Automatic driving vehicle lane changing method considering surrounding vehicle interaction

Abstract

The invention provides an automatic driving vehicle lane changing method considering surrounding vehicle interaction. The method comprises the steps of obtaining state information of an automatic driving vehicle and other vehicles in the same scene at the current moment; driving behavior characteristics are analyzed, a fuzzy reasoning method is adopted to estimate a vehicle aggressiveness coefficient, and a social potential field model considering a driving style is constructed; based on collision risk assessment and vehicle interaction behavior analysis, constructing a non-cooperative game model, and solving non-cooperative game equilibrium to obtain an optimal lane changing decision and acceleration; taking the longitudinal acceleration of the lane changing decision variable output by the game layer as a reference input signal of an NMPC controller; and based on the vehicle kinematic model and the potential field information, generating a driving track of the autonomous vehicle by using an NMPC controller and outputting corresponding motion planning information. The upper layer makes a decision through a game model, and the lower layer performs transverse and longitudinal coupling planning based on nonlinear model predictive control, so that smooth and controllable path tracking is realized, and the real-time performance and stability of the system are improved.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a lane-changing method for autonomous vehicles that takes into account vehicle-to-vehicle interaction. Background Technology

[0002] With the rapid development of intelligent connected and autonomous driving technologies, vehicles have evolved from simple longitudinal following to comprehensive control systems encompassing complex driving behaviors such as lane changing, overtaking, and obstacle avoidance. Lane changing is a fundamental driving operation, crucial for ensuring driving safety and improving vehicle efficiency. In autonomous driving technology, decision planning is a core function. It utilizes relevant algorithmic models to integrate surrounding environmental information, vehicle position, speed, and time factors to generate safe and effective decisions. This approach not only ensures the safety of lane changing but also significantly improves its efficiency, contributing to better overall traffic flow and enhancing driving comfort.

[0003] The complexity of lane-changing decisions lies in their inherent interactive characteristics. In real-world traffic scenarios, autonomous vehicles do not exist in isolation but rather in a dynamic and uncertain environment comprised of other vehicles, pedestrians, and traffic infrastructure. Lane-changing operations require significant maneuvers in both the lateral and longitudinal dimensions, inevitably triggering reactions from surrounding vehicles and creating a continuous game-theoretic process. The vehicle needs to predict the potential behaviors of other vehicles and adjust its lane-changing strategy accordingly to avoid collisions while completing the lane change as efficiently as possible. Furthermore, the varying driving styles and intentions of different drivers (including human drivers and autonomous driving systems) further increase the difficulty of environmental perception and the uncertainty of prediction. Making safe, efficient, and comfortable decisions in such a complex environment is a significant challenge facing the field of autonomous driving.

[0004] First, most current lane-changing game theory methods focus primarily on the longitudinal distance or acceleration competition between the lead vehicle and the obstacle vehicle, lacking effective connection with the generation of the vehicle's lateral trajectory. This leads to the risk of abrupt changes, inconsistencies, or even violations of kinematic feasibility at the control execution level, reducing the actual feasibility of lane-changing strategies.

[0005] Secondly, in the existing NMPC control framework, obstacle vehicles are often simplified as static objects or rule-driven entities. The "leader-follower" structure and game behavior modeling are not introduced, nor are the differences in driving styles between individual vehicles considered. This makes it difficult to truly reflect the dynamic game behavior in complex traffic flow environments, thus affecting the robustness and prediction accuracy of obstacle avoidance strategies.

[0006] Finally, some rule-driven or simple objective function optimization methods only constrain the minimum safe distance in the design, without fully incorporating passenger comfort (such as changes in lateral and longitudinal acceleration) into the evaluation and control objectives of lane-changing strategies. This can easily lead to uncomfortable actions such as sudden acceleration and sharp turns while successfully avoiding obstacles, thus reducing the riding experience. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides an autonomous vehicle lane-changing method that considers vehicle-to-vehicle interaction. This invention integrates non-cooperative game theory and nonlinear models to achieve closed-loop coupling between lane-changing decisions and trajectory planning, thereby improving risk perception and obstacle avoidance robustness in complex multi-vehicle interaction environments.

[0008] The technical solution of the present invention is: a lane-changing method for autonomous vehicles considering vehicle-to-vehicle interaction, comprising the following steps:

[0009] S1) Obtain the current state information of the current autonomous vehicle and other vehicles in the same scene as the current autonomous vehicle;

[0010] S2) Based on the NGSIM dataset, driving behavior characteristics are analyzed, and the vehicle aggression coefficient is estimated using fuzzy inference methods to construct a social potential field model that considers driving style.

[0011] S3) Based on collision risk assessment and vehicle interaction behavior analysis, a non-cooperative game model is constructed, and the non-cooperative game equilibrium is solved to obtain the optimal lane-changing decision and acceleration.

[0012] S4) Use the longitudinal acceleration of the lane-changing decision variable output by the game layer as the reference input signal for the NMPC controller;

[0013] S5) Based on the vehicle kinematics model and potential field information, the NMPC controller is used to generate the driving trajectory of the autonomous vehicle and output the corresponding motion planning information.

[0014] Preferably, in step S2), driving behavior characteristics are analyzed based on the NGSIM dataset. Specifically, vehicle speed and vehicle acceleration motion parameters are extracted from the NGSIM dataset, and the vehicle speed distribution and acceleration change characteristics under different driving scenarios are analyzed to obtain the statistical distribution characteristics of speed and acceleration parameters under each driving scenario. Based on the statistical distribution characteristics, the range of values ​​of vehicle motion parameters under different traffic conditions is determined, and a classification standard for speed and acceleration intervals is established.

[0015] Preferably, in step S2), the vehicle aggression coefficient is estimated using a fuzzy inference method. Specifically, the vehicle aggression coefficient ρ is estimated by inputting the longitudinal velocity and longitudinal acceleration of adjacent vehicles into a fuzzy controller. The driving aggression is quantified using three types of fuzzy values: conservative, normal, and aggressive. The velocity and acceleration are fuzzified into five levels using S-shaped and triangular membership functions. Fuzzy inference rules are established to form a mapping relationship between velocity, acceleration, and aggression, thereby obtaining the aggression coefficient ρ.

