An assisted driving vehicle lane-changing decision method for heterogeneous traffic flow environment and with human-like dynamic decision characteristics

Abstract

The application discloses a lane-changing decision-making method for an assisted driving vehicle in a heterogeneous traffic flow environment and with human-like dynamic decision-making characteristics, relates to the field of lane-changing decision-making of an assisted driving vehicle, and aims to solve the problem that the behavior strategy generated by an algorithm in the prior art deviates from the actual expectation of a human driver, and adopts the technical scheme that information such as a host vehicle and a surrounding vehicle environment is taken as reference information of a decision-making module, the decision-making module adopts a PPO-DDM fusion module, the PPO model and the DDM model are dynamically weighted according to traffic conditions, an environment self-adaptive parameter adjustment mechanism submodule and an evidence accumulation process numerical implementation submodule are introduced into a DDM decision-making submodule, the decision-making module can predict the future behavior of surrounding vehicles, the decision-making not only has the continuity feature of gradual adjustment, which is convenient for a human driver to accurately predict subsequent behavior, but also can be adaptively adjusted according to real-time traffic scene changes.

Description

Technical Field

[0001] This invention relates to the field of lane change decision-making technology for assisted driving vehicles, specifically a lane change decision-making method for assisted driving vehicles that is designed for heterogeneous traffic flow environments and possesses human-like dynamic decision-making characteristics. Background Technology

[0002] With the gradual deployment of assisted driving vehicles (ADAS) in open road scenarios, how to achieve safe, efficient, and human-machine interface-compatible lane-changing behavior in complex traffic environments has become a core issue that urgently needs to be addressed. Especially in high-density interaction areas such as weaving zones, lane-changing behavior is frequent and highly risky, and existing ADAS systems still have shortcomings in understanding human driving behavior and responding to high-risk lane-changing decisions. Existing research on lane-changing decisions based on reinforcement learning, while able to learn the optimal driving strategy through trial and error in simulated environments, still faces numerous challenges when dealing with complex traffic scenarios. First, reinforcement learning methods typically lack a dynamic characterization of the decision-making process during modeling, ignoring the continuous characteristics of decision accumulation and gradual adjustments in human driving behavior, leading to discrepancies between the algorithm-generated behavioral strategies and the actual expectations of human drivers. Second, although some studies incorporate the operational status information of surrounding vehicles as environmental input, they generally fail to predict their future behavior, potentially resulting in overly aggressive or conservative strategy execution, reducing system safety and efficiency. Finally, the objective functions in current reinforcement learning frameworks often use static weights to balance multiple objectives, lacking the ability to adaptively adjust according to real-time traffic scene changes, thus limiting the model's adaptability and robustness in dynamic and changing environments. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to overcome the existing defects and provide a lane-changing decision method for assisted driving vehicles with human-like dynamic decision-making characteristics for heterogeneous traffic flow environments, which can effectively solve the problems in the background art.

[0004] To achieve the above objectives, this invention discloses a lane-changing decision-making method for assisted driving vehicles with human-like dynamic decision-making characteristics in heterogeneous traffic flow environments. The technical solution adopted includes the following steps:

[0005] Step 1: Obtain the vehicle's operating status information, the relative status information of surrounding vehicles, traffic scene information, and environmental context information. The traffic environment perception module then transmits this information to the multi-dimensional state space module.

[0006] Step 2: The multi-dimensional state space module converts the vehicle operation status information output by the traffic environment perception module into vehicle characteristic codes, converts the relative status information of surrounding vehicles into environmental interaction codes, converts traffic scene information into traffic scene codes, converts environmental context information into environmental context codes, and finally realizes multi-dimensional feature fusion through encoder fusion, and outputs the fused result as a state vector and passes it to the dynamic fusion decision module.

[0007] Step 3: The dynamic fusion decision module makes a decision on the next driving state of the vehicle based on the current state vector output result. The decision depends on the accumulation of evidence from the current state vector output result. When the evidence accumulates to a threshold, the lane change intention is determined and the result is output. The output result is the lane change intention, the total reaction time, and the overall confidence level.

[0008] As a preferred embodiment of the present invention, the dynamic fusion decision module includes a PPO-DDM decision submodule and a weight adaptive mechanism submodule. The PPO-DDM decision submodule includes a PPO decision submodule and a DDM decision submodule. The PPO decision submodule uses an Actor network to output lane-changing decision results and acceleration values ​​based on PPO. The DDM decision submodule outputs lane-changing decision results based on DDM. The output results of the PPO-DDM decision submodule are input to the weight adaptive mechanism submodule. The weight adaptive mechanism submodule dynamically allocates the lane-changing decision weights output by the PPO decision submodule and the DDM decision submodule according to changes in weather, traffic density, and risk level factors. In low-risk, low-density scenarios, the weight of the PPO decision submodule is increased, and in high-risk or severe weather conditions, the weight of the DDM decision submodule is increased, obtaining the final weighted fusion decision result as the module output.

[0009] As a preferred embodiment of the present invention, the calculation formula for the weighted fusion decision result is as follows:

[0010]

[0011] in, This represents the final, synthesized probability distribution of actions. The action probability distribution output by the PPO decision submodule; The action probability distribution of the DDM decision submodule; and These are the weight coefficients for the PPO decision submodule and the DDM decision submodule, respectively.

[0012]

[0013]

[0014] in, These are the default weights for DDM under the basic conditions; This represents the overall risk score; This is a normalized value for traffic density, reflecting the degree of congestion.

[0015] The comprehensive risk score It consists of at least one of collision risk, road environment risk, and traffic rule / road structure risk, and is obtained through normalization. Risk score within the range;

[0016] The comprehensive risk score Calculate as follows:

[0017]

[0018] in, This indicates that the value will be cropped to... interval; Indicates a collision risk indicator; Indicates road environmental risk indicators; Indicates traffic rule / road structure risk indicators; Let be the weight coefficient, and satisfy... .

[0019] As a preferred embodiment of the present invention, the DDM decision-making submodule includes a DDM model construction submodule, an environmental adaptive parameter adjustment mechanism submodule, an evidence accumulation process numerical implementation submodule, and a confidence level and decision quality quantification submodule. The results of the DDM model construction submodule and the environmental adaptive parameter adjustment mechanism submodule are input to the evidence accumulation process numerical implementation submodule for evidence accumulation, ultimately obtaining the lane-changing intention and total reaction time. The evidence accumulation results of the evidence accumulation process numerical implementation submodule are input to the confidence level and decision quality quantification submodule, which outputs a comprehensive confidence level.

[0020] As a preferred embodiment of the present invention, the decision rule of the DDM model construction submodule is as follows:

[0021]

[0022] in, The accumulated evidence value at time n. For decision-making reaction time, The decision threshold represents the amount of evidence required to trigger a decision.

[0023] Decision reaction time This includes the time it takes for evidence to first reach the threshold and the perception delay, calculated using the following formula:

[0024]

[0025] in, The time when the evidence first reaches the threshold. This includes non-decision time such as perception delay, cognitive processing, and action execution time.

[0026] The DDM model is:

[0027]

[0028] in, The accumulated evidence value at time n. The drift rate indicates the average direction and speed at which evidence accumulates per unit of time, in relation to a decision. Random noise is used to represent the uncertainty of information.

[0029] Drift rate The calculation formula is:

[0030]

[0031] in, Indicates the first One source of evidence at any time The contribution of drift rate; For the first The weight of each source of evidence; This represents the total number of evidence sources.

