Double-level decision-making method of urban expressway autonomous lane changing system

Abstract

The invention discloses a double-level decision-making method of an urban expressway autonomous lane changing system, which comprises an upper-level decision-making method and a lower-level decision-making method, the upper-level decision-making method is used for calculating lane passing efficiency, and the lower-level decision-making method is used for analyzing lane changing safety. According to the method, a closed-loop behavior decision-making system of lane keeping-lane changing preparation-lane changing execution is constructed, the driving behavior of the intelligent vehicle can be described more accurately, and a clearer framework is provided for subsequent decision-making control. According to the method, the lane passing efficiency driving strategy is adopted for the autonomous lane changing intention, quantification and modeling are carried out by referring to the psychological mechanism that a human driver generates the lane changing intention when encountering a low-speed front vehicle, the decision logic under the actual driving situation is better met, and the intelligent vehicle lane changing decision is more reasonable and humanized. The invention provides a depreciation function calculation method, the influence of the traffic condition on the vehicle driving efficiency can be meticulously measured from the lane level, and a more scientific and more targeted basis is provided for intelligent vehicle lane changing decision making.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle auxiliary control, and in particular to a dual-level decision-making method for an autonomous lane-changing system on an urban expressway. Background Art

[0002] The current mainstream behavioral decision-making methods for intelligent vehicles can be roughly divided into the following two typical paradigms:

[0003] The first category is data-driven approaches, typified by end-to-end deep learning models. Supported by high-quality, large-scale datasets, these approaches are able to generate human-like driving trajectories in complex and changing traffic environments, demonstrating strong environmental adaptability. However, their performance is highly dependent on the coverage and sample quality of the training data, and the model's internal decision-making mechanisms lack transparency and interpretability, making them difficult to meet in autonomous driving scenarios with extremely high system safety requirements. Furthermore, when addressing low-frequency, long-tail scenarios, these approaches suffer from insufficient generalization and limited behavioral controllability.

[0004] The second category is rule-based decision-making methods. Their core advantages lie in their clear logical structure, strong behavioral traceability, and the ability to clearly define safety boundaries and ensure behavioral controllability through formal modeling. These methods, provided they are well-designed, offer high verifiability and stability. However, as traffic scenarios become more complex, the cost of building and maintaining rule-based systems rises significantly, leading to rule base expansion and decreased reasoning efficiency.

[0005] In the specific application context of urban expressways, the problem boundaries are relatively clear due to the highly structured nature of the traffic environment, which typically does not involve unstructured obstacles such as pedestrians and non-motorized vehicles, and rarely features complex road configurations such as irregular intersections. Based on reasonable abstraction and modeling of this scenario, it is possible to construct a complete and consistent state space description system. Under this premise, the use of rule-based decision-making methods is more conducive to achieving stable, highly interpretable, and easily verifiable decision outputs, which has high engineering feasibility and practical deployment value.

[0006] At present, the autonomous lane change decision-making method in multi-lane scenarios on urban expressways, based on rule-based decision-making, has the following shortcomings:

[0007] Most existing methods for intelligent vehicle driving state classification are relatively crude and lack detailed classification. This makes it difficult to accurately characterize the vehicle's driving characteristics and behavioral transitions at different stages. This leads to unclear logic when building behavioral decision-making systems, making it difficult to fully and accurately describe the driving process of intelligent vehicles in complex multi-lane scenarios.

[0008] Traditional methods for generating lane change intentions rely solely on simple factors like inter-vehicle distance and speed differential, failing to fully consider the psychological decision-making mechanisms of human drivers. This disconnect from the complex psychological processes of humans, such as the desire to change lanes when encountering a slower vehicle ahead due to unsatisfactory speeds, results in a lack of rationality and human considerations in intelligent vehicle lane change decisions. In real-world urban expressway scenarios, these decisions may not align with the habits of actual traffic participants, impacting traffic flow and safety.