[0016] Preferably, in step S2), the expression for the social potential field model is:

[0017] P = P NV (X,Y)+P r (Y);

[0018] In the formula, P is the total potential energy field, P NV (X,Y) represents the potential energy field of adjacent vehicles, with the superscript NV indicating other vehicles adjacent to or near the main vehicle; P r (Y) represents the potential energy field at the road boundary, and (X,Y) represents the coordinate position of the main vehicle.

[0019] Preferably, in step S2), the expression for the potential energy field of the adjacent vehicles is:

[0020]

[0021] In the formula, (X NV ,Y NV () indicates the position of adjacent vehicles; λ is a function of pre-collision time and driving initiative; λ0 represents the control parameter; ∈ represents a local minimum; TTC represents pre-collision time; ξ is a parameter related to vehicle length and driving aggression; y =a y W NV For parameters related to vehicle width; L NV W is the length of adjacent vehicles; NV The width of adjacent vehicles; ρ represents the vehicle aggression coefficient; a is the proportionality coefficient; a x a y These are the x and y axis components of the scaling factor a.

[0022] Preferably, in step S2), the road boundary potential energy field P r The expression for (Y) is:

[0023] P r (Y)=exp(-(Y c -|YY c |) 2)+ηexp(-(YY c ) 2 );

[0024] In the formula, Y c η represents the lateral position coordinates of the lane centerline; η represents the scaling factor.

[0025] Preferably, in step S3), a non-cooperative game model is constructed based on collision risk assessment and vehicle interaction behavior analysis, specifically including the following steps:

[0026] S31) The vehicle is taken as a game participant and the game strategy space is defined; the game strategy of the main vehicle includes three lateral decisions: changing lanes to the left, keeping lanes, and changing lanes to the right, and three longitudinal decisions: accelerating, decelerating, and maintaining speed.

[0027] The game strategy of adjacent vehicles includes the corresponding acceleration adjustment range;

[0028] S32) Construct the total cost function of the main vehicle C HV The total cost function C of adjacent vehicles NV The collision risk is calculated based on the time to collision (TTC) before the collision and then embedded into the safety cost function.

[0029] S33) By analyzing vehicle interaction behavior, establish the mutual influence relationship between vehicles, including longitudinal following behavior and lateral lane change response behavior.

[0030] Preferably, in step S32), the total cost function C of the main vehicle in the non-cooperative game model... HV for:

[0031]

[0032] In the formula, The cost function for the safety of the main vehicle, with the superscript HV indicating the main vehicle; The main vehicle comfort cost function; The main vehicle traffic efficiency cost function; The main vehicle safety weighting coefficient; The main vehicle comfort weighting coefficient; The main vehicle efficiency weighting coefficient.

[0033] Preferably, in step S32), the cost function of the main vehicle safety is... Represented as:

[0034]

[0035] In the formula, σ = {-1, 0, 1} represents {left lane change, no lane change, right lane change}; This represents the longitudinal cost of the main vehicle; This indicates the lateral cost of the main vehicle; The longitudinal safety factor of the main vehicle; This is the longitudinal relative speed between the vehicle in front and the main vehicle in this lane; The longitudinal safety distance between the vehicle in front and the main vehicle in this lane; ε is the regularization parameter to avoid division by zero; TTC HV Time before the main vehicle collision; T safe For safe time thresholds; The weighting coefficient for the main vehicle; The longitudinal relative speed between the vehicle in front and the main vehicle; The longitudinal safety distance between the vehicle in front and the vehicle in front; TTC NV This represents the time before the collision between adjacent vehicles.

[0036] Preferably, in step S32), the main vehicle comfort cost function... Represented as:

[0037]

[0038] In the formula, The longitudinal and lateral weighting coefficients for the main vehicle. The longitudinal acceleration of the main vehicle, This is lateral acceleration.

[0039] Preferably, in step S32), the main vehicle traffic efficiency cost function Represented as:

[0040]

[0041] In the formula, This is the longitudinal speed of the vehicle; The longitudinal speed of the vehicle in front; This represents the maximum longitudinal speed of the lane.

[0042] Preferably, in step S32), the adjacent vehicle total cost function C of the non-cooperative game model... NV Represented as:

[0043]

[0044] In the formula, A cost function that takes into account the safety of adjacent vehicles; The comfort cost function for adjacent vehicles; The traffic efficiency cost function for adjacent vehicles; The safety weighting coefficient for adjacent vehicles. The comfort weighting coefficient for adjacent vehicles. This is the efficiency weighting coefficient for adjacent vehicles.

[0045] Preferably, in step S32), the cost function that considers the safety of adjacent vehicles... Represented as:

[0046]

[0047] In the formula, σ = {-1, 0, 1} represents {left lane change, no lane change, right lane change}; Indicates the longitudinal cost of adjacent vehicles; Indicates the lateral cost of adjacent vehicles; The longitudinal safety factor for adjacent vehicles; The longitudinal relative speed between the vehicle in front and the adjacent vehicle; The longitudinal safety distance between the vehicle in front and the adjacent vehicle; TTC NV The time before the collision of adjacent vehicles; T safe ε is the safe time threshold; ε is the regularization parameter to avoid division by zero.

[0048] Preferably, in step S3), the adjacent vehicle comfort cost function Represented as:

[0049]

[0050] In the formula, The weighting coefficients for adjacent vehicles. This represents the longitudinal acceleration of the adjacent vehicle.

[0051] Preferably, in step S3), the adjacent vehicle traffic efficiency cost function Represented as:

[0052]

[0053] In the formula, The longitudinal speed of the adjacent vehicle This represents the maximum longitudinal speed of the lane.