[0032] As a preferred technical solution of the present invention, the environmental adaptive parameter adjustment mechanism submodule calculates the model-related dynamic thresholds, including dynamic drift rate and adaptive decision threshold, based on the DDM model construction submodule.

[0033] The dynamic drift rate is calculated as follows:

[0034]

[0035] in, Based on the drift rate, This represents the model's basic drift rate bias term. The weighting coefficient controls the strength of the influence of the PPO strategy output on decision-making tendency; The PPO network's intention to change lanes in the current state can be derived from the softmax probability output by the PPO network.

[0036]

[0037] in, This is the final drift rate; This is a weather correction factor; This is a correction factor for lane-changing type;

[0038] The adaptive decision threshold calculation method is as follows:

[0039]

[0040] in, Basic decision threshold; This is a weather correction factor; This is the lane-changing type coefficient.

[0041] As a preferred embodiment of the present invention, the evidence includes PPO preference evidence, safety urgency evidence, and traffic efficiency evidence;

[0042] The method for calculating the PPO preference evidence is as follows:

[0043]

[0044] in, Indicates the state The probability of the PPO strategy choosing the "lane change" action; Indicates the state The probability of PPO strategy selecting the "lane keeping" action; The weighting coefficients for PPO preference evidence;

[0045] The method for calculating the evidence of security urgency is as follows:

[0046]

[0047] in, The scenario-based instantaneous risk assessment method is used for calculation. Weighting coefficients for evidence of security urgency;

[0048] The method for calculating the traffic efficiency evidence is as follows:

[0049]

[0050] in, As a traffic efficiency indicator, it can be calculated based on the difference in traffic flow density between lanes. The weighting coefficients for evidence of traffic efficiency;

[0051] The weighting coefficients of the PPO preference evidence, the weighting coefficients of the safety urgency evidence, and the weighting coefficients of the traffic efficiency evidence satisfy the following: .

[0052] As a preferred embodiment of the present invention, the evidence accumulation is discretized using the Euler-Markov approximation, including the following steps:

[0053] Step 31: Initialize the accumulated evidence value and time counter;

[0054] Step 32: Within each time step, calculate the evidence increment consisting of the drift term and the diffusion term;

[0055] The drift term is the product of the drift rate and the time step.

[0056] The diffusion term is the product of a Gaussian random number and a diffusion coefficient;

[0057] Step 33: Overlay the current cumulative evidence value and the evidence increment as a new cumulative evidence value. Compare the new cumulative evidence value with the decision threshold. When the evidence exceeds the threshold for the first time, determine the final decision of lane changing or lane keeping and its confidence level. If the threshold is not reached within the maximum time range, the direction is determined by default based on the evidence symbol, and the confidence level is set to 0.5.

[0058] As a preferred embodiment of the present invention, in step 33, boundary conditions and a reflection boundary mechanism are introduced in the model implementation to ensure the numerical stability of the evidence accumulation process. When the absolute value of the accumulated evidence value approaches the decision threshold, the accumulated decision value is considered to have entered the soft triggering interval, and the soft boundary triggering probability is calculated. :

[0059]

[0060] in, Represents the absolute value of accumulated evidence. The decision threshold for changing lanes or staying straight. The sensitivity coefficient for triggering reflection boundary. This is the starting value of the soft-trigger interval. ;

[0061] Based on the soft boundary trigger probability Adaptive correction is applied to the drift rate, i.e., based on the original drift rate. Based on the superposition and Additional drift terms in the same direction accelerate the arrival of evidence toward the current decision direction at the threshold, reduce the probability of prolonged oscillations around the threshold, and improve decision efficiency and stability.

[0062] in, This is a sign indicator function used to indicate the direction of the current accumulated evidence value; when A timeout of +1 indicates a bias towards lane changing. When the value is negative, the output is −1, indicating bias to maintain lane position.

[0063] As a preferred embodiment of the present invention, the method for calculating the comprehensive confidence level is as follows:

[0064]

[0065] in, , For the weighting coefficients, satisfying + + =1;

[0066] The strength of evidence is calculated as follows:

[0067]

[0068] in, For the final accumulated value of evidence, This is the decision threshold; the closer the strength of evidence is to 1, the more sufficient and credible the evidence is for the decision to change lanes or stay in lane.

[0069] To ensure consistency in decision-making time, the calculation method is as follows:

[0070]

[0071] in, For the actual decision-making time, The expected value is based on experience and can be calculated from the median historical decision-making time in similar scenarios (same weather, same speed range, same lane change type). This is a sensitivity parameter for time consistency; the closer the decision time consistency is to 1, the more the model's time-series performance matches expectations.

[0072] For drift rate stability, the calculation method is as follows:

[0073]

[0074] in, and Let Variance and mean of the drift rate be represented, respectively. To prevent small constants with a denominator of zero; the closer the drift rate stability is to 1, the smoother the drift rate process and the more reliable the decision.

[0075] Compared with the prior art, the beneficial effects of the present invention are as follows: By inputting information such as the vehicle information, the relative situation of surrounding vehicles, and traffic scenarios into the decision-making module, and combining the accumulated evidence in the decision-making module, the present invention enables the decision-making module to predict the future behavior of surrounding vehicles. This not only enables the decision to have the continuous characteristics of gradual adjustment, thus better conforming to the behavioral logic of human driving behavior and facilitating accurate prediction of subsequent behavior by human drivers, but also enables adaptive adjustment based on real-time changes in traffic scenarios.

[0076] Furthermore, by introducing the PPO-DDM decision submodule, the PPO-DDM model it provides can autonomously switch between the PPO model and the DDM model according to the current traffic conditions, and choose between aggressive driving or conservative driving according to the driving environment, making lane-changing decisions more flexible and improving comfort.

[0077] Furthermore, the application of this invention can effectively reduce the risk of traffic accidents and significantly improve driving safety. Meanwhile, test results show that the PPO-DDM model exhibits faster convergence speed and stronger learning ability during the model training phase. In terms of model performance, the average decision consistency score of the PPO-DDM model increased from 7.2 points for the basic PPO model to 9.1 points, an increase of 26.4%. In scenarios of vehicle deceleration and high-speed lane separation, the average score further improved to 9.4 points, indicating that the model has a significant advantage in dynamic decision consistency. This performance improvement stems from the DDM's coherent modeling of lane-changing behavior. Regarding response performance, this invention improves the drift rate amplitude and enhances the consistency of evidence direction through multi-source evidence fusion. Simultaneously, it introduces a soft-triggered reflection boundary mechanism to reduce oscillations near the threshold and long-tail delay, allowing evidence to accumulate and reach the decision threshold more quickly, thereby shortening the decision time and reducing model response latency. The average response time was reduced from 0.37 seconds to 0.22 seconds, a reduction of 40.5%, significantly enhancing safety redundancy in emergency traffic situations. In terms of decision success rate, the average level improved from 85.7% to 94.1%, achieving an effective balance between speed and accuracy, further validating the advantages of the fusion framework in multi-scenario driving decision-making. This can help companies save significant R&D and testing costs and accelerate the market launch of assisted driving products. Attached Figure Description

[0078] Figure 1 This is a schematic diagram of the PPO-DDM dynamic fusion lane-changing decision model of the present invention;

[0079] Figure 2 This is a SUMO traffic simulation scene diagram of the present invention;

[0080] Figure 3 This is a comparison chart of the learning curves of the model of this invention;

[0081] Figure 4 This is a comparative analysis chart of the overall performance of the model of this invention;

[0082] Figure 5 This is a comparative analysis chart of the stability of the model of this invention. Detailed Implementation

[0083] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0084] Example 1

[0085] like Figure 1 As shown, this invention proposes a PPO-DDM dynamic fusion lane-changing decision model, including a traffic environment perception module, a multi-dimensional state space module, and a dynamic fusion decision module. The dynamic fusion decision module includes a PPO-DDM decision submodule and a weight adaptive mechanism submodule. The PPO-DDM decision submodule includes a PPO decision submodule and a DDM decision submodule. The DDM decision submodule includes a DDM model construction submodule, an environmental adaptive parameter adjustment mechanism submodule, a numerical implementation submodule of the evidence accumulation process, and a confidence and decision quality quantification submodule.

