A lane-changing trajectory planning method for autonomous driving vehicles based on multiplayer game

Abstract

The present invention belongs to the field of autonomous driving technology, and specifically relates to a lane-changing trajectory planning method for an autonomous driving vehicle based on multiplayer games, which solves the problem that existing decision-making methods cannot take into account dynamic trajectory planning and have poor applicability in complex vehicle-road environments. The technical key points are to determine the vehicle's lane-changing strategy in real time through a two-person or multiplayer game method, and complete trajectory planning. Specifically, a support vector regression prediction model is constructed to predict the state of the leading vehicle and determine the feasibility of lane changing; during the lane changing process, a two-person game model between the vehicle and the vehicle behind the target lane is constructed, and after the vehicle fails to change lanes and decides to return to the original lane, a multiplayer game model involving the original lane and the vehicle behind the target lane is constructed; according to the real-time lane changing decision and the predicted state of the leading vehicle, a quintic polynomial is used to dynamically fit the lane-changing trajectory to ensure the safety and stability of the lane changing process. The present invention realizes safe lane changing of autonomous driving vehicles in complex vehicle-road environments.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and in particular relates to a lane-changing trajectory planning method for an autonomous driving vehicle based on multiplayer game. Background Art

[0002] Autonomous vehicles use key technologies such as environmental perception, decision-making, planning and vehicle control, and have significant advantages over manually driven vehicles in terms of traffic safety and traffic efficiency. With the development of autonomous driving technology, the share of