[0054] Preferably, in step S3), a non-cooperative dynamic game theory method is used, with the lead vehicle as the leader and adjacent vehicles as followers, to solve for the non-cooperative game equilibrium and obtain the optimal lane-changing decision σ. * and optimal longitudinal acceleration As a reference input for the lower-level NMPC controller, it is specifically as follows:

[0055]

[0056] In the formula, σ * The optimal lane-changing strategy for the main vehicle serves as the overtaking decision. This represents the optimal longitudinal acceleration of the main vehicle; a NVThe acceleration of the vehicles participating in the game is defined by the boundary of the acceleration that the vehicles can achieve; C HV With C NV γ represents the total decision cost of the game between the main vehicle and its adjacent vehicles; 2 (a HV σ) represents the set of optimal decision candidate actions made by other vehicles following the decision action of the master vehicle; σ∈{-1,0,1} represents the master vehicle changing lanes to the left, following in the original lane, and changing lanes to the right, respectively; a HV The acceleration of the main vehicle; Indicates γ 2 (a HV ,σ) is defined as the set described in the following curly braces; These are the minimum and maximum longitudinal acceleration constraints for the vehicle on road number μ, respectively; These are the minimum and maximum longitudinal speed limits for vehicles in lane μ, respectively.

[0057] Preferably, in step S5), a vehicle kinematic model is constructed with the vehicle's longitudinal position, lateral position, heading angle, and longitudinal velocity as state variables, and the front wheel steering angle and longitudinal acceleration as control variables. A composite potential field function of the vehicle's repulsive potential field and the lane centerline constraint potential field is established. A nonlinear model predictive control (NMPC) architecture is adopted. Under the premise of satisfying the vehicle kinematic constraints, control input constraints, collision safety constraints, and lateral deviation of the reference path constraints, the optimal control sequence is solved through a rolling time-domain optimization strategy to realize the trajectory planning and motion control of the autonomous vehicle in the dynamic traffic environment, and output complete motion planning information including position trajectory, velocity, heading angle sequence, and control commands.

[0058] Preferably, in step S5), the vehicle kinematic model is:

[0059]

[0060] In the formula, ζ kin(τ+1) X represents the state vector of the vehicle at time τ+1; kin (τ+1) represents the longitudinal position coordinate of the vehicle at time τ+1; Y kin (τ+1) represents the lateral position coordinate of the vehicle at time τ+1;

[0061] δ represents the heading angle of the vehicle at time τ+1; kin (τ+1) represents the front wheel steering angle of the vehicle at time τ+1; Δt represents the discretization time step; v xkin (τ) represents the longitudinal velocity of the vehicle at time τ; δ represents the heading angle of the vehicle at time τ; fkin(τ) represents the front wheel steering angle of the vehicle at time τ; L kin τ represents the vehicle wheelbase; τ represents the current discrete time.

[0062] Preferably, in step S5), the risk field is embedded into the output of the vehicle model, and the risk assessment result is used as the objective function term in the nonlinear model predictive control (NMPC) to drive the planning result away from the high-risk area, i.e.:

[0063]

[0064] In the formula, y kin (τ) represents the potential energy field output value of the vehicle at time τ, which serves as the risk assessment term in the NMPC objective function; g represents the potential energy field mapping function, which maps the vehicle state and control input to risk potential energy values; u kin (τ) represents the vehicle's control input vector at time τ, including the front wheel steering angle and longitudinal acceleration; P represents the total potential energy field function, which consists of the potential energy fields of neighboring vehicles and the potential energy field of the road boundary.

[0065] Preferably, in step S5), by minimizing the cost function J(τ), which comprehensively considers path risk assessment, course change smoothness, control input overhead, and road-keeping error, a collision-free and dynamically feasible reference path is generated after solving the optimization problem in each time period.

[0066]

[0067] In the formula, N p Indicates the prediction time domain; It represents the potential field strength and is used to guide vehicles to move along the potential field gradient and avoid high potential energy regions. The penalty term representing the change in yaw rate is used to suppress drastic changes in vehicle yaw motion and ensure driving stability; N c Indicates control over the time domain; This represents a penalty term for controlling changes in the input. The penalty for lateral position deviation is used to keep the vehicle within the target lane; v max Indicates the maximum permissible speed; v represents the current vehicle speed; u represents the maximum permissible speed. min u max These represent the minimum and maximum constraints of the control input, respectively; u(τ+i|τ) represents the predicted value of the control input at time τ to the future time τ+i. Represents the optimal control input sequence; u represents the optimal control input increment at time τ; kin (τ|τ) represents the actual control input at the current moment; This represents the optimal control input increment vector.

[0068] The beneficial effects of this invention are as follows:

[0069] 1. This invention constructs a game model between the master vehicle and the target vehicle, and introduces driving style parameters to realize response prediction and strategy optimization of the target vehicle's behavior, which has stronger adaptability and safety in multi-vehicle interaction scenarios.

[0070] 2. This invention significantly improves the accuracy of game strategy by introducing driving style parameters to model the opponent's reaction behavior; at the same time, it introduces nonlinear model predictive control (NMPC) and multi-source potential field perception mechanism in the control layer to dynamically model obstacles, road boundaries and driving risks, and guide the master vehicle to generate a smooth and risk-controllable trajectory.

[0071] 3. This invention achieves smooth and controllable path tracking by making decisions based on a game theory model at the upper level and performing horizontal and vertical coupled planning based on a nonlinear model predictive control at the lower level, thereby improving the real-time performance and stability of the system. Attached Figure Description

[0072] Figure 1 This is a schematic flowchart of the method of the present invention;

[0073] Figure 2 This is a flowchart illustrating the fuzzy reasoning method of the present invention;

[0074] Figure 3 This is a flowchart of the non-cooperative game model of the present invention;

[0075] Figure 4 This is a schematic diagram illustrating the construction of the vehicle kinematics model of the present invention;

[0076] Figure 5 This is a simulation diagram of the game between the main vehicle and the target lane (HV vs NV+LV) in this invention;

[0077] Figure 6 This is a simulation diagram of the game between the main vehicle and the target lane (HV vs NV+LV1+LV2) of the present invention;

[0078] Figure 7 This is a simulation diagram of the game between the main vehicle and the target lane (HV vs NV+LV1+LV2) of the present invention;

[0079] Figure 8 This is a schematic diagram showing the speed of each vehicle in three cases in the embodiments of the present invention. Detailed Implementation

[0080] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0081] like Figure 1As shown, this embodiment provides a lane-changing method for autonomous vehicles that considers vehicle-to-vehicle interaction, including the following steps:

[0082] S1) Obtain the current state information of the current autonomous vehicle and other vehicles in the same scene as the current autonomous vehicle;

[0083] The status information includes: the vehicle's longitudinal position coordinates, lateral position coordinates, heading angle, longitudinal velocity, lateral velocity, longitudinal acceleration, lateral acceleration, yaw rate, front wheel steering angle, vehicle geometric parameters, and historical information on the vehicle's motion trajectory.