[0086] This invention also discloses a lane-changing decision-making method for assisted driving vehicles based on the above-mentioned PPO-DDM dynamic fusion lane-changing decision-making model, which is suitable for heterogeneous traffic flow environments and has human-like dynamic decision-making characteristics. The technical solution adopted includes the following steps:

[0087] Step 1: The traffic environment perception module acquires vehicle operating status information, relative status information of surrounding vehicles, traffic scene information, and environmental context information through onboard sensors and roadside equipment. The information from this module will serve as the input information for the multi-dimensional state space module.

[0088] Step 2: The multi-dimensional state space module converts the vehicle's operating state information output by the traffic environment perception module into vehicle characteristic codes, the relative state information of surrounding vehicles into environmental interaction codes, the traffic scene information into traffic scene codes, and the environmental context information into environmental context codes. Finally, multi-dimensional feature fusion is achieved through encoder fusion, and the fused result is used as a state vector input to the dynamic fusion decision module; specifically as follows:

[0089] The vehicle's operating status information includes: vehicle speed v, acceleration a, position (x, y), heading angle θ, and steering wheel angle. The output vehicle characteristic code is: ;

[0090] The relative status information of surrounding vehicles includes the relative distances of key vehicles such as the vehicle in front, the vehicle behind, the vehicle to the left front, the vehicle to the left rear, the vehicle to the right front, and the vehicle to the right rear. Relative velocity Collision Time Tolerance (TTC); Output environment interaction code is ;

[0091] The traffic scene information includes: coded scene type, road structure, and traffic density; it uses one-hot encoding of scene type (highway / city / congestion), numerical road features (number of lanes, curvature, speed limit), and statistical features (average vehicle speed, density), and the output traffic scene code is as follows.

[0092] ;

[0093] The environmental context information includes: weather, time of day, visibility, and road state (e.g., smooth, potholed, muddy, icy, etc.); the output environmental context encoding is...

[0094] ;

[0095] The encoder fusion result is as follows:

[0096] .

[0097] Step 3: The dynamic fusion decision module makes a decision on the next driving state of the vehicle based on the current state information of the vehicle, and outputs the lane change intention, total reaction time and comprehensive confidence level.

[0098] The input of the PPO-DDM decision submodule is the state vector output by the multidimensional state space module. The PPO decision submodule of the PPO-DDM decision submodule uses an Actor network to output lane change decision results and acceleration values ​​based on PPO. This is existing technology. The specific methods for outputting lane change decision results and acceleration values ​​based on PPO will not be elaborated here. The DDM decision submodule of the PPO-DDM decision submodule outputs lane change decision results based on DDM.

[0099] The DDM decision submodule's DDM model construction submodule abstracts the DDM model into a binary decision problem of "whether to change lanes," i.e., changing lanes or staying straight; the environmental adaptive parameter adjustment mechanism submodule, based on the DDM model construction submodule, calculates the model-related dynamic thresholds, including dynamic drift rate and adaptive decision threshold.

[0100] The outputs of the DDM model construction submodule and the environmental adaptive parameter adjustment mechanism submodule serve as inputs to the evidence accumulation process numerical realization submodule. The outputs of the evidence accumulation process numerical realization submodule are lane-changing decision, total decision time, and decision confidence, which are then used as inputs to the confidence and decision quality quantification submodule. The confidence and decision quality quantification submodule is used to comprehensively quantify the final state, time consistency, and drift rate volatility of the evidence accumulation process to obtain a comprehensive confidence level, ensuring the interpretability and practical usability of the model output.

[0101] Specifically, the decision rule for the DDM model construction submodule is defined as follows:

[0102]

[0103] in, Let t be the accumulated evidence value at time t, and t be the decision reaction time. The initial value is usually set to t. ; The decision threshold represents the amount of evidence required to trigger a decision. ), which is usually set to 1.05; z is the initial bias, reflecting the initial decision tendency, and is usually set to 0.

[0104] Decision reaction time It consists of two parts: the time it takes for evidence to first reach the threshold and the perception delay. The specific calculation formula is as follows:

[0105]

[0106] in, The time when the evidence first reaches the threshold. This includes non-decision time such as perception delay, cognitive processing, and action execution time.

[0107] To adapt to complex traffic decision-making environments, the DDM model is extended to a form of multi-source evidence fusion:

[0108]

[0109] in, The accumulated evidence value at time n. The drift rate indicates the average direction and speed at which evidence accumulates per unit of time, in relation to a decision. Random noise is used to represent the uncertainty of information.

[0110] Drift rate The calculation formula is:

[0111]

[0112] in, Indicates the first One source of evidence at any time The contribution of drift rate; For the first The weight of each source of evidence; The total number of evidence sources;

[0113] In the environmental adaptive parameter adjustment mechanism submodule, the dynamic drift rate is calculated as follows:

[0114]

[0115] in, Base drift rate; This represents the model's base drift rate bias term, which is usually set to 0. The weighting coefficient controls the strength of the influence of the PPO strategy output on decision-making tendency, and its value range is [range missing]. ; The intention preference of the PPO network for lane changing in the current state can be obtained from the softmax probability output by the PPO network. The method of outputting the softmax probability by the PPO network is existing technology and will not be described in detail here.

[0116]

[0117] in, This is the final drift rate; This is a weather correction factor, with a range of values. The specific values ​​are shown in Table 1; This is a lane-change type correction factor, with a value range of... The specific values ​​are shown in Table 2.

[0118] Table 1 Example value table

[0119]

[0120] Table 2: Example value table

[0121]

[0122] In the environmental adaptive parameter adjustment mechanism submodule, the adaptive decision threshold is calculated as follows:

[0123]

[0124] in, The basic decision threshold, consistent with the previous text, is set at 1.05; This is a weather correction factor; This is the lane-changing type coefficient.

[0125] In the numerical implementation submodule of the evidence accumulation process, evidence accumulation involves PPO preference evidence, safety urgency evidence, and traffic efficiency evidence. The specific calculation methods are as follows:

[0126] The method for calculating PPO preference evidence is as follows:

[0127]

[0128] in, Indicates the state The probability of the PPO strategy choosing the "lane change" action; Indicates the state The probability of the PPO strategy choosing the "keep lane" action is obtained from the softmax probability output by the PPO network. The weighting coefficients for PPO preference evidence are used to measure its importance in decision-making. The specific values ​​of the weight coefficients can be adjusted according to the training scenario requirements and training results.

[0129] The method for calculating evidence of safety urgency is as follows:

[0130]

[0131] in, The scenario-based instantaneous risk assessment method is used for calculation. The weighting coefficient for evidence of security urgency represents the relative influence of that factor in the decision-making process. The specific value can be determined according to the requirements of the training scenario. Adjustment.

[0132] The calculation method used is existing technology; for details, please refer to:

[0133] Guo YJ, Chen YY, Gu X, Guo JF, Zheng SY, Zhou Y T. Dynamictraffic graph based risk assessment of multivehicle lane interactions changecenarios [J]. Physica A: Statistical Mechanics and its Applications, 2024,643: 129791.

