculate the simulated control sequence and the observed sequence using the least squares criterion The formula for the sum of squared residuals between them is as follows: ; in, The time length of the observation sequence. For discrete time points; By performing multiple rounds of iterative optimization, we find the solution that... Minimal target parameter set .

[0012] Furthermore, the steps for online correction of the baseline personalized cost function using the objective parameter set include: A weighted fusion algorithm is used to fuse the solved target parameter set with the original baseline parameters to generate instantaneous cost function parameters; The weighted fusion algorithm introduces an intent confidence factor, which dynamically adjusts the fusion ratio of the target parameter set based on the length of the observation sequence and the confidence level in the deviation identification stage. The weight coefficient vector in the baseline personalized cost function is replaced with the parameters of the instantaneous cost function to complete the online reconstruction of the cost function model.

[0013] Furthermore, the steps for generating a replanning lane-changing trajectory that matches the driver's immediate intent, based on current environmental perception data, include: Simultaneously retrieve real-time environmental data from the fusion perception of millimeter-wave radar and cameras to obtain the position, velocity, and acceleration vector of obstacles in the target lane and adjacent lanes; Using the online-corrected baseline personalized cost function as the optimization objective, and with the vehicle's current position, attitude, and speed as the initial boundary conditions, a nonlinear programming problem is constructed. By incorporating road geometry and topology constraints, vehicle maximum lateral acceleration limits, and actuator response delay compensation into the constraints, this nonlinear programming problem is solved in milliseconds using a sequential quadratic programming algorithm. The output series of spatiotemporal path points constitutes the replanning and lane-changing trajectory.

[0014] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: By inferring the driver's intentions in real time and correcting the planning model online, the adaptability problem of personalized driving assistance systems is effectively solved. The continuous deviation between the driver's actions and the expected trajectory is treated as unplanned behavior. New parameters reflecting the driver's current preferences are quickly calculated through inverse optimization, and the trajectory generation strategy is immediately adjusted. This allows the system to shift from relying on historical patterns to following the driver's current intentions in real time, thus solving the problem of model rigidity.

[0015] As a result, system decisions are synchronized with driver expectations, reducing human-machine conflicts caused by the system insisting on its own way. Drivers feel that the system understands and responds promptly to their operational intentions, thereby enhancing their trust in the automated functions and preventing a decline in trust.

[0016] Furthermore, by identifying deviations through dynamic thresholds and fusing parameters using confidence level assessments, the accuracy and robustness of the adjustments are ensured. This avoids frequent or unstable adjustments due to misjudgments, ensuring the consistent reliability of personalized functions. It also prevents function abandonment and resource waste due to poor user experience, guaranteeing the long-term practical value of the system. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a personalized lane-changing trajectory planning method for human-machine co-driving vehicles in complex environments, provided in this application embodiment. Detailed Implementation

[0018] This application provides a personalized human-machine co-driving vehicle lane-changing trajectory planning method in complex environments. This solves the problem in the prior art where the personalized driving model is fixed and cannot adapt to the dynamic changes in the driver's state and immediate intentions in real time. By identifying the deviation between the driver's operation and the system planning in real time and inversely calculating the implicit driving preference parameters to correct the trajectory planning model online, the autonomous driving system realizes the transformation from executing a fixed style to dynamically synchronizing with the driver's real-time intentions, thereby reducing human-machine decision-making conflicts and improving driving trust and system utilization.

[0019] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0020] like Figure 1 The diagram shown is a flowchart of a personalized lane-changing trajectory planning method for human-machine co-driving vehicles in complex environments provided in this application embodiment. The method includes the following steps: acquiring the driver's historical driving data and constructing a benchmark personalized cost function, and generating an initial lane-changing reference trajectory based on the benchmark personalized cost function. The system collects driver input control signals during vehicle operation in real time and extracts the corresponding theoretical control sequence based on the initial lane-changing reference trajectory. Deviation between the input control signal and the theoretical control sequence is identified. When unplanned behavior is determined to exist, the input control signal is used as the observation sequence to input into the reverse reasoning process. Inverse optimization is performed within the parameter space of the baseline personalized cost function to solve for the target parameter set that minimizes the difference between the theoretically recommended trajectory and the observed sequence; The baseline personalized cost function is corrected online using the target parameter set, and a replanning lane-changing trajectory that matches the driver's real-time intentions is generated by combining the current environmental perception data.