[0084] S2) Based on the NGSIM dataset, driving behavior characteristics are analyzed, and the vehicle aggression coefficient is estimated using fuzzy inference methods to construct a social potential field model that considers driving style.

[0085] In this embodiment, driving behavior features are analyzed based on the NGSIM dataset, specifically as follows:

[0086] Vehicle speed and acceleration motion parameters were extracted from the NGSIM dataset, and the distribution of vehicle speed and acceleration variation characteristics under different driving scenarios were analyzed to obtain the statistical distribution characteristics of speed and acceleration parameters under each driving scenario. Based on the statistical distribution characteristics, the range of values ​​of vehicle motion parameters under different traffic conditions was determined, and a classification standard for speed and acceleration intervals was established.

[0087] like Figure 2 As shown, the fuzzy inference method is used to estimate the vehicle aggression coefficient. Specifically, the longitudinal velocity and longitudinal acceleration of adjacent vehicles are input into the fuzzy controller to estimate the vehicle aggression coefficient ρ. The driving aggression is quantified using three types of fuzzy values: conservative, normal, and aggressive. The velocity and acceleration are fuzzified into five levels through S-shaped and triangular membership functions. Fuzzy inference rules are established to form a mapping relationship between velocity, acceleration, and aggression, thereby obtaining the aggression coefficient ρ.

[0088] In this embodiment, the driving styles include aggressive, moderate, and conservative, and the expression for the social potential field model is:

[0089] P = P NV (X,Y)+P r (Y);

[0090] In the formula, P is the total potential energy field, P NV (X,Y) represents the potential energy field of adjacent vehicles, with the superscript NV indicating other vehicles adjacent to or near the main vehicle; P r (Y) represents the potential energy field at the road boundary, and (X,Y) represents the coordinate position of the main vehicle.

[0091] The expression for the potential energy field of the adjacent vehicles is as follows:

[0092]

[0093] In the formula, (X NV ,Y NV () indicates the position of adjacent vehicles; λ is a function of pre-collision time and driving initiative; λ0 represents the control parameter; ∈ represents a local minimum; TTC represents pre-collision time; ξ is a parameter related to vehicle length and driving aggression; y =a y W NV For parameters related to vehicle width; L NV W is the length of adjacent vehicles; NV The width of adjacent vehicles; ρ represents the vehicle aggression coefficient; a is the proportionality coefficient; a x a y These are the x and y axis components of the scaling factor a.

[0094] Wherein, the road boundary potential energy field P r The expression for (Y) is:

[0095] P r (Y)=exp(-(Y c -|YY c |) 2 )+ηexp(-(YY c ) 2 );

[0096] In the formula, Y c η represents the lateral position coordinates of the lane centerline; η represents the scaling factor.

[0097] S3) Based on collision risk assessment and vehicle interaction behavior analysis, a non-cooperative game model is constructed, and the non-cooperative game equilibrium is solved to obtain the optimal lane-changing decision and the optimal lateral acceleration.

[0098] like Figure 3 As shown, in this embodiment, a non-cooperative game model is constructed based on collision risk assessment and vehicle interaction behavior analysis, specifically including the following steps:

[0099] S31) The vehicle is taken as a game participant and the game strategy space is defined; the game strategy of the main vehicle includes three lateral decisions: changing lanes to the left, keeping lanes, and changing lanes to the right, and three longitudinal decisions: accelerating, decelerating, and maintaining speed.

[0100] The game strategy of adjacent vehicles includes the corresponding acceleration adjustment range;

[0101] S32) Construct the total cost function of the main vehicle C HV The total cost function C of adjacent vehicles NV The collision risk is calculated based on the time to collision (TTC) before the collision and then embedded into the safety cost function.

[0102] Wherein: the total cost function of the main vehicle in the non-cooperative game model is c. HV for:

[0103]

[0104] In the formula, The cost function for the safety of the main vehicle, with the superscript HV indicating the main vehicle; The main vehicle comfort cost function; The main vehicle traffic efficiency cost function; The main vehicle safety weighting coefficient; The main vehicle comfort weighting coefficient; The main vehicle efficiency weighting coefficient.

[0105] The cost function of the main vehicle safety mentioned above. Represented as:

[0106]

[0107] In the formula, σ = {-1, 0, 1} represents {left lane change, no lane change, right lane change}; This represents the longitudinal cost of the main vehicle; This indicates the lateral cost of the main vehicle; The longitudinal safety factor of the main vehicle; This is the longitudinal relative speed between the vehicle in front and the main vehicle in this lane; The longitudinal safety distance between the vehicle in front and the main vehicle in this lane; ε is the regularization parameter to avoid division by zero; TTC HV Time before the main vehicle collision; T safe For safe time thresholds; The weighting coefficient for the main vehicle; The longitudinal relative speed between the vehicle in front and the main vehicle; The longitudinal safety distance between the vehicle in front and the vehicle in front; TTC NV This represents the time before the collision between adjacent vehicles.

[0108] in:

[0109]

[0110]

[0111] In the formula, a HV The current longitudinal acceleration of the main vehicle; a LVThis represents the current longitudinal acceleration of the vehicle in front; the superscript LV indicates the vehicle in front. HV This indicates the current longitudinal speed of the main vehicle; v LV Indicates the current longitudinal speed of the vehicle in front; This is the longitudinal speed of the vehicle; The longitudinal speed of the vehicle in front; This indicates the coordinates of the vehicle ahead at the location marked μ on the road; This indicates the coordinates of the main vehicle's position on the road marked μ; l o This represents the safety factor taking into account vehicle length; v NV Indicates the current longitudinal speed of the adjacent vehicle; a NV Indicates the current longitudinal acceleration of the adjacent vehicle; This represents the longitudinal velocity of adjacent vehicles on road number μ+σ; This represents the longitudinal speed of the main vehicle on road number σ; This represents the longitudinal and lateral coordinates of adjacent vehicles at the road number μ+σ.