[0134] The method for calculating traffic efficiency evidence is as follows:

[0135]

[0136] in, Traffic efficiency is an indicator that can be calculated based on differences in lane traffic flow density; this is existing technology and will not be elaborated upon here. The weighting coefficient for traffic efficiency evidence indicates the proportion of this factor in the decision-making trade-off. The specific value can be adjusted according to the training scenario requirements and training results; among them, ,For example .

[0137] The numerical implementation of the evidence accumulation process is discretized using the Euler-Markov approximation. First, the accumulated evidence value and time counter are initialized. Within each time step, the evidence increment is calculated, consisting of a drift term (the product of the drift rate and the time step size) and a diffusion term (the product of a Gaussian random number and the diffusion coefficient). After each update, the accumulated evidence is compared with the decision threshold. When the evidence exceeds the threshold (positive or negative) for the first time, the final decision of lane changing or lane keeping and its confidence level are determined. If the threshold is not reached within the maximum time range, the direction is determined by default based on the evidence sign, and the confidence level is set to 0.5.

[0138] The numerical implementation process of the above evidence accumulation process is as follows:

[0139] Step a, Initialization: Let Set the maximum decision time window. (Preferred time: 2-4 seconds);

[0140] Step b, for each output: lane-changing decision, total decision time, and initial decision confidence. Time step Next update of evidence increment:

[0141]

[0142] Step c, update And determine:

[0143] ·like Then output "Lane Change";

[0144] ·like Then the output will be Keep Lane;

[0145] Otherwise, continue accumulating until... ;

[0146] If in If the threshold is not reached, the default will be based on... The symbol indicates direction (X) Output lane change (otherwise keep lane), and set the initial confidence level to 0.5.

[0147] Step d, obtain Then, calculate the total reaction time:

[0148]

[0149] Furthermore, boundary conditions and a reflection boundary mechanism are introduced into the model implementation to ensure the numerical stability of the evidence accumulation process. The boundary conditions define the binary decision output when the accumulated evidence first reaches the threshold, while the reflection boundary, when the evidence approaches the threshold range (soft trigger range, such as 0.8–1.0a), avoids long-term oscillation of the accumulated evidence by increasing the trigger probability or additional drift rate, thereby improving the model's decision efficiency and security. The implementation method is as follows:

[0150] Calculate the soft boundary trigger probability :

[0151]

[0152] in, Represents the absolute value of accumulated evidence. The decision threshold for changing lanes or staying straight. For the sensitivity coefficient of the reflection boundary trigger, the preferred value is... , This is the starting value of the soft-trigger interval. The value is 0.8.

[0153] When the absolute value of the accumulated evidence enters the soft trigger zone Calculate the soft boundary trigger probability. The system adaptively corrects the drift rate based on the stated probability, that is, at the original drift rate... Based on the superposition of Additional drift terms in the same direction accelerate the arrival of evidence towards the current decision direction at the threshold, reduce the probability of prolonged oscillations around the threshold, and improve decision-making efficiency and stability. Among these, This is a sign indicator function used to indicate the direction of the current accumulated evidence value; when When it is positive, output +1 (i.e. ) indicates a tendency to change lanes, when When it is negative, output -1 (i.e. This indicates a bias towards keeping the lane, thereby accelerating the amount of evidence. Exceeding the decision threshold .

[0154] The calculation method for the overall confidence score output by the confidence score and decision quality quantification submodule is as follows:

[0155]

[0156] in, , For the weighting coefficients, satisfying + + =1, for example , ;

[0157] Evidence strength is used to measure the proportion of the final accumulated evidence relative to the decision threshold, representing the degree of "certainty" of the decision output. Its calculation method is as follows:

[0158]

[0159] in, For the final accumulated value of evidence, This is the decision threshold; the closer the strength of evidence is to 1, the more sufficient and credible the evidence is for the decision to change lanes or stay in lane.

[0160] To ensure consistency in decision time, this parameter characterizes the degree of matching between the model's output decision time and the expected value, reflecting the model's temporal stability across multiple simulations. The calculation method is as follows:

[0161]

[0162] in, For the actual decision-making time, The expected value is based on experience and can be calculated from the median historical decision-making time in similar scenarios (same weather, same speed range, same lane change type). For time consistency sensitivity parameters, the closer the decision time consistency is to 1, the better the model's time-series performance matches expectations; a value of [value missing] is recommended. ;

[0163] To determine the drift rate stability, which reflects the fluctuation of the drift rate during the decision-making process, the calculation method is as follows:

[0164]

[0165] in, and Let Variance and mean of the drift rate be represented, respectively. To prevent small constants with denominators of zero, a value of [value missing] is recommended. The closer the drift rate stability is to 1, the smoother the drift rate process and the more reliable the decision-making.

[0166] In step 3, the input of the weight adaptive mechanism submodule is the output of the PPO-DDM decision submodule. This module considers weather, traffic density, and risk level factors, and dynamically allocates the lane-changing decision weights of the PPO and DDM decision submodules based on changes in these factors. In low-risk, low-density scenarios, PPO dominates, emphasizing the flexibility and efficiency of the decision. In high-risk or severe weather conditions, the DDM weight is increased to make the decision more biased towards the cautious lane-changing habits of human drivers. The final weighted fusion decision result is then used as the module output. The specific formula for allocating the lane-changing decision weights is as follows:

[0167]

[0168] in, This represents the probability distribution of actions after fusion. The action probability distribution output by the PPO policy network (obtained by softmax normalization of the Actor network output); The action probability distribution output by the DDM evidence accumulation mechanism (obtained by multiple random simulations of the statistical boundary arrival frequency, or by monotonic mapping of the normalized ratio of the evidence accumulation value to the threshold). and Let be the weighting coefficients for PPO and DDM, respectively, and satisfy the following:

[0169]

[0170] To achieve adaptability to different traffic risks and congestion levels, this invention preferably calculates dynamically based on risk scores and traffic density. The calculation formula is as follows:

[0171]

[0172] in, The default weights for DDM under basic conditions are preferably set according to weather conditions: 0.4 for sunny days and 0.6 for rainy days; other weather conditions can also be set according to preset rules or by looking up a table. The comprehensive risk score has a range of values. It is used to characterize the risk level of multiple factors such as collision risk and road risk. The higher the risk, the more likely it is to increase the weight of DDM to enhance the prudence and interpretability of decision-making. This is the normalized value of traffic density, with a range of values ​​of [value range missing]. It is used to characterize the degree of road congestion. The higher the degree of congestion, the more likely it is to increase the weight of DDM in order to improve the stability of decision-making.

[0173] The comprehensive risk score It consists of at least one of collision risk, road environment risk, and traffic rule / road structure risk, and is obtained through normalization. Risk score within the range. Calculate as follows:

[0174]

[0175] in, This indicates that the value will be cropped to... interval; Indicates a collision risk indicator; Indicates road environmental risk indicators; Indicates traffic rule / road structure risk indicators; , , Let be the weight coefficient, and satisfy... Weights can be selected This is to reflect the principle of prioritizing collision risk.

[0176] 1) Collision risk indicators The calculation (preferably based on TTC)

[0177] Collision risk is calculated using the minimum collision time margin (TTC):

[0178]

[0179] in, The set of targets that may potentially conflict with the vehicle (which may include vehicles in front of and behind the vehicle in the current lane and the target lane, etc.); For the vehicle and the target vehicle The relative distance; The closing velocity; To prevent division by zero of small constants. Based on a threshold. Normalizing it yields:

[0180]

[0181] in, This indicates that the value will be cropped to... interval; The collision risk threshold is preferably set to 4 seconds in high-speed scenarios and 3 seconds in low-speed urban scenarios.