[0009] Existing decision-making methods often use conventional traffic parameters (such as traffic volume and average speed) as decision variables, lacking in-depth analysis of lane-level traffic conditions. This inability to accurately measure the impact of traffic conditions on vehicle efficiency at the microscopic level of lanes results in decision variables that fail to fully and meticulously reflect actual lane capacity differences, making it difficult for intelligent vehicles to make scientific and accurate lane change decisions. Summary of the Invention

[0010] To address the aforementioned issues with existing technologies, this invention aims to provide a two-tiered decision-making approach for an autonomous lane change system on urban expressways. By meticulously categorizing vehicle driving states, a closed-loop behavioral decision-making system is constructed to accurately describe driving behavior. Based on human driving psychology, a lane efficiency-driven strategy is employed to determine lane change intentions, making decision-making more rational and human-like. A loss function calculation method is proposed, using the adjacent lane efficiency index as a decision variable to accurately measure lane efficiency and enhance the scientificity and accuracy of decision-making.

[0011] In order to achieve the above-mentioned object, the technical solution of the present invention is as follows: a two-level decision-making method for an autonomous lane-changing system on an urban expressway, wherein the autonomous lane-changing system is installed in a controller of the vehicle.

[0012] The vehicle is equipped with an autonomous lane-changing system in a multi-lane traffic scenario on an urban expressway. The driving states of the vehicle on the urban expressway are defined as including lane centering state, preparing to change lanes to the left, preparing to change lanes to the right, executing left lane change state, and executing right lane change state.

[0013] The lane centering state of the ego vehicle is defined as the basic state of the ego vehicle. The lane in which the ego vehicle is traveling is defined as the driving lane, the adjacent lane to the left of the driving lane is defined as the left lane, and the adjacent lane to the right of the driving lane is defined as the right lane.

[0014] The dual-level decision-making method includes an upper-level decision-making method and a lower-level decision-making method. The upper-level decision-making method calculates lane efficiency, and the lower-level decision-making method analyzes lane change safety. Specifically, the method includes the following steps:

[0015] A. The autonomous lane change system controls the vehicle's driving state to a center-maintaining state.

[0016] B. The autonomous lane change system determines whether the vehicle is blocked by a slow vehicle ahead. If so, go to step C; otherwise, go to step A.

[0017] C. The autonomous lane change system calculates lane efficiency indices for the left lane, the driving lane, and the right lane, and compares them to determine the preferred lane. If the left lane efficiency index is higher, proceed to step D. If the right lane efficiency index is higher, proceed to step E. If the efficiency indices for both lanes are the same and higher than the driving lane efficiency index, proceed to step D. If the driving lane efficiency index is higher, proceed to step A.

[0018] D. Generate a left lane change motivation and prepare to execute the left lane change. The autonomous lane change system generates a longitudinal occupancy map of the target lane and evaluates whether the current lane change opportunity allows it. If so, the left lane change is executed. After the lane change is complete, the system proceeds to step A. Otherwise, the vehicle remains in a left-hand lane preparation state until a suitable lane change opportunity is found. If a suitable lane change opportunity cannot be found within 5 seconds, the system proceeds to step A.

[0019] E. Generate a right lane change motivation and prepare to execute the right lane change. The autonomous lane change system generates a longitudinal occupancy map of the target lane and evaluates whether the current lane change opportunity allows it. If so, the right lane change is executed. After the lane change is complete, the system proceeds to step A. Otherwise, the vehicle remains in a right-hand lane preparation state until a suitable lane change opportunity is found. If a suitable lane change opportunity cannot be found within 5 seconds, the system proceeds to step A.

[0020] Furthermore, the method for calculating lane traffic efficiency is as follows:

[0021] Cruising and following represent two opposing driving states in road traffic. The unobstructed state is called the free cruising state. Using the free cruising state as a baseline, we quantitatively analyze the mileage loss caused by following behavior under different traffic conditions. Based on this, we propose the concept of a "virtual cruising trajectory." Specifically, regardless of whether there are slower vehicles ahead that obstruct the ego vehicle's cruising, assume that the ego vehicle starts from its current speed and acceleration, adjusts to a preset cruising speed, and then maintains a constant speed. The resulting longitudinal displacement sequence over a specific future time interval is defined as the virtual cruising trajectory.