[0021] Furthermore, the steps for obtaining the driver's historical driving data and constructing a baseline personalized cost function include: The vehicle-mounted sensor cluster collects lane-changing behavior data of the target driver within a preset historical period. The lane-changing behavior data includes the vehicle's lateral displacement, longitudinal acceleration, and relative distance to surrounding obstacles. A benchmark personalized cost function is constructed using a multi-objective optimization framework. The benchmark personalized cost function is composed of a weighted sum of a safety subfunction, a comfort subfunction, and an efficiency subfunction. The safety sub-function quantifies the collision risk between the vehicle and the road boundary and surrounding vehicles by establishing a potential field model. The comfort sub-function evaluates driving smoothness by calculating the rate of change of the curvature of the trajectory and the longitudinal impact. The efficiency sub-function measures traffic efficiency by calculating the deviation between lane change completion time and expected vehicle speed. Through statistical analysis algorithms, the driver's historical average preference weights for safety, comfort, and efficiency are extracted from the lane-changing behavior data. These historical average preference weights are used as the initial weight vector of the baseline personalized cost function, thereby constructing a decision benchmark that reflects the driver's long-term driving style. Based on this, a smooth initial lane-changing reference trajectory is solved using the minimum principle in the current lane-changing scenario.

[0022] Furthermore, the steps of real-time acquisition of driver input control signals during vehicle operation and extraction of the corresponding theoretical control sequence based on the initial lane-changing reference trajectory include: The high-frequency sampling module continuously reads the physical quantities of the driver's actions on the vehicle's actuators at fixed time steps. These physical quantities include steering wheel angle, accelerator pedal opening, and brake pedal pressure. The above physical quantities are then converted into standard control vector form to form input control signals. Simultaneously acquire the vehicle's current yaw rate, longitudinal velocity, and lateral acceleration motion state parameters to construct a vehicle dynamics model; The initial lane-changing reference trajectory is input into the vehicle dynamics model, and feedforward control simulation calculation is performed to derive the theoretical sequence of control commands that the driver should apply in order for the vehicle to follow the initial lane-changing reference trajectory completely. The collected input control signals and the calculated theoretical control sequence are strictly aligned on the time axis to form a dual-channel data stream containing both measured and theoretical values. This provides basic data support for subsequent deviation quantification analysis and ensures that the system can detect the driver's minor corrections or significant counter-movements to the current automatically planned path.

[0023] Furthermore, the steps for identifying the deviation between the input control signal and the theoretical control sequence include: A sliding time window is set, and within each sampling period, a subset of the input control signals and the corresponding subset of the theoretical control sequence within the sliding time window are extracted; Calculate the spatiotemporal trajectory distance between the subset of input control signals and the subset of theoretical control sequences, wherein the spatiotemporal trajectory distance is measured by a dynamic time warping algorithm or a weighted Euclidean distance formula; A dynamic deviation threshold based on the variance of the driver's historical operations is introduced, and the formula for the dynamic deviation threshold is expressed as: ; In the formula, Indicates the dynamic deviation threshold. This represents the expected value of the control deviation under historical normal driving conditions. The standard deviation represents the historical bias. This represents the sensitivity adjustment coefficient; When the calculated spatiotemporal trajectory distance exceeds the dynamic deviation threshold in multiple consecutive sampling periods, it is determined that the driver's current driving expectation has deviated from the historical average style, that is, it is identified as an unplanned behavior. The input control signal in this period is immediately locked as the observation sequence, and the intent interpretation process is triggered to analyze the driver's real real-time driving needs, so as to avoid the system blindly executing the old logic.

[0024] Furthermore, the steps of performing inverse optimization within the parameter space of the baseline personalized cost function to calculate the target parameter set that minimizes the difference between the theoretically recommended trajectory and the observed sequence include: The parameter vector to be optimized in the baseline personalized cost function is defined as follows: This includes weighting coefficients for safety, comfort, and efficiency. Construct an inverse optimization objective function to measure the fit between the predictive control sequence generated under a specific parameter vector and the observed sequence actually executed by the driver; The Bayesian inference method is used to find a set of optimal solutions in the pre-defined parameter vector search space, so that the inverse optimization objective function under the solution achieves the minimum value; During the optimization process, vehicle dynamics constraints are used as hard boundary conditions to ensure that the searched target parameter set can not only explain the driver's intention, but also that the generated trajectory is physically feasible. The reverse optimization process essentially treats the driver's actual operation as a real-time expression of environmental evaluation standards. Through algorithms, it reverse-engineers the psychological evaluation indicators that can induce the operation, thereby transforming the abstract driving intention into quantifiable parameter indicators, i.e., the target parameter set, and realizing a deep reverse mapping from behavioral characteristics to decision-making logic.

[0025] Furthermore, the specific mathematical implementation process of the reverse optimization is as follows: For each candidate parameter vector within the search space The trajectory generation operator is called to calculate the corresponding simulated trajectory curve. ; The simulated trajectory curve is obtained through an inverse dynamics model. Converted into analog control sequence ; Calculate the simulated control sequence and the observed sequence using the least squares criterion The formula for the sum of squared residuals between them is as follows: ; in, The time length of the observation sequence. For discrete time points; By performing multiple rounds of iterative optimization, we find the solution that... Minimal target parameter set .