[0182] 2) Road environmental risk indicators Calculation

[0183]

[0184] in, This indicates that the value will be cropped to... interval; This represents the road surface adhesion coefficient of the road segment at the current travel time t. This indicates the reference adhesion coefficient for dry road surfaces (e.g., 0.8). This indicates a low adhesion limit (e.g., 0.3).

[0185] 3) Traffic rules / road structure risk indicators The calculation (based on road rules and map constraints)

[0186] Traffic rule / road structure risk indicators are used to characterize the degree of limitation in lane changing feasibility, and can be defined based on high-precision maps or lane line recognition results as follows:

[0187] ;

[0188] Finally, the output includes the lane change intention, total reaction time, and overall confidence level.

[0189] Step 4: The input to the vehicle lane-changing control system is the lane-changing intention, total reaction time, and overall confidence level output by the dynamic fusion decision module. The control system sets up a behavioral state machine. When the decision is to change lanes and the overall confidence level exceeds a preset threshold and passes the safety executability gating, it enters the lane-changing preparation state. In the preparation state, the system maintains the lane and adjusts the longitudinal speed to meet the target lane clearance requirements. When the corresponding execution trigger moment is reached and the gating conditions are still met, the system switches to the lane-changing execution state and calls the trajectory planning module to generate a lane-changing trajectory that meets the vehicle dynamics constraints. The underlying steering and longitudinal controllers track the trajectory to complete the lane change. If the risk indicators deteriorate or the confidence level drops below the threshold during execution, the abort logic is triggered to return to the lane-keeping state, thereby achieving interpretable lane-changing timing control and safe and robust execution. This step is existing technology, and the specific method will not be described in detail.

[0190] Model training

[0191] The PPO-DDM dynamic fusion lane-changing decision model proposed in Example 1 was trained using the publicly available aerial trajectory dataset UCF-SST.

[0192] (a) Training data preprocessing

[0193] The preprocessing objectives are: to unify coordinates, extract features, construct labels (lane change events / decision times), and perform normalization and slicing.

[0194] (1) Coordinates and lane representation

[0195] Convert global coordinates to lane coordinates (Frenet or "s / d along lane centerline"):

[0196] : Along the road direction

[0197] Lateral offset (lane centerline is 0)

[0198] Lane marking / assignment: based on The current lane and the target lane are determined by the lane boundaries.

[0199] (2) Lane change event and decision time tag

[0200] Lane change begins: lateral speed Exceeding a threshold (e.g., 0.2 m / s) or lateral deviation across lane lines;

[0201] Lane change complete: The vehicle's center enters the target lane and its lateral speed decreases;

[0202] Decision time tag (suggested definition: the time difference between the emergence of a lane-changing opportunity and the start of execution):

[0203] Lane change opportunity: The first moment when the target lane meets the minimum clearance / rule-permitted condition;

[0204] Execution begins: The moment when the threshold is first exceeded.

[0205] (3) State characteristics

[0206] composition

[0207] The recommendations should include at least (and be normalized):

[0208] Bicycle: Speed wait;

[0209] Vehicle in front / behind in the target lane: relative distance Relative velocity Time to Collision (TTC) / Time to Front Distance (THW);

[0210] Lane level: Current lane and target lane density, average speed;

[0211] Rules / Map: Is lane changing allowed? Distance to exit? Is there construction / lane end?

[0212] Weather: Visibility, rainfall intensity, or weather rating (used for...) ).

[0213] (4) Calculation of risk and density indicators

[0214] :Depend on Or predict the collision probability to be normalized to [0,1];

[0215] Use window length The number of vehicles in the vehicle or the speed-based congestion index is normalized to [0,1].

[0216] It is recommended to perform the following for all continuous features:

[0217] Clipping (removing extreme values, e.g., P1–P99)

[0218] z-score or min-max normalization

[0219] Missing values ​​are filled (0 or last observation carry).

[0220] (5) Data slicing and partitioning

[0221] Sequence window: For example, each trajectory is cut into segments of 5–10 seconds in length;

[0222] Training / validation / testing is divided into "scenario / road segment" (to avoid leakage within the same road segment).

[0223] (II) Training Environment

[0224] (1) Hardware environment

[0225] CPU: 16–64 cores x86 (such as AMD EPYC / Intel Xeon)

[0226] GPU: NVIDIA RTX 4090 (24GB) or A100 (40 / 80GB) × 1–4

[0227] Memory: 64–256GB

[0228] Storage: NVMe SSD ≥ 2TB

[0229] (2) Software environment

[0230] OS: Ubuntu 22.04 LTS

[0231] Python: 3.10

[0232] Deep Learning: PyTorch 2.x + CUDA 12.x

[0233] RL library: Stable-Baselines3 / RLlib (choose one)

[0234] Simulation: SUMO (including TraCI) or CARLA

[0235] Training interface: Gymnasium / OpenAI Gym-style environment encapsulation

[0236] Log: TensorBoard / Weights & Biases (optional)

[0237] (III) Training Process

[0238] Training of the PPO-DDM dynamic fusion lane-changing decision model (in a simulation environment)

[0239] (1) Training objectives and rewards

[0240] Comprehensive Rewards (Example):

[0241]

[0242] Safety Strong collision penalty, low TTC penalty

[0243] efficiency Target vehicle speed tracking, traffic efficiency

[0244] Comfort Lateral / longitudinal acceleration and jerk penalty

[0245] rule Solid line / No entry / Penalties for crossing boundary

[0246] (2) PPO core hyperparameters

[0247] Optimizer: AdamW

[0248] Initial learning rate (actor / critic): 3e-4

[0249] Weight decay: 1e-4

[0250] Discount factor γ: 0.99

[0251] GAEλ: 0.95

[0252] clip range: 0.2

[0253] entropy coef: 0.01

[0254] Value loss coefficient: 0.5

[0255] max grad norm: 0.5

[0256] rollout length (number of sampling steps per cycle): 2048

[0257] Batch size (total samples updated each time): 65536

[0258] mini-batch size: 2048

[0259] epochs per update: 10

[0260] Gradient accumulation steps: 4 (Enabled when video memory is insufficient)

[0261] Total training steps: 3e7 – 1e8 (commonly used: 5e7)

[0262] Convergence strategy (early stop): If the average reward on the validation set does not improve after N=10 consecutive evaluations, stop or reduce the learning rate.

[0263] (3) Learning rate decay strategy

[0264] Warmup: The first 1e5 steps linearly transition from 1e-4 to 3e-4.

[0265] Decay: cosine decay to 3e-5

[0266] Decay cycle: once for the entire training steps (or once every 1e7 steps).

[0267] Simulation verification

[0268] To verify the effectiveness of the hierarchical decision architecture and multi-scenario adaptive reward function of the proposed PPO-DDM dynamic fusion lane-changing decision model, this embodiment constructs a simulation implementation and verification platform based on the SUMO (Simulation of Urban Mobility) traffic simulator and reinforcement learning. This platform can simulate real road traffic environments with high fidelity, flexibly support the modeling and dynamic testing of various driving scenarios, and provide a solid technical foundation for the development and evaluation of intelligent driving algorithms.