[0022] Assume that the computational time domain of the virtual cruise trajectory is t∈[0,t2], within which there are variable acceleration and uniform speed phases. Assume that the speed regulation time range of the variable acceleration phase is t∈[0,t1], and the speed regulation task ends at time t1. The virtual cruise trajectory within the time ranges of the variable acceleration phase [0,t1] and the uniform speed phase [t1,t2] is:

[0023]

[0024] Among them, x vir1 (t) Trajectory of strain acceleration stage, x vir2 (t) corresponds to the trajectory of the uniform motion phase, and:

[0025] x vir1 (t) = A1X

[0026]

[0027] X=[tt 2 t 3 ] T

[0028] x vir2 (t) = v2(t-t1) + x vir1 (t)

[0029] Where a jerk is the set jerk, i.e., the time derivative of acceleration, a0, v0, and v2 are the initial acceleration, initial vehicle speed, and uniform vehicle speed, respectively. A1 is the coefficient matrix, and X is the state variable matrix.

[0030] The displacement time series of the standard distance following vehicle is called the standard following trajectory:

[0031] x fol (t) = x f (t)-D

[0032] Where x f (t) is the longitudinal displacement trajectory of the preceding vehicle in the lane at time t in the prediction time domain, and D is the longitudinal distance increment. The upper bound of the standard following trajectory is x fol,max (t i ), the lower bound is x fol,min (t i The standard following trajectory is determined only by the effective perception range and prediction time domain of the vehicle. The longitudinal distance increment D is determined by the longitudinal length of the vehicle l self and the longitudinal length of the preceding vehicle l i The calculation formula is as follows:

[0033]

[0034] Taking the virtual cruise trajectory as the benchmark, the longitudinal mileage loss of each lane under the standard following trajectory is calculated, and the loss function is defined as:

[0035]

[0036] in, k ti is the compensation function, which is positively correlated with time. i k corresponding to the moment ti =mt i , where m is a constant greater than 0 that determines the rate at which the compensation function increases over time; larger m indicates faster growth. The penalty function describes the degree of hindrance experienced by the ego vehicle during lane following over a period of time. From the driver's perspective, this means the expected longitudinal distance does not meet expectations.

[0037] The lower bound of the loss function is:

[0038]

[0039] The upper bound of the loss function is:

[0040]

[0041] Since the virtual cruise trajectory serves as a unified reference for all lanes, it depends only on the initial state of the ego vehicle and the predicted duration at the current moment, and is independent of the lane selected. Furthermore, the upper and lower bounds of the standard following trajectory are independent of the specific lane selected, but are instead determined by the vehicle's effective perception range and the predicted time domain. Therefore, at any given moment, the loss function corresponding to each lane has the same upper and lower bounds. Accordingly, the loss function for each lane is normalized to the [0,1] interval based on its upper and lower bounds, and the loss index is defined as:

[0042] Where a = left, mid, right

[0043] Among them, the damage index of the lane where the vehicle is traveling is expressed as N mid If the target lane is on the left side of the driving lane, the lane loss index is expressed as N left If the target lane is on the right side of the driving lane, the lane loss index is expressed as N right express.

[0044] The left lane traffic efficiency index is defined as:

[0045] η left =N mid -N left

[0046] The right lane traffic efficiency index is defined as:

[0047] η right =N mid -N right

[0048] The left and right adjacent lane efficiency index represents the relative degree of congestion in the target lane relative to the driving lane, and its value range is [-1, 1]. A negative value indicates that the target lane is more congested than the current lane, and the ego vehicle's autonomous lane change system will not be motivated to change lanes. A positive value indicates that the target lane is more unobstructed, and the ego vehicle's autonomous lane change system will be motivated to change lanes. A value of 0 indicates that the target lane has no difference in efficiency compared to the driving lane, and the ego vehicle's autonomous lane change system will not be motivated to change lanes. Furthermore, a larger absolute value of the index indicates a more significant degree of congestion or unobstructed traffic. This allows for a quantitative characterization and representation of the efficiency of adjacent lanes.

[0049] Furthermore, the switching between the driving states of the vehicle follows the following logic:

[0050] A1. When the vehicle ahead in the driving lane slightly blocks the ego vehicle, the ego vehicle's autonomous lane change system executes the upper-level decision-making method, generating a lane change motivation. The ego vehicle adjusts from maintaining lane center to moving closer to the lane with higher traffic efficiency, moving to the left or right of the ego vehicle's lane, entering the lane change preparation state. Simultaneously, the ego vehicle activates its turn signal, ready to change lanes to the adjacent target lane. The lane change preparation state is divided into two states: preparing to change left and preparing to change right, corresponding to the ego vehicle moving to the left and right of the driving lane, respectively.