[0112] The aforementioned vehicle comfort cost function Represented as:

[0113]

[0114] In the formula, The longitudinal and lateral weighting coefficients for the main vehicle. The longitudinal acceleration of the main vehicle, This is lateral acceleration.

[0115] The aforementioned vehicle traffic efficiency cost function Represented as:

[0116]

[0117] In the formula, This is the longitudinal speed of the vehicle; The longitudinal speed of the vehicle in front; This represents the maximum longitudinal speed of the lane.

[0118] In this embodiment, the total cost function C of adjacent vehicles in the non-cooperative game model is... NV Represented as:

[0119]

[0120] In the formula, A cost function that takes into account the safety of adjacent vehicles; The comfort cost function for adjacent vehicles; The traffic efficiency cost function for adjacent vehicles; The safety weighting coefficient for adjacent vehicles. The comfort weighting coefficient for adjacent vehicles. This is the efficiency weighting coefficient for adjacent vehicles.

[0121] The cost function that considers the safety of adjacent vehicles Represented as:

[0122]

[0123] In the formula, σ = {-1, 0, 1} represents {left lane change, no lane change, right lane change}; Indicates the longitudinal cost of adjacent vehicles; Indicates the lateral cost of adjacent vehicles; The longitudinal safety factor for adjacent vehicles; The longitudinal relative speed between the vehicle in front and the adjacent vehicle; The longitudinal safety distance between the vehicle in front and the adjacent vehicle; TTC NV The time before the collision of adjacent vehicles; T safe ε is the safe time threshold; ε is the regularization parameter to avoid division by zero.

[0124] in:

[0125]

[0126] In the formula, This represents the longitudinal velocity of the vehicle ahead at position μ+σ on road. This represents the longitudinal velocity of the adjacent vehicle at position μ+σ on road. Represents the longitudinal relative speed between the vehicle in front and the adjacent vehicle. This represents the longitudinal and lateral coordinates of adjacent vehicles at the location of road number μ+σ;

[0127] This represents the longitudinal and lateral coordinates of the vehicle ahead at the location of road number μ+σ; Indicates the longitudinal safety distance between the vehicle in front and the adjacent vehicle; This represents the longitudinal safety factor of adjacent vehicles, with a safety factor of 1 or 0, used to determine whether there is a risk of a rear-end collision; when When the speed of the adjacent vehicle is greater than that of the vehicle in front, there is a risk of a rear-end collision; when When there is no risk of a rear-end collision, l o This represents the safety factor taking into account the vehicle length.

[0128] The adjacent vehicle comfort cost function Represented as:

[0129]

[0130] In the formula, The weighting coefficients for adjacent vehicles. This represents the longitudinal acceleration of the adjacent vehicle.

[0131] The adjacent vehicle traffic efficiency cost function Represented as:

[0132]

[0133] In the formula, The longitudinal speed of the adjacent vehicle This represents the maximum longitudinal speed of the lane.

[0134] S33) By analyzing vehicle interaction behavior, establish the mutual influence relationship between vehicles, including longitudinal following behavior and lateral lane change response behavior.

[0135] A non-cooperative dynamic game theory approach is used, with the lead vehicle as the leader and adjacent vehicles as followers, to solve for the non-cooperative game equilibrium and obtain the optimal lane-changing decision σ. * and optimal longitudinal acceleration As a reference input for the lower-level NMPC controller, it is specifically as follows:

[0136]

[0137]

[0138] In the formula, σ * The optimal lane-changing strategy for the main vehicle serves as the overtaking decision. This represents the optimal longitudinal acceleration of the main vehicle; a NV The acceleration of the vehicles participating in the game is defined by the boundary of the acceleration that the vehicles can achieve; C HV With C NV γ represents the total decision cost of the game between the main vehicle and its adjacent vehicles; 2 (a HV σ) represents the set of optimal decision candidate actions made by other vehicles following the decision action of the master vehicle; σ∈{-1,0,1} represents the master vehicle changing lanes to the left, following in the original lane, and changing lanes to the right, respectively; a HV The acceleration of the main vehicle; Indicates γ 2 (a HV ,σ) is defined as the set described in the following curly braces; These are the minimum and maximum longitudinal acceleration constraints for the vehicle on road number μ, respectively; These are the minimum and maximum longitudinal speed limits for vehicles in lane μ, respectively.

[0139] S4) Use the longitudinal acceleration of the lane-changing decision variable output by the game layer as the reference input signal for the NMPC controller;

[0140] S5) Based on the vehicle kinematics model and potential field information, the NMPC controller is used to generate the driving trajectory of the autonomous vehicle and output the corresponding motion planning information.

[0141] In this embodiment, a vehicle kinematic model is constructed with the vehicle's longitudinal position, lateral position, heading angle, and longitudinal velocity as state variables, and the front wheel steering angle and longitudinal acceleration as control variables. A composite potential field function of the vehicle's repulsive potential field and the lane centerline constraint potential field is established. A nonlinear model predictive control (NMPC) architecture is adopted. Under the premise of satisfying vehicle kinematic constraints, control input constraints, collision safety constraints, and lateral deviation of the reference path constraints, the optimal control sequence is solved through a rolling time-domain optimization strategy. This enables the autonomous vehicle to perform trajectory planning and motion control in a dynamic traffic environment and outputs complete motion planning information including position trajectory, velocity, heading angle sequence, and control commands.

[0142] Among them, such as Figure 4 As shown, the vehicle kinematic model is as follows:

[0143]

[0144] In the formula, X represents the state vector of the vehicle at time τ+1; kin (τ+1) represents the longitudinal position coordinate of the vehicle at time τ+1; Y kin (τ+1) represents the lateral position coordinate of the vehicle at time τ+1; δ represents the heading angle of the vehicle at time τ+1; kin (τ+1) represents the front wheel steering angle of the vehicle at time τ+1; Δt represents the discretization time step; v xkin (τ) represents the longitudinal velocity of the vehicle at time τ; δ represents the heading angle of the vehicle at time τ; fkin (τ) represents the front wheel steering angle of the vehicle at time τ; L kin τ represents the vehicle wheelbase; τ represents the current discrete time.