[0269] (I) Construction of SUMO Traffic Simulation Scenario

[0270] To verify the model's performance in relatively realistic environments, the dataset scenarios used for model training were modeled proportionally in SUMO, constructing four typical highway driving scenarios: a vehicle slowing down, lane changing and merging, merging, and diverging. Different types of background vehicles were introduced into each scenario to simulate diverse driving behaviors in real traffic environments, increasing the policy challenges faced by the agent. Simulation scenario diagrams are shown below. Figure 2 As shown.

[0271] (1) Scenarios of the vehicle in front slowing down

[0272] This scenario simulates a common traffic situation where a vehicle ahead suddenly decelerates. Starting from step 5 of the simulation, the vehicle ahead decelerates uniformly from 16 m / s to 10 m / s over 30 steps. The deceleration uses a smooth non-linear curve with a deceleration rate of approximately -0.2 m / s². The assisted driving vehicle follows the vehicle ahead at a slightly lower speed (15 m / s) in the same lane, needing to flexibly adjust lane changes and acceleration during the vehicle's deceleration to ensure driving safety and efficiency. Every 20 steps, background vehicles are generated in adjacent lanes, with driving styles randomly selected, including normal (60%), cautious (30%), and aggressive (10%). Different vehicle types differ in speed (14-18 m / s), minimum safe following distance (1.5-3.0 m), etc., creating diverse traffic flows.

[0273] (2) Lane change insertion scene

[0274] This scenario primarily tests the agent's response to lane-changing behavior from vehicles in adjacent lanes, evaluating its perception and decision-making performance in lateral interaction scenarios. The preceding vehicle randomly departs from lane 1 or 3 (adjacent lanes), while the following vehicle (assisted driving vehicle) starts in lane 2 (middle lane). In the first step of the simulation, if the distance between the preceding and following vehicles exceeds 10 meters, a lane-changing operation is performed within 5 seconds, inserting into lane 2. This disrupts the agent's longitudinal following rhythm, forcing it to reassess safety and implement appropriate driving strategies. Every 20 steps, new vehicles are randomly generated in lanes 1 and 3, with speeds ranging from 14-18 m / s and vehicle spacing of 2-3 seconds to make the simulation more closely resemble real-world traffic conditions.

[0275] (3) Merging scenario

[0276] The merging scenario is primarily used to test the merging and lane-changing strategies of assisted driving vehicles. Vehicles on the ramp have an initial speed of 10 m / s and are positioned at 70% of the ramp's end, simulating a real-world scenario of about to enter the main road. The merging decision is adapted in stages, from "approaching the merging point" (<10 meters from the merging zone) to "merging in progress" and then to "merging complete." At each step, a different type of new main road vehicle is generated with a 10% probability, with a density controlled at 1200-1800 vehicles / hour / lane to ensure challenge and diversity during the merging process.

[0277] (4) Diversion scenario

[0278] The exit diversion scenario examines the agent's lane preparation and safe departure capabilities in a highway exit scenario. The assisted driving vehicle starts in lane 1 (middle lane), and exit guidance is triggered in step 10 of the simulation. The goal is to exit from lane 0 (the rightmost lane), simulating the advance warning signs and diversion requirements of a real highway exit area. Approximately 70% of the vehicles eventually exit, requiring the assisted driving vehicle to change lanes and slow down in advance. Some vehicles are traveling in exit lane 0 but continue to travel straight at high speed, creating an "exit lane occupancy" scenario, increasing the complexity of the agent's lane-changing and deceleration decisions.

[0279] (II) Phased Training Methods

[0280] To ensure the gradual adaptation of assisted driving vehicles under different risks and complexities, this embodiment introduces a four-stage learning mechanism: (1) The first stage only includes scenarios of deceleration and merging of the preceding vehicle, with low background vehicle density and large safety distance; (2) The second stage adds lane-changing scenarios, and assisted driving vehicles face more lateral interaction challenges; (3) The third stage introduces exit diversion scenarios, testing lane selection and safe lane-changing capabilities; (4) The fourth stage superimposes multiple scenarios, severe weather (rainy days), and high-density traffic to simulate complex real-world environments. The difficulty of the course is dynamically adjusted through background vehicle behavior characteristics (minimum distance, speed distribution, etc.) and traffic density parameters (0.3-0.8), forming a multi-level and diversified verification system. In addition to traffic scenarios, the platform supports randomized weather conditions (such as rainy days and sunny days), with the highest probability of rainy days being 30%. In rainy weather scenarios, the agent needs to dynamically adjust the following distance, safety threshold, and lane-changing intention to further improve its generalization ability and robustness. To improve learning stability, the platform incorporates state normalization and reward normalization modules to reduce oscillations and numerical shifts during training, ensuring model robustness and convergence.

[0281] (III) Simulation Interface

[0282] The simulation platform constructed in this embodiment strictly follows the Gymnasium standard, and the core interface of the encapsulated simulation environment includes the following parts:

[0283] (1) State space: contains a 12-dimensional vector, covering information such as vehicle speed, surrounding traffic, lane change safety, risk, current lane position, scene progress and weather;

[0284] (2) Action space: a two-dimensional vector, corresponding to acceleration and lane-changing decision respectively;

[0285] (3) Reward function: integrates multi-scenario adaptive rewards to respond in real time to vehicle safety, efficiency, comfort and social perception factors.

[0286] (iv) Comparative Analysis

[0287] This embodiment compares and analyzes the PPO-DDM dynamic fusion lane-changing decision model (PPO-DDM model) and the PPO reinforcement learning model (base model) across multiple dimensions. The comparative experiment covers 10,000 training episodes, comprehensively evaluating the models' performance in key metrics such as learning speed, performance stability, and environmental adaptability.

[0288] (1) Comparison of learning ability and convergence characteristics

[0289] like Figure 3 As shown, the PPO-DDM model exhibits relatively good initial performance (4.33) in the early training phase (Episode 100), while the base model starts from a negative value (-454.91), with an initial performance difference of 459.24 points. This indicates that the PPO-DDM model has significant advantages in parameter initialization and prior knowledge integration. The PPO-DDM model achieved a 37.44-point improvement from Episode 100 to 1000 Episodes, while the base model achieved a 282.35-point improvement during the same period. Although the absolute value is larger, the PPO-DDM model has a higher learning efficiency in terms of relative improvement rate. The PPO-DDM model only needs 52 Episodes to achieve 3 consecutive positive scores, demonstrating extremely fast positive learning ability, while the base model requires 957 Episodes, a difference of 18 times. This huge difference indicates that the PPO-DDM model has a significant advantage in rapid environmental adaptation. The PPO-DDM model achieved a peak performance of 202.89 in Episode 10000, which is 152.36 points higher than the base model's 50.53 (approximately a 4-fold performance difference), fully demonstrating the long-term learning advantage of the PPO-DDM model.

[0290] (2) Comprehensive evaluation of model performance

[0291] This embodiment evaluates the overall performance of the two models using four dimensions: average score, maximum score, standard deviation of score, and proportion of positive scores. Figure 4(As shown). First, the average score of the PPO-DDM model is 124.56, which is 60.46 points higher than the base model's 64.10, representing a 94.3% improvement. This significant difference indicates that the PPO-DDM architecture has a substantial advantage in overall performance stability. As the average score is a key indicator of a model's daily performance, the advantage of the PPO-DDM model means it can provide more reliable and consistent performance output in actual deployments. Second, the maximum score of the PPO-DDM model reaches 421.04, surpassing the base model's 302.72, a lead of 118.32 points (a 39.0% performance improvement). This result shows that the PPO-DDM fusion architecture not only performs excellently in average performance but also demonstrates strong potential in peak performance mining. The advantage in maximum score proves that the fusion architecture has a strong ability to handle complex scenarios and extreme cases. Furthermore, the standard deviation of the PPO-DDM model's score was 90.46, significantly lower than the base model's 150.55, a difference of 60.09 points (a 39.9% improvement in stability). The lower standard deviation indicates that the PPO-DDM model exhibits less performance fluctuation and higher reliability across different test scenarios. Finally, the PPO-DDM model achieved a positive score percentage of 65%, 10 percentage points higher than the base model's 55%. This metric demonstrates that the PPO-DDM model maintains stronger consistency in achieving positive performance, reflecting its superior learning effectiveness and environmental adaptability.