[0051] A2. If the traffic efficiency of the target lane decreases significantly after entering the lane change preparation state, the lane change motivation should be canceled, the turn signal should be turned off, and the lane should be returned to the center position to avoid the risks of unfavorable lane changes.

[0052] A3. When the vehicle ahead in the driving lane significantly blocks the ego vehicle, the ego vehicle's autonomous lane change system executes the upper-level decision-making method, generating a lane change motivation. The ego vehicle activates its turn signal within the driving lane and enters the lane change preparation state. The lower-level decision-making method assesses whether the current window of opportunity exists to ensure the safety of the entire lane change process. If the safe lane change conditions are met, the ego vehicle switches from the lane change preparation state to the lane change execution state.

[0053] A4: Once the vehicle has completely entered the target lane, the lane change operation is considered complete, and the vehicle's autonomous lane change system switches to a centering state within the target lane.

[0054] A5. Whether in lane centering mode or preparing to change lanes, the vehicle must maintain lane centering for a certain period of time before switching to another state to ensure the stability of behavioral decisions and driving safety.

[0055] Furthermore, the lower-level decision-making method incorporates a corresponding decision-making mechanism that performs real-time analysis and assessment of the future driving trends of other vehicles in the target lane and their relative motion relationships with the ego vehicle when the ego vehicle is preparing to change lanes. This decision-making is based on the generation of a longitudinal occupancy map of the target lane and analysis of the optimal lane change timing.

[0056] The target lane longitudinal occupancy map is defined as the set of longitudinal spatiotemporal areas occupied by all obstacles in a given lane over a period of time in the future, in a scenario where the vehicle is simplified as a point mass.

[0057] The steps for generating the target lane longitudinal occupancy map are as follows:

[0058] B1. Predict the future driving path of the target lane vehicle based on its existing motion parameters.

[0059] B2. In the longitudinal and lateral dimensions, based on collision geometry principles and appropriate expansion of the target lane vehicle size as needed, construct graphs of the longitudinal position and lateral position of the target lane vehicle over time.

[0060] B3. Determine the time series of target lane occupancy by analyzing the graph of lateral position changes over time.

[0061] B4. Based on the obtained target lane occupancy time series, extract relevant information from the longitudinal position change over time graph within the corresponding time period, thereby generating a target lane longitudinal occupancy map.

[0062] B5. After obtaining the target lane longitudinal occupancy map, define the appropriate lane change opportunity as the segment on the time axis of the target lane longitudinal occupancy map corresponding to when the ego vehicle is in the lane change preparation state and the target lane is idle. If multiple appropriate lane change opportunities exist, analyze and select the optimal one, which will serve as the basis for the autonomous lane change system to execute the next-level decision-making method.

[0063] The analysis method of the optimal lane change timing is as follows:

[0064] The desired trajectory of the ego vehicle in the lane change preparation state is plotted into the target lane's longitudinal occupancy map. The intersection of the desired trajectory and the occupied area in the target lane's longitudinal occupancy map is used to determine the lane change opportunity. Analysis reveals that overlap between the desired trajectory and the occupied area indicates an unfeasible area conflict between the ego vehicle and the target lane, preventing the lane change. Non-overlapping areas indicate no area conflict in the target lane, and the ego vehicle will execute the lane change.

[0065] Since the lane change action of the ego vehicle takes a certain amount of time, the safety threshold of the lane change time is defined as T s, the time windows that meet the lane change timing requirements in the target lane longitudinal occupancy diagram are T r , where r = 1, 2, 3...

[0066] When a certain time window T r >Safety threshold T s When , it means that the vehicle has the conditions to safely execute the lane change decision within this time window, and this time is the appropriate lane change opportunity. During the lane change process, the smaller the lateral acceleration of the vehicle, the smoother the lane change action and the higher the comfort. On the basis of meeting safety requirements, the time window with the highest comfort is the optimal time window, and the corresponding time is the optimal lane change opportunity. In addition, different safety thresholds T are set according to actual conditions. s , to meet diverse driving needs.