[0145] By embedding the risk field into the vehicle model output, the risk assessment results are used as the objective function term in the nonlinear model predictive control (NMPC), driving the planning results away from high-risk areas.

[0146]

[0147] In the formula, y kin (τ) represents the potential energy field output value of the vehicle at time τ, which serves as the risk assessment term in the NMPC objective function; g represents the potential energy field mapping function, which maps the vehicle state and control input to risk potential energy values; u kin(τ) represents the vehicle's control input vector at time τ, including the front wheel steering angle and longitudinal acceleration; P represents the total potential energy field function, which consists of the potential energy fields of neighboring vehicles and the potential energy field of the road boundary.

[0148] By minimizing the cost function J(τ), which comprehensively considers path risk assessment, course change smoothness, control input overhead, and road-keeping error, a collision-free and dynamically feasible reference path is generated after solving the optimization problem in each time period.

[0149]

[0150] In the formula, N p Indicates the prediction time domain; It represents the potential field strength and is used to guide vehicles to move along the potential field gradient and avoid high potential energy regions. The penalty term representing the change in yaw rate is used to suppress drastic changes in vehicle yaw motion and ensure driving stability; N c Indicates control over the time domain; This represents a penalty term for controlling changes in the input. The penalty for lateral position deviation is used to keep the vehicle within the target lane; v max Indicates the maximum permissible speed; v represents the current vehicle speed; u represents the maximum permissible speed. min u max These represent the minimum and maximum constraints of the control input, respectively; u(τ+i|τ) represents the predicted value of the control input at time τ to the future time τ+i. Represents the optimal control input sequence; u represents the optimal control input increment at time τ; kin (τ|τ) represents the actual control input at the current moment; This represents the optimal control input increment vector.

[0151] Example 2

[0152] This embodiment verifies the effectiveness of the proposed method by constructing multi-vehicle interactive lane-changing scenarios with different driving styles. The simulation includes three typical cases, each configured with traffic participants of varying aggression levels, to verify the adaptability and robustness of the driving style-based game-theoretic lane-changing algorithm in complex traffic environments. Figure 5-8 As shown:

[0153] Scene setup:

[0154] Case Study 1: HV vs NV (Intense) + LV (Cautious)

[0155] Case Study 2: HV vs NV (Intense) + LV1 (Intense) + LV2 (Cautious)

[0156] Case Study 3: HV vs NV1 (Standard) + NV2 (Standard) + LV (Standard)

[0157] Each case was conducted on a simulated road 250 meters long, with a time span of 10 seconds. The lanes were distributed at three locations: Y = -2m, Y = 2m, and Y = 6m. The aggression coefficient ρ of different vehicles was identified through a fuzzy inference system and embedded into the social potential field model.

[0158] Simulation Result Analysis

[0159] 1. Trajectory Planning Results

[0160] Case 1: For example Figure 5 As shown, in a mixed cautious-aggressive driving style environment, the driver vehicle (HV) successfully changed lanes from its initial position Y=6m to Y=2m. Facing a cautious leading vehicle (LV) (maintaining a stable trajectory at Y=6m) and an aggressive adjacent vehicle (NV) (located at Y=2m), the driver vehicle smoothly changed lanes downwards. Due to the cautious characteristics of LV (low aggression coefficient ρ), its potential field range is relatively small, providing ample conditions for the driver vehicle to leave its current lane; while the aggressive NV, due to its high aggression coefficient, generates a stronger repulsive potential field, allowing the driver vehicle to successfully insert into the target lane through precise timing.

[0161] Case 2: For example Figure 6 As shown, in a highly complex environment with a mix of aggressive and cautious driving styles, the master vehicle starts from Y=6m, successfully changes lanes to the Y=2m lane, but returns to the Y=6m lane in the X=50-100m range to complete the final positioning. Faced with the complex combination of aggressive NV (Y=2m), LV1 (Y=6m), and cautious LV2 (Y=2m), the master vehicle adopts a "lane-change first, then return" strategy. This complex lane-changing behavior demonstrates the algorithm's dynamic decision-making capability in a mixed driving style environment: the master vehicle first successfully enters the Y=2m lane, and then, to avoid potential conflicts with aggressive vehicles and ensure long-term driving safety, the system intelligently chooses to return to the relatively safe Y=6m lane.

[0162] Case 3: For example Figure 7As shown, in a coordinated environment with a normal driving style, vehicles are distributed in three lanes: Y=-2m, Y=2m, and Y=6m. The master vehicle successfully changed lanes from Y=2m to Y=6m, with a smoother overall trajectory and a more coordinated lane-changing process, demonstrating the good interaction characteristics between vehicles under a normal driving style. In this normal driving style environment, all vehicles (LV remains at Y=2m, NV1 at Y=6m, and NV2 at Y=-2m) exhibit balanced driving characteristics, neither overly conservative nor overly aggressive, providing an ideal traffic environment for the master vehicle. The master vehicle selected the optimal lane-changing path through a game-theoretic decision-making algorithm, smoothly transitioning from the initial Y=2m lane to the target Y=6m lane, avoiding the influence areas of other vehicles, and achieving an efficient and safe lane-changing operation. This coordination reflects the balanced potential field characteristics between vehicles under a normal driving style, making lane-changing decisions more optimized and smoother.

[0163] Case 1: Speed ​​Characteristics

[0164] like Figure 8 As shown in (a), the main vehicle's HV: rapidly accelerates from an initial 20m / s to 25m / s, demonstrating an aggressive lane-changing overtaking strategy. The speed curve shows a clear acceleration trend in the first 4 seconds, and then tends to stabilize.

[0165] Cautious Leading Car LV: Maintains a steady and moderate speed increase, gradually accelerating from 20m / s to 22m / s, reflecting the conservative characteristics of a cautious driving style.