[0292] (3) Training stability and fluctuation characteristics

[0293] The PPO-DDM model demonstrated significant stability and robustness during reinforcement learning training. For example... Figure 5 As shown, the PPO-DDM model The average value is -0.15, significantly lower than the base model's -0.84, indicating a milder policy update and stronger convergence. The standard deviation decreased from 4.31 to 1.16, a 73.1% reduction in volatility, effectively avoiding oscillations during training and improving learning consistency and predictability. Regarding value network learning, the PPO-DDM model... The average value decreased from 2.07 in the base model to 0.54, and the error was reduced by 73.9%, indicating that it has a more accurate state value estimation capability; The standard deviation also decreased from 1.06 to 0.42, with volatility reduced by 60.4%, further enhancing the stability of value learning and providing more reliable gradient signals for the policy network. Overall, the PPO-DDM model achieved a dual reduction in both the mean and volatility of loss in both policy and value dimensions, forming a stable and efficient Actor-Critic training loop. This coordinated characteristic not only accelerates the model's convergence speed and reduces the time and computational resources required for training, but also improves the reliability and repeatability of model training, significantly optimizing the instability and performance degradation risks of the base model during training.

[0294] (4) Comparative analysis of model performance

[0295] A comprehensive comparative analysis was conducted on the overall performance metrics of the trained PPO-DDM model and the base model, as well as their performance in four typical traffic scenarios. The test scenarios included vehicle deceleration, lane changing / merging, merging, and diverging, with 1000 simulation experiments performed for each scenario. Through multi-dimensional evaluation, the PPO-DDM model demonstrated significant advantages in safety, response speed, and decision quality.

[0296] 1. Overall performance index comparison and analysis

[0297] Table 3 illustrates the performance improvements of the PPO-DDM model compared to the base model across four dimensions. The table employs both absolute and relative improvement indicators to provide a more comprehensive comparative perspective. The average decision consistency score improved from 7.2 points in the base model to 9.1 points in the PPO-DDM model, an absolute increase of 1.9 points and a relative increase of 26.4%, highlighting the significant advantages of the PPO-DDM model in the predictability and consistency of decision-making. The average response time decreased from 0.37 seconds to 0.22 seconds, an absolute reduction of 0.15 seconds, representing an improvement of 40.5%. In assisted driving applications, this 0.15-second difference corresponds to a braking distance of approximately 2.5 meters at 60 km / h, significantly improving safety in emergency situations. The average success rate increased from 85.7% to 94.1%, an absolute increase of 8.4%, indicating that while accelerating response speed, the PPO-DDM model still maintains a high level of decision success rate, achieving a balance between rapid response and high-quality decision-making.

[0298] Table 3 Comparison of Overall Performance Indicators

[0299]

[0300] Note: Positive relative improvement indicates performance improvement, negative response time indicates time reduction (performance improvement), the average decision consistency is based on the arithmetic mean of four scenarios, and all indicators have passed the statistical significance test.

[0301] 2. Comparative analysis of performance indicators in various scenarios

[0302] ①Scenario of the vehicle in front slowing down

[0303] In this scenario, the PPO-DDM model exhibited only 0.56 collision events, a 34% reduction compared to the base model, significantly lowering the probability of rear-end collisions. Simultaneously, the reduced average safe distance of 2.8 meters and decision delay, while ensuring safety, avoided the impact of overly conservative driving strategies on traffic efficiency. The 1.2 m / s speed adjustment range reflects a smoother and more gentle driving style (as shown in Table 4).

[0304] Table 4 Comparison of performance indicators in the scenario of deceleration of the vehicle in front

[0305] ②

[0306] ③ Insert lane change scene

[0307] In this scenario, the PPO-DDM model achieved a lane-changing success rate of 93.5%, demonstrating good gap judgment and lane-changing timing capabilities. Crucially, the number of collisions decreased from 1.8 to 0.7, a reduction of 61%, significantly lowering the risk of side collisions. The average lane-changing time of 3.2 seconds ensured both operational efficiency and smoothness during the lane-changing process. The minimum safe clearance of 1.8 meters further illustrates the comprehensiveness and accuracy of the system's safety assessment (as shown in Table 5).

[0308] Table 5 Performance Indicators Comparison for Lane Change Scenarios

[0309] ④

[0310] ⑤ Merging scenario

[0311] The PPO-DDM model achieved a 96.2% success rate in merging scenarios, significantly outperforming the base model's 92.5%, and reduced merging time by 12.7%, thus improving traffic efficiency. Meanwhile, the 87.6% gap acceptance rate reflects a good balance between risk control and efficiency, avoiding both overly conservative and overly aggressive merging operations. Particularly noteworthy is the reduction in the number of conflict scenarios from 3.2 to 2.6, a decrease of 23.1%, indicating a significant reduction in the interference of merging behavior on main road traffic flow, which is beneficial to improving the overall capacity and safety of the weaving area (as shown in Table 6).

[0312] Table 6 Comparison of Performance Indicators in Merging and Lane Changing Scenarios

[0313] ⑥

[0314] ⑦ Traffic splitting scenario

[0315] The PPO-DDM model demonstrated superior predictive capabilities in exit scenarios, achieving an advance lane-changing distance of 25 meters, a 21% improvement over the base model, effectively mitigating safety risks caused by sudden lane changes. Its exit deceleration smoothness score reached 8.9, indicating the algorithm's ability to plan deceleration in advance as it approaches the exit, further optimizing passenger comfort. Furthermore, the lane occupancy time (1.8 seconds) and rear vehicle impact index (1.2 seconds) were significantly lower than the base model, reflecting its superior performance in operational efficiency and human-machine interaction compatibility (as shown in Table 7).

[0316] Table 7 Comparison of Performance Indicators for Traffic Diversion and Lane Changing Scenarios

[0317]

[0318] The above verification and experiments show that, compared with the basic model, the PPO-DDM model can significantly improve the safety and consistency of decision-making and reduce the frequency of high-risk lane-changing behaviors in complex scenarios such as multi-vehicle interaction, traffic flow density fluctuations and driving intention uncertainty.