[0067] When two time windows with similar lateral acceleration and no essential difference in safety and comfort are present during a lane change, the window with the closer start time is selected to complete the lane change as quickly as possible and ensure traffic efficiency. If an optimal time window has already begun and the vehicle is currently within it, that optimal time window is abandoned and the search continues for a subsequent optimal time window that has not yet begun to ensure the safe completion of the lane change process.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] 1. Detailed and logically coherent state classification: This invention subdivides the driving states of intelligent vehicles on urban expressways into five categories: lane centering, preparing for a left lane change, preparing for a right lane change, executing a left lane change, and executing a right lane change. This establishes a closed-loop behavioral decision-making system for lane keeping, preparing for a lane change, and executing a lane change. Compared to the more general state classification used in some studies, this detailed and logically coherent state setting more accurately describes intelligent vehicle driving behavior and provides a clearer framework for subsequent decision-making and control.

[0070] 2. Lane Change Intention Strategy Adapted to Human Driving Logic: This invention's autonomous lane change intention strategy utilizes a lane efficiency-driven strategy, quantifying and modeling the psychological mechanisms that trigger lane change intentions in human drivers when encountering a slow-moving vehicle ahead. Unlike existing research that relies solely on simple factors like distance and speed, this strategy better aligns with the decision-making logic of actual driving situations, making lane change decisions for intelligent vehicles more rational and user-friendly.

[0071] 3. Innovation in Decision Variable Calculation: This paper proposes a loss function calculation method, based on which the adjacent lane efficiency index is derived as the decision variable for lane switching. Unlike conventional traffic parameters that may be used in existing research, this innovative calculation method can more meticulously measure the impact of traffic conditions on vehicle efficiency at the lane level, providing a more scientific and targeted basis for intelligent vehicle lane change decisions, thereby improving the accuracy and effectiveness of decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a schematic flow diagram of the present invention. DETAILED DESCRIPTION

[0073] The present invention will be further described below with reference to the accompanying drawings.

[0074] like Figure 1 As shown in the figure, in a multi-lane urban expressway scenario, when a vehicle encounters a slow vehicle ahead, the efficiency of the current lane may be lower than that of the left or right lanes. The vehicle's autonomous lane change system calculates the lane efficiency index and compares it to determine the optimal lane change, deciding whether to change lanes left or right, or remain in the center.

[0075] When the autonomous lane change system calculates the lane efficiency index and indicates that the left lane is more efficient, it will initiate a left lane change and prepare to execute the left lane change. The vehicle will then pull left within its current lane and activate its left turn signal to alert surrounding vehicles of its impending lane change. The autonomous lane change system then generates a longitudinal occupancy map of the target lane and assesses whether the current lane change opportunity is permitted. If the assessment indicates safety and comfort, the lane change opportunity is considered permitted.

[0076] If the assessment indicates that the lane change is permitted, the vehicle will execute the left lane change while ensuring safety. After completing the lane change, the vehicle will maintain lane centering and continue driving in the new lane.

[0077] If the assessment result is that the lane change is not allowed: the vehicle will maintain a left-hand driving standby state until a suitable lane change opportunity is found. If a suitable lane change opportunity cannot be found within 5 seconds, the vehicle will return to the lane centering state and continue driving in the current lane.

[0078] When the autonomous lane change system calculates the lane efficiency index and indicates that the right lane is more efficient, it will initiate a lane change attempt and prepare to proceed. The vehicle will then pull right within its current lane and activate its right turn signal to alert surrounding vehicles of its impending lane change. The autonomous lane change system then generates a longitudinal occupancy map of the target lane and assesses whether the lane change opportunity is permitted. If the assessment indicates safety and comfort, the lane change opportunity is considered permitted.

[0079] If the assessment indicates that the lane change is permitted, the vehicle will execute the right lane change while ensuring safety. After completing the lane change, the vehicle will maintain lane centering and continue driving in the new lane.

[0080] If the assessment result is that the lane change is not allowed: the vehicle will maintain a ready state of driving right until a suitable lane change opportunity is found. If a suitable opportunity cannot be found within 5 seconds, the vehicle will return to the lane centering state and continue driving in the current lane.

[0081] In some cases, if the traffic efficiency indexes of both lanes are the same and lower than the current lane, the vehicle's autonomous lane change system will not generate a lane change motivation, and the vehicle will continue to stay centered in the current lane to avoid the potential risks of unnecessary lane changes. If the traffic efficiency indexes of both lanes are the same and higher than the current lane, the autonomous lane change system's lane change motivation is to prepare for a left lane change, and subsequent judgment rules are the same as if the left lane has higher traffic efficiency.