[0026] To prevent drastic parameter fluctuations from causing an inconsistent driving experience, a regularization term is introduced during the optimization process to constrain the range of change between the target parameter set and the baseline parameters. This ensures that the parameter evolution process satisfies the immediate intention expression while also taking into account the smooth transition of style. The final target parameter set can accurately depict the trade-offs made by the driver in the current emergency or special psychological state.

[0027] Furthermore, the steps for online correction of the baseline personalized cost function using the objective parameter set include: A weighted fusion algorithm is used to fuse the solved target parameter set with the original baseline parameters to generate instantaneous cost function parameters; The weighted fusion algorithm introduces an intent confidence factor, which dynamically adjusts the fusion ratio of the target parameter set based on the length of the observation sequence and the confidence level in the deviation identification stage. The weight coefficient vector in the baseline personalized cost function is replaced with the instantaneous cost function parameters to complete the online reconstruction of the cost function model; The modified cost function can reflect the driver's latest perception of the current environmental risks in real time. For example, when the driver shows an aggressive acceleration intention, the modified function will significantly reduce the cost penalty of the efficiency subfunction and moderately increase the tolerance for safe distance. This parameter correction mechanism based on operational feedback allows the system to move away from static style labels and instead track the driver's psychological state through dynamic parameter drift, providing precise logical support for generating highly coordinated lane-changing trajectories.

[0028] Furthermore, the steps for generating a replanning lane-changing trajectory that matches the driver's immediate intent, based on current environmental perception data, include: Simultaneously retrieve real-time environmental data from the fusion perception of millimeter-wave radar and cameras to obtain the position, velocity, and acceleration vector of obstacles in the target lane and adjacent lanes; Using the online-corrected baseline personalized cost function as the optimization objective, and with the vehicle's current position, attitude, and speed as the initial boundary conditions, a nonlinear programming problem is constructed. By incorporating road geometry and topology constraints, vehicle maximum lateral acceleration limits, and actuator response delay compensation into the constraints, this nonlinear programming problem is solved in milliseconds using a sequential quadratic programming algorithm. The output series of spatiotemporal path points is the replanning lane-changing trajectory, which deeply integrates the driver's micro-operation cues and the system's safety defense logic at the logical level. The generated replanning lane-changing trajectory has instant responsiveness, which can offset the decision lag caused by the original initial lane-changing reference trajectory, making the vehicle's automated lane-changing action behave as if the driver were personally controlling it, thereby eliminating the control conflict between humans and machines.

[0029] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0030] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0031] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0032] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0033] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0034] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A personalized lane-changing trajectory planning method for human-machine co-driving vehicles in complex environments, characterized in that, Includes the following steps: Acquire the driver's historical driving data and construct a baseline personalized cost function, and generate an initial lane change reference trajectory based on the baseline personalized cost function; The system collects driver input control signals during vehicle operation in real time and extracts the corresponding theoretical control sequence based on the initial lane-changing reference trajectory. Deviation between the input control signal and the theoretical control sequence is identified. When unplanned behavior is determined to exist, the input control signal is used as the observation sequence to input into the reverse reasoning process. Inverse optimization is performed within the parameter space of the baseline personalized cost function to solve for the target parameter set that minimizes the difference between the theoretically recommended trajectory and the observed sequence; The baseline personalized cost function is corrected online using the target parameter set, and a replanning lane-changing trajectory that matches the driver's real-time intentions is generated by combining the current environmental perception data.

2. The personalized lane-changing trajectory planning method for human-machine co-driving vehicles in complex environments as described in claim 1, characterized in that, The steps for obtaining a driver's historical driving data and constructing a baseline personalized cost function include: The vehicle-mounted sensor cluster collects lane-changing behavior data of the target driver within a preset historical period. A baseline personalized cost function is constructed, which is composed of a weighted sum of a safety subfunction, a comfort subfunction, and an efficiency subfunction. The safety sub-function quantifies the collision risk between the vehicle and the road boundary and surrounding vehicles by establishing a potential field model. The comfort sub-function evaluates driving smoothness by calculating the rate of change of the curvature of the trajectory and the longitudinal impact. The efficiency sub-function measures traffic efficiency by calculating the deviation between lane change completion time and expected vehicle speed. Through statistical analysis algorithms, the driver's historical average preference weights for safety, comfort, and efficiency are extracted from the lane-changing behavior data. These historical average preference weights are used as the initial weight vector of the benchmark personalized cost function, thereby constructing a decision benchmark that reflects the driver's long-term driving style.