[0319] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A lane-changing decision-making method for assisted driving vehicles with human-like dynamic decision-making characteristics, characterized in that, Includes the following steps: Step 1: Obtain the vehicle's operating status information, the relative status information of surrounding vehicles, traffic scene information, and environmental context information. The traffic environment perception module then transmits this information to the multi-dimensional state space module. Step 2: The multi-dimensional state space module converts the vehicle operation status information output by the traffic environment perception module into vehicle characteristic codes, converts the relative status information of surrounding vehicles into environmental interaction codes, converts traffic scene information into traffic scene codes, converts environmental context information into environmental context codes, and finally realizes multi-dimensional feature fusion through encoder fusion, and outputs the fused result as a state vector and passes it to the dynamic fusion decision module. Step 3: The dynamic fusion decision module makes a decision on the next driving state of the vehicle based on the current state vector output result of the vehicle. The decision depends on the accumulation of evidence from the current state vector output result of the vehicle. When the evidence accumulates to a threshold, the lane change intention is determined and the result is output. The output result is the lane change intention, the total reaction time and the comprehensive confidence level. The dynamic fusion decision module includes a PPO-DDM decision submodule and a weight adaptive mechanism submodule. The PPO-DDM decision submodule includes a PPO decision submodule and a DDM decision submodule. The PPO decision submodule uses an Actor network to output lane-changing decision results and acceleration values ​​based on PPO. The DDM decision submodule outputs lane-changing decision results based on DDM. The output results of the PPO-DDM decision submodule are input to the weight adaptive mechanism submodule. The weight adaptive mechanism submodule dynamically allocates the lane-changing decision weights output by the PPO decision submodule and the DDM decision submodule according to changes in weather, traffic density, and risk level factors. In low-risk, low-density scenarios, the weight of the PPO decision submodule is increased, and in high-risk or severe weather conditions, the weight of the DDM decision submodule is increased, obtaining the final weighted fusion decision result as the module output. The DDM decision-making submodule includes a DDM model construction submodule, an environmental adaptive parameter adjustment mechanism submodule, an evidence accumulation process numerical implementation submodule, and a confidence level and decision quality quantification submodule. The results of the DDM model construction submodule and the environmental adaptive parameter adjustment mechanism submodule are input to the evidence accumulation process numerical implementation submodule for evidence accumulation, ultimately obtaining the lane-changing intention and total reaction time. The evidence accumulation results of the evidence accumulation process numerical implementation submodule are input to the confidence level and decision quality quantification submodule, which outputs the comprehensive confidence level.

2. The lane-changing decision-making method for assisted driving vehicles with human-like dynamic decision-making characteristics in heterogeneous traffic flow environments, as described in claim 1, is characterized in that... The formula for calculating the weighted fusion decision result is as follows: in, This represents the final, synthesized probability distribution of actions. The action probability distribution output by the PPO decision submodule; The action probability distribution of the DDM decision submodule; Hehe These are the weight coefficients for the PPO decision submodule and the DDM decision submodule, respectively. in, These are the default weights for DDM under the basic conditions; This represents the overall risk score; This is a normalized value for traffic density, reflecting the degree of congestion.

3. The lane-changing decision-making method for assisted driving vehicles with human-like dynamic decision-making characteristics in heterogeneous traffic flow environments, as described in claim 2, is characterized in that... The decision rules for the DDM model construction submodule are as follows: in, The accumulated evidence value at time n. For decision-making reaction time, The decision threshold represents the amount of evidence required to trigger a decision. Decision reaction time This includes the time it takes for evidence to first reach the threshold and the perception delay, calculated using the following formula: in, The time when the evidence first reaches the threshold. This is a non-decision-making time; The DDM model is: in, The accumulated evidence value at time n. For drift rate, It is random noise; Drift rate The calculation formula is: in, Indicates the first One source of evidence at any time The contribution of drift rate; For the first The weight of each source of evidence; This represents the total number of evidence sources.

4. The lane-changing decision-making method for assisted driving vehicles with human-like dynamic decision-making characteristics in heterogeneous traffic flow environments, as described in claim 3, is characterized in that: The environmental adaptive parameter adjustment mechanism submodule, based on the DDM model construction submodule, calculates model-related dynamic thresholds, including dynamic drift rate and adaptive decision threshold. The dynamic drift rate is calculated as follows: in, Based on the drift rate, This represents the model's basic drift rate bias term. These are the weighting coefficients; This represents the PPO network's intention preference for lane changing in the current state. in, This is the final drift rate; This is a weather correction factor; This is a correction factor for lane-changing type; The adaptive decision threshold calculation method is as follows: in, Basic decision threshold; This is a weather correction factor; This is the lane-changing type coefficient.

5. The lane-changing decision-making method for assisted driving vehicles with human-like dynamic decision-making characteristics in heterogeneous traffic flow environments, as described in claim 4, is characterized in that: The evidence includes evidence of PPO preference, evidence of safety urgency, and evidence of traffic efficiency; The method for calculating the PPO preference evidence is as follows: in, Indicates the state The probability of the PPO strategy choosing the "lane change" action; Indicates the state Under these conditions, the probability of the PPO strategy selecting the "lane keeping" action; The weighting coefficients for PPO preference evidence; The method for calculating the evidence of security urgency is as follows: in, The scenario-based instantaneous risk assessment method is used for calculation. Weighting coefficients for evidence of security urgency; The method for calculating the traffic efficiency evidence is as follows: in, As a traffic efficiency indicator, it can be calculated based on the difference in traffic flow density between lanes. The weighting coefficients for evidence of traffic efficiency; The weighting coefficients of the PPO preference evidence, the weighting coefficients of the safety urgency evidence, and the weighting coefficients of the traffic efficiency evidence satisfy the following: .

6. The lane-changing decision-making method for assisted driving vehicles with human-like dynamic decision-making characteristics in heterogeneous traffic flow environments, as described in claim 5, is characterized in that: The evidence accumulation is discretized using the Euler-Markov approximation, including the following steps: Step 31: Initialize the accumulated evidence value and time counter; Step 32: Within each time step, calculate the evidence increment consisting of the drift term and the diffusion term; The drift term is the product of the drift rate and the time step. The diffusion term is the product of a Gaussian random number and a diffusion coefficient; Step 33: Overlay the current cumulative evidence value and the evidence increment as a new cumulative evidence value. Compare the new cumulative evidence value with the decision threshold. When the evidence exceeds the threshold for the first time, determine the final decision of lane changing or lane keeping and its confidence level. If the threshold is not reached within the maximum time range, the direction is determined by default based on the evidence symbol, and the confidence level is set to 0.

5.

7. The lane-changing decision-making method for assisted driving vehicles with human-like dynamic decision-making characteristics in heterogeneous traffic flow environments, as described in claim 6, is characterized in that: In step 33, when the absolute value of the accumulated evidence value approaches the decision threshold, the accumulated decision value is considered to have entered the soft triggering range, and the soft boundary triggering probability is calculated. : in, Represents the absolute value of accumulated evidence. The decision threshold for changing lanes or staying straight. The sensitivity coefficient for triggering reflection boundary. This is the starting value of the soft-trigger interval. ; Based on the soft boundary trigger probability Adaptive correction is applied to the drift rate, i.e., based on the original drift rate... Based on the superposition and Additional drift terms in the same direction accelerate the arrival of evidence toward the current decision direction at the threshold, reduce the probability of prolonged oscillations around the threshold, and improve decision efficiency and stability. in, This is a sign indicator function used to indicate the direction of the current evidence accumulation value; when A timeout of +1 indicates a bias towards lane changing. When the value is negative, the output is −1, indicating bias to maintain lane position.

8. The lane-changing decision-making method for assisted driving vehicles with human-like dynamic decision-making characteristics in heterogeneous traffic flow environments, as described in claim 1, is characterized in that: The method for calculating the overall confidence level is as follows: in, For the weighting coefficients, satisfying ; The strength of evidence is calculated as follows: in, For the final accumulated value of evidence, This is the decision threshold; To ensure consistency in decision-making time, the calculation method is as follows: in, For the actual decision-making time, For the expected value based on experience, This is a sensitivity parameter for time consistency. For drift rate stability, the calculation method is as follows: in, and Let Variance and mean of the drift rate be represented, respectively. To prevent small constants with a denominator of zero.

Images