[0082] During the entire lane-changing process, the autonomous lane-changing system will continuously monitor changes in the surrounding environment and dynamically adjust the lane-changing strategy to ensure that the vehicle can drive safely and efficiently in complex multi-lane traffic scenarios on urban expressways.

[0083] At this point, a closed-loop behavioral decision-making process of lane centering, lane change preparation, and lane change execution is achieved.

[0084] The present invention is not limited to this embodiment, and any equivalent concepts or modifications within the technical scope disclosed by the present invention are included in the protection scope of the present invention.

Claims

1. A two-level decision-making method for an autonomous lane-changing system on an urban expressway, wherein the autonomous lane-changing system is installed in a vehicle controller; The vehicle is equipped with an autonomous lane change system in a multi-lane urban expressway traffic scenario. The vehicle's driving states on the urban expressway are defined as lane centering, preparing to change left, preparing to change right, executing left lane change, and executing right lane change. The lane centering state of the ego vehicle is defined as the basic state of the ego vehicle. The lane the ego vehicle is traveling in is defined as the driving lane, the adjacent lane to the left of the driving lane is defined as the left lane, and the adjacent lane to the right of the driving lane is defined as the right lane. Its characteristics are: The dual-level decision-making method includes an upper-level decision-making method and a lower-level decision-making method. The upper-level decision-making method calculates lane efficiency, and the lower-level decision-making method analyzes lane change safety. Specifically, the method includes the following steps: A. The autonomous lane change system controls the vehicle's driving state to a center-maintaining state; B. The autonomous lane change system determines whether the vehicle is blocked by a slow vehicle ahead. If so, go to step C; otherwise, go to step A. C. The autonomous lane change system calculates lane efficiency indices for the left lane, the driving lane, and the right lane, and compares them to determine the preferred lane. If the left lane efficiency index is higher, proceed to step D. If the right lane efficiency index is higher, proceed to step E. If the efficiency indices of both lanes are the same and higher than the driving lane efficiency index, proceed to step D. If the driving lane efficiency index is higher, proceed to step A. D. Generate a left lane change motivation and prepare to execute the left lane change. The autonomous lane change system generates a longitudinal occupancy map of the target lane and evaluates whether the current lane change opportunity allows the lane change. If so, the left lane change is executed. After the lane change is completed, the vehicle proceeds to step A. Otherwise, the vehicle remains in a left-hand lane preparation state until a suitable lane change opportunity is found. If a suitable lane change opportunity cannot be found within 5 seconds, the vehicle proceeds to step A. E. Generate motivation to change lanes to the right and prepare to implement the right lane change; the autonomous lane change system generates a longitudinal occupancy map of the target lane and evaluates whether the current lane change opportunity allows a lane change; if so, execute the right lane change; after the lane change is completed, proceed to step A; otherwise, the vehicle will maintain a preparatory state of driving on the right until a suitable lane change opportunity is found; if a suitable lane change opportunity cannot be found within 5 seconds, proceed to step A.