[0166] Aggressive Adjacent Vehicle (NV): Starting from a low initial speed of 15m / s, it continuously accelerates to 20m / s, demonstrating the aggressive vehicle's pursuit intention.

[0167] Case 2 Speed ​​Characteristics:

[0168] like Figure 8 As shown in (b), the main vehicle HV also exhibits strong acceleration, increasing from 20m / s to 25m / s, maintaining its dominance in complex multi-vehicle environments.

[0169] Aggressive Vehicle LV1: Exhibits an acceleration pattern similar to the HV, accelerating from 15m / s to 20m / s, demonstrating its competitive nature and aggressive driving style.

[0170] Aggressive Adjacent Vehicle NV: Maintains a relatively stable speed increase, from 17m / s to 21m / s

[0171] Cautious vehicle LV2: Exhibits the most conservative speed strategy, gradually increasing from 20m / s to 22m / s, providing a relaxed speed environment for the lead vehicle to change lanes.

[0172] Case 3: Speed ​​Characteristics

[0173] like Figure 8 As shown in (c), the main vehicle HV continues to maintain excellent acceleration performance, accelerating from 20m / s to 25m / s, achieving a smooth speed increase in normal driving conditions.

[0174] The standard NV1 vehicle: steadily accelerates from 15m / s to 20m / s, demonstrating balanced driving characteristics.

[0175] The standard NV2 vehicle exhibits a similar speed pattern to the NV1, increasing from 16 m / s to 20.3 m / s, reflecting consistency with typical driving styles.

[0176] Standard vehicle LV: Increased from 17m / s to 20.5m / s, reflecting the consistency of general driving style and the differentiation of driving style:

[0177] Cautious vehicles: Speed ​​increases are the most moderate, with relatively low acceleration.

[0178] Aggressive vehicles: exhibit stronger acceleration intent and competitiveness, but still within reasonable limits.

[0179] Standard vehicles exhibit balanced speed characteristics, neither too conservative nor too aggressive.

[0180] Game Theory Interaction Effect Verification: The speed curves of all three cases show obvious inter-vehicle interaction response characteristics. The lead vehicle can achieve safe lane changes through appropriate acceleration strategies under different driving style environments, while other vehicles adjust their speeds accordingly based on their respective driving style characteristics. This verifies that the non-cooperative game model can effectively capture and simulate vehicle interaction behaviors under different driving styles, achieving personalized speed coordination and safe lane-changing operations.

[0181] The embodiments and descriptions above are merely illustrative of the principles and preferred embodiments of the present invention. Various changes and modifications may be made to the present invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed.

Claims

1. A lane-changing method for autonomous vehicles considering vehicle-to-vehicle interaction, characterized in that, Includes the following steps: S1) Obtain the current state information of the current autonomous vehicle and other vehicles in the same scene as the current autonomous vehicle; S2) Based on the NGSIM dataset, driving behavior characteristics are analyzed, and the vehicle aggression coefficient is estimated using fuzzy inference methods to construct a social potential field model that considers driving style. S3) Based on collision risk assessment and vehicle interaction behavior analysis, a non-cooperative game model is constructed, and the non-cooperative game equilibrium is solved to obtain the optimal lane-changing decision and acceleration. S4) Use the longitudinal acceleration of the lane-changing decision variable output by the game layer as the reference input signal for the NMPC controller; S5) Based on the vehicle kinematics model and potential field information, the NMPC controller is used to generate the driving trajectory of the autonomous vehicle and output the corresponding motion planning information.

2. The lane-changing method for autonomous vehicles considering vehicle-to-vehicle interaction according to claim 1, characterized in that: In step S2), the expression for the social potential field model is: P=P NV (X,Y)+P r (Y); In the formula, P is the total potential energy field, P NV (X,Y) represents the potential energy field of adjacent vehicles, with the superscript NV indicating other vehicles adjacent to or near the main vehicle; P r (Y) represents the potential energy field at the road boundary, and (X,Y) represents the coordinate position of the main vehicle.

3. The lane-changing method for autonomous vehicles considering vehicle-to-vehicle interaction according to claim 2, characterized in that: In step S3), a non-cooperative game model is constructed based on collision risk assessment and vehicle interaction behavior analysis, specifically including the following steps: S31) The vehicle is taken as a game participant and the game strategy space is defined; the game strategy of the main vehicle includes three lateral decisions: changing lanes to the left, keeping lanes, and changing lanes to the right, and three longitudinal decisions: accelerating, decelerating, and maintaining speed. The game strategy of adjacent vehicles includes the corresponding acceleration adjustment range; S32) Construct the total cost function of the main vehicle C HV The total cost function C of adjacent vehicles NV The collision risk is calculated based on the time to collision (TTC) before the collision and then embedded into the safety cost function. S33) By analyzing vehicle interaction behavior, establish the mutual influence relationship between vehicles, including longitudinal following behavior and lateral lane change response behavior.

4. The lane-changing method for autonomous vehicles considering vehicle-to-vehicle interaction according to claim 3, characterized in that: In step S32), the total cost function C of the main vehicle in the non-cooperative game model is... HV for: In the formula, The cost function for the safety of the main vehicle, with the superscript HV indicating the main vehicle; The main vehicle comfort cost function; The main vehicle traffic efficiency cost function; The main vehicle safety weighting coefficient; The main vehicle comfort weighting coefficient; The main vehicle efficiency weighting coefficient.

5. The lane-changing method for autonomous vehicles considering vehicle-to-vehicle interaction according to claim 4, characterized in that: In step S32), the cost function of the main vehicle safety... Represented as: In the formula, σ = {-1, 0, 1} represents {left lane change, no lane change, right lane change}; This represents the longitudinal cost of the main vehicle; This indicates the lateral cost of the main vehicle; The longitudinal safety factor of the main vehicle; This is the longitudinal relative speed between the vehicle in front and the main vehicle in this lane; The longitudinal safety distance between the vehicle in front and the main vehicle in this lane; ε is the regularization parameter to avoid division by zero; TTC HV Time before the main vehicle collision; T safe For safe time thresholds; The weighting coefficient for the main vehicle; The longitudinal relative speed between the vehicle in front and the main vehicle; The longitudinal safety distance between the vehicle in front and the vehicle in front; TTC NV This represents the time before the collision between adjacent vehicles.