3. The personalized lane-changing trajectory planning method for human-machine co-driving vehicles in complex environments as described in claim 1, characterized in that, The steps of real-time acquisition of driver input control signals during vehicle operation and extraction of the corresponding theoretical control sequence based on the initial lane-changing reference trajectory include: The physical quantities applied by the driver to the vehicle's actuators are continuously read at fixed time steps to form input control signals; Simultaneously acquire the vehicle's current yaw rate, longitudinal velocity, and lateral acceleration motion state parameters to construct a vehicle dynamics model; The initial lane-changing reference trajectory is input into the vehicle dynamics model, and feedforward control simulation calculations are performed to derive the theoretical control sequence. The acquired input control signals and the calculated theoretical control sequence are strictly aligned on the time axis to form a dual-channel data stream containing both measured and theoretical values.

4. The personalized lane-changing trajectory planning method for human-machine co-driving vehicles in complex environments as described in claim 1, characterized in that, The steps for identifying the deviation between the input control signal and the theoretical control sequence include: A sliding time window is set, and within each sampling period, a subset of the input control signals and the corresponding subset of the theoretical control sequence within the sliding time window are extracted; Calculate the spatiotemporal trajectory distance between the subset of input control signals and the subset of theoretical control sequences; A dynamic deviation threshold based on the variance of the driver's historical operations is introduced, and the formula for the dynamic deviation threshold is expressed as: ； In the formula, Indicates the dynamic deviation threshold. This represents the expected value of the control deviation under historical normal driving conditions. The standard deviation represents the historical bias. This represents the sensitivity adjustment coefficient; When the calculated spatiotemporal trajectory distance exceeds the dynamic deviation threshold in multiple consecutive sampling periods, it is determined that the driver's current driving expectation has deviated from the historical average style, i.e., it is identified as an unplanned behavior.

5. The personalized lane-changing trajectory planning method for human-machine co-driving vehicles in complex environments as described in claim 1, characterized in that, The steps involved in performing inverse optimization within the parameter space of the baseline personalized cost function to calculate the target parameter set that minimizes the difference between the theoretically recommended trajectory and the observed sequence include: The parameter vector to be optimized in the baseline personalized cost function is defined as follows: This includes weighting coefficients for safety, comfort, and efficiency. Construct an inverse optimization objective function to measure the fit between the predictive control sequence generated under a specific parameter vector and the observed sequence actually executed by the driver; The Bayesian inference method is used to find a set of optimal solutions in the pre-defined parameter vector search space, such that the inverse optimization objective function under the solution achieves the minimum value.

6. The personalized lane-changing trajectory planning method for human-machine co-driving vehicles in complex environments as described in claim 1, characterized in that, The specific mathematical implementation process of the reverse optimization is as follows: For each candidate parameter vector within the search space The trajectory generation operator is called to calculate the corresponding simulated trajectory curve. ; The simulated trajectory curve is obtained through an inverse dynamics model. Converted into analog control sequence ; Calculate the simulated control sequence and the observed sequence using the least squares criterion The formula for the sum of squared residuals between them is as follows: ； in, The time length of the observation sequence. For discrete time points; By performing multiple rounds of iterative optimization, we find the solution that... Minimal target parameter set .

7. The personalized lane-changing trajectory planning method for human-machine co-driving vehicles in complex environments as described in claim 1, characterized in that, The steps for online correction of the baseline personalized cost function using the objective parameter set include: A weighted fusion algorithm is used to fuse the solved target parameter set with the original baseline parameters to generate instantaneous cost function parameters; The weighted fusion algorithm introduces an intent confidence factor, which dynamically adjusts the fusion ratio of the target parameter set based on the length of the observation sequence and the confidence level in the deviation identification stage. The weight coefficient vector in the baseline personalized cost function is replaced with the parameters of the instantaneous cost function to complete the online reconstruction of the cost function model.

8. The personalized lane-changing trajectory planning method for human-machine co-driving vehicles in complex environments as described in claim 1, characterized in that, The steps involved in generating a replanning lane-changing trajectory that matches the driver's immediate intent, based on current environmental perception data, include: Simultaneously retrieve real-time environmental data from the fusion perception of millimeter-wave radar and cameras to obtain the position, velocity, and acceleration vector of obstacles in the target lane and adjacent lanes; Using the online-corrected baseline personalized cost function as the optimization objective, and with the vehicle's current position, attitude, and speed as the initial boundary conditions, a nonlinear programming problem is constructed. By incorporating road geometry and topology constraints, vehicle maximum lateral acceleration limits, and actuator response delay compensation into the constraints, this nonlinear programming problem is solved in milliseconds using a sequential quadratic programming algorithm. The output series of spatiotemporal path points constitutes the replanning and lane-changing trajectory.

Images