2. The dual-level decision-making method for an autonomous lane change system on an urban expressway according to claim 1, characterized in that: The method for calculating lane traffic efficiency is as follows: Cruising and following represent two opposing driving states in road traffic. The unobstructed state is called the free cruising state. Using the free cruising state as a baseline, we quantitatively analyze the mileage loss associated with following behavior under different traffic conditions. Based on this, we propose the concept of a "virtual cruising trajectory." Specifically, regardless of whether there are slower vehicles ahead that obstruct the ego vehicle's cruising, assume that the ego vehicle starts from its current speed and acceleration, adjusts to a preset cruising speed, and then maintains a constant speed. The resulting longitudinal displacement sequence over a specific future time interval is defined as the virtual cruising trajectory. Assume that the calculation time domain of the virtual cruise trajectory is t∈[0,t2], within which there are variable acceleration motion phases and uniform speed motion phases; suppose that the speed regulation time range of the variable acceleration motion phase is t∈[0,t1], and the speed regulation task ends at time t1; then the virtual cruise trajectory within the time range of the variable acceleration motion phase [0,t1] and the uniform speed motion phase [t1,t2] is: Among them, x vir1 (t) Trajectory of strain acceleration stage, x vir2 (t) corresponds to the trajectory of the uniform motion phase, and: x vir1 (t)=A1X X=[t t 2 t 3 ] T x vir2 (t)=v2(t-t1)+x vir1 (t) Where a jerk is the set jerk, i.e., the time derivative of acceleration, a0, v0, and v2 are the initial acceleration, initial vehicle speed, and uniform vehicle speed, respectively. A1 is the coefficient matrix, and X is the state variable matrix. The displacement time series of the standard distance following vehicle is called the standard following trajectory: x fol (t)=x f (t)-D Where x f (t) is the longitudinal displacement trajectory of the preceding vehicle in the lane at time t in the prediction time domain, D is the longitudinal distance increment; the upper bound of the standard following trajectory is x fol,max (t i ), the lower bound is x fol,min (t i ); the standard following trajectory is determined only by the effective perception range and prediction time domain of the vehicle; the longitudinal distance increment D is determined by the longitudinal length of the vehicle l self and the longitudinal length of the preceding vehicle l i The calculation formula is as follows: Taking the virtual cruise trajectory as the benchmark, the longitudinal mileage loss of each lane under the standard following trajectory is calculated, and the loss function is defined as: in, k ti is the compensation function, which is positively correlated with time; t i k corresponding to the moment ti =mt i , where m is a constant greater than 0 that determines the rate at which the compensation function grows over time; the larger the m, the faster the growth. The loss function is used to describe the degree of hindrance that the ego vehicle will experience during the following process in a certain lane over a period of time in the future. From the driver's perspective, this means that the expected longitudinal travel distance does not reach the expected value. The lower bound of the loss function is: The upper bound of the loss function is: Since the virtual cruise trajectory serves as a unified reference for all lanes, it depends only on the initial state of the ego vehicle and the predicted duration at the current moment and is independent of the lane selection. Furthermore, the upper and lower bounds of the standard following trajectory are independent of the specific lane selection and are instead determined by the vehicle's effective perception range and prediction time domain. Therefore, at any given moment, the loss function corresponding to each lane has the same upper and lower bounds. Based on this, the loss function for each lane is normalized to the interval [0, 1] based on its upper and lower bounds, and the loss index is defined as: Where a = left, mid, right Among them, the damage index of the lane where the vehicle is traveling is expressed as N mid If the target lane is on the left side of the driving lane, the lane loss index is expressed as N left If the target lane is on the right side of the driving lane, the lane loss index is expressed as N right express; The left lane traffic efficiency index is defined as: or left =N mid -N left The right lane traffic efficiency index is defined as: or right =N mid -N right The left and right adjacent lane efficiency index represents the relative degree of congestion of the target lane relative to the driving lane, and its value range is [-1, 1]. A negative index indicates that the target lane is more congested than the current lane, and the ego vehicle's autonomous lane change system has no incentive to change lanes. A positive index indicates that the target lane is more unobstructed, and the ego vehicle's autonomous lane change system has an incentive to change lanes. A value of 0 indicates that the target lane has no difference in efficiency compared to the driving lane, and the ego vehicle's autonomous lane change system has no incentive to change lanes. In addition, a larger absolute value of the index indicates a more significant degree of congestion or unobstructed traffic. This quantitatively characterizes and represents the efficiency of adjacent lanes.