6. The lane-changing method for autonomous vehicles considering vehicle-to-vehicle interaction according to claim 5, characterized in that: In step S32), the total cost function C of adjacent vehicles in the non-cooperative game model is... NV Represented as: In the formula, A cost function that takes into account the safety of adjacent vehicles; The comfort cost function for adjacent vehicles; The traffic efficiency cost function for adjacent vehicles; The safety weighting coefficient for adjacent vehicles. The comfort weighting coefficient for adjacent vehicles. This is the efficiency weighting coefficient for adjacent vehicles.

7. The lane-changing method for autonomous vehicles considering vehicle-to-vehicle interaction according to claim 6, characterized in that: In step S32), the cost function that considers the safety of adjacent vehicles... Represented as: In the formula, σ = {-1, 0, 1} represents {left lane change, no lane change, right lane change}; Indicates the longitudinal cost of adjacent vehicles; Indicates the lateral cost of adjacent vehicles; The longitudinal safety factor for adjacent vehicles; The longitudinal relative speed between the vehicle in front and the adjacent vehicle; The longitudinal safety distance between the vehicle in front and the adjacent vehicle; TTC NV The time before the collision of adjacent vehicles; T safe ε is the safe time threshold; ε is the regularization parameter to avoid division by zero.

8. The lane-changing method for autonomous vehicles considering vehicle-to-vehicle interaction according to claim 7, characterized in that: In step S3), a non-cooperative dynamic game theory method is used, with the main vehicle as the leader and adjacent vehicles as followers, to solve for the non-cooperative game equilibrium and obtain the optimal lane-changing decision σ. * and optimal longitudinal acceleration As a reference input for the lower-level NMPC controller, it is specifically as follows: stσ∈{-1,0,1}; In the formula, σ * The optimal lane-changing strategy for the main vehicle serves as the overtaking decision. This represents the optimal longitudinal acceleration of the main vehicle; a NV The acceleration of the vehicles participating in the game is defined by the boundary of the acceleration that the vehicles can achieve; C HV With C NV γ represents the total decision cost of the game between the main vehicle and its adjacent vehicles; 2 (a HV σ) represents the set of optimal decision candidate actions made by other vehicles following the decision action of the master vehicle; σ∈{-1,0,1} represents the master vehicle changing lanes to the left, following in the original lane, and changing lanes to the right, respectively; a HV The acceleration of the main vehicle; Indicates γ 2 (a HV ,σ) is defined as the set described in the following curly braces; These are the minimum and maximum longitudinal acceleration constraints for the vehicle on road number μ, respectively; These are the minimum and maximum longitudinal speed limits for vehicles in lane μ, respectively.

9. A lane-changing method for autonomous vehicles considering vehicle-to-vehicle interaction according to claim 8, characterized in that: In step S5), a vehicle kinematic model is constructed with the vehicle's longitudinal position, lateral position, heading angle, and longitudinal velocity as state variables, and the front wheel steering angle and longitudinal acceleration as control variables. A composite potential field function of the vehicle repulsive potential field and the lane centerline constraint potential field is established. The nonlinear model predictive control (NMPC) architecture is adopted. Under the premise of satisfying the vehicle kinematic constraints, control input constraints, collision safety constraints and reference path lateral deviation change constraints, the optimal control sequence is solved by the rolling time domain optimization strategy to realize the trajectory planning and motion control of autonomous vehicles in dynamic traffic environment, and output complete motion planning information including position trajectory, speed, heading angle sequence and control commands. The vehicle kinematic model is as follows: In the formula, X represents the state vector of the vehicle at time τ+1; kin (τ+1) represents the longitudinal position coordinate of the vehicle at time τ+1; Y kin (τ+1) represents the lateral position coordinate of the vehicle at time τ+1; δ represents the heading angle of the vehicle at time τ+1; kin (τ+1) represents the front wheel steering angle of the vehicle at time τ+1; Δt represents the discretization time step; v xkin (τ) represents the longitudinal velocity of the vehicle at time τ; δ represents the heading angle of the vehicle at time τ; fkin (τ) represents the front wheel steering angle of the vehicle at time τ; L kin τ represents the vehicle wheelbase; τ represents the current discrete time.

10. A lane-changing method for autonomous vehicles considering vehicle-to-vehicle interaction according to claim 9, characterized in that: In step S5), the risk field is embedded into the output of the vehicle model, and the risk assessment result is used as the objective function term in the nonlinear model predictive control (NMPC) to drive the planning result away from the high-risk area. In the formula, y kin (τ) represents the potential energy field output value of the vehicle at time τ, which serves as the risk assessment term in the NMPC objective function; g represents the potential energy field mapping function, which maps the vehicle state and control input to risk potential energy values; u kin (τ) represents the vehicle's control input vector at time τ, including the front wheel steering angle and longitudinal acceleration; P represents the total potential energy field function, which consists of the potential energy fields of neighboring vehicles and the potential energy field of the road boundary. By minimizing the cost function J(τ), which comprehensively considers path risk assessment, course change smoothness, control input overhead, and road-keeping error, a collision-free and dynamically feasible reference path is generated after solving the optimization problem in each time period. st0≤v≤v max ,u min ≤u(τ+i|τ)≤u max ; In the formula, N p Indicates the prediction time domain; It represents the potential field strength and is used to guide vehicles to move along the potential field gradient and avoid high potential energy regions. The penalty term representing the change in yaw rate is used to suppress drastic changes in vehicle yaw motion and ensure driving stability; N c Indicates control over the time domain; This represents a penalty term for controlling changes in the input. The penalty for lateral position deviation is used to keep the vehicle within the target lane; v max Indicates the maximum permissible speed; v represents the current vehicle speed; u represents the maximum permissible speed. min u max These represent the minimum and maximum constraints of the control input, respectively; u(τ+i|τ) represents the predicted value of the control input at time τ to the future time τ+i. Represents the optimal control input sequence; u represents the optimal control input increment at time τ; kin (τ|τ) represents the actual control input at the current moment; This represents the optimal control input increment vector.

Images