3. The dual-level decision-making method for an autonomous lane change system on an urban expressway according to claim 1, characterized in that: The switching between the driving states of the vehicle follows the following logic: A1. When the vehicle ahead in the driving lane slightly blocks the ego vehicle, the ego vehicle's autonomous lane change system executes the upper-level decision-making method, generating a lane change motivation. The ego vehicle adjusts from maintaining lane center to moving closer to the lane with higher traffic efficiency, that is, moving to the left or right of the ego vehicle's lane, entering the lane change preparation state. Simultaneously, the ego vehicle activates its turn signal, ready to change lanes to the adjacent target lane. The lane change preparation state is divided into two states: the left lane change preparation state and the right lane change preparation state, corresponding to the ego vehicle moving to the left and right of the driving lane, respectively. A2. If the traffic efficiency of the target lane decreases significantly after entering the lane change preparation state, the driver should cancel the lane change intention, turn off the turn signal, and return to the current lane centering state to avoid the risks of unfavorable lane changes. A3. When the vehicle ahead in the driving lane significantly blocks the ego vehicle, the ego vehicle's autonomous lane change system executes the upper-level decision-making method to generate a lane change motivation. The ego vehicle activates its turn signal within the driving lane and enters the lane change preparation state. The lower-level decision-making method evaluates whether the current window of opportunity exists to ensure the safety of the entire lane change process. If the conditions for a safe lane change are met, the ego vehicle switches from the lane change preparation state to the lane change execution state. A4: Once the ego vehicle has completely entered the target lane, the lane change operation is considered complete, and the ego vehicle's autonomous lane change system switches to the target lane centering state. A5. Whether in lane centering mode or preparing to change lanes, the vehicle must maintain lane centering for a certain period of time before switching to another state to ensure the stability of behavioral decisions and driving safety.

4. The dual-level decision-making method for an autonomous lane change system on an urban expressway according to claim 1, characterized in that: The lower-level decision-making method designs a corresponding decision-making mechanism for real-time analysis and evaluation of the future driving trends of other vehicles in the target lane and their relative motion relationship with the ego vehicle when the ego vehicle is in a state of preparing to change lanes; The decision is based on the generation of the longitudinal occupancy map of the target lane and the analysis of the optimal lane change timing; The target lane longitudinal occupancy map is defined as: the set of longitudinal spatiotemporal areas occupied by all obstacles in a given lane over a period of time in the future, in a scenario where the ego vehicle is simplified as a point mass. The steps for generating the target lane longitudinal occupancy map are as follows: B1. Predict the future driving path of the target lane vehicle based on its existing motion parameters; B2. In the longitudinal and lateral dimensions, based on collision geometry principles and by appropriately expanding the target lane vehicle size as needed, construct graphs of the longitudinal and lateral position changes of the target lane vehicle over time. B3. Determine the time series of target lane occupancy by analyzing the graph of lateral position changes over time; B4. Based on the obtained target lane occupancy time series, extract relevant information from the longitudinal position change over time graph within the corresponding time period, thereby generating a target lane longitudinal occupancy map; B5. After obtaining the target lane longitudinal occupancy map, define the appropriate lane change opportunity as the segment on the time axis of the target lane longitudinal occupancy map corresponding to when the ego vehicle is in the lane change preparation state and the target lane is idle. If multiple appropriate lane change opportunities exist, analyze and select the optimal one, which will serve as the basis for the autonomous lane change system of the ego vehicle to execute the next-level decision-making method. The analysis method of the optimal lane change timing is as follows: The expected trajectory of the ego vehicle following the vehicle in the lane change preparation state is plotted into the target lane longitudinal occupancy map, and the intersection point is taken from the occupied area in the target lane longitudinal occupancy map to obtain the lane change opportunity situation; After analysis, it is found that if the desired trajectory of the ego vehicle overlaps with the occupied area, it means that there is an unfeasible area conflict between the ego vehicle and the target lane, and the lane change cannot be performed. The non-overlapping part means that there is no area conflict in the target lane, and the ego vehicle can perform the lane change. Since the lane change action of the ego vehicle takes a certain amount of time, the safety threshold of the lane change time is defined as T s , the time windows that meet the lane change timing requirements in the target lane longitudinal occupancy diagram are T r , where r = 1, 2, 3, ...; When a certain time window T r >Safety threshold T s When , it means that the vehicle has the conditions to safely execute the lane change decision within the time window, and this time is the appropriate lane change time; during the lane change process, the smaller the lateral acceleration of the vehicle, the smoother the lane change action and the higher the comfort; on the basis of meeting safety requirements, the time window with the highest comfort is the optimal time window, and the corresponding time is the optimal lane change time; in addition, different safety thresholds T are set according to actual conditions. s To meet diverse driving needs; When two time windows with similar lateral acceleration and no essential difference in safety and comfort appear during a lane change, the time window with a closer start time is selected to complete the lane change as soon as possible and ensure traffic efficiency. If an optimal time window has already started and the vehicle is currently within it, the optimal time window is abandoned and the search continues for a subsequent optimal time window that has not yet started at the current moment to ensure the safe completion of the entire lane change process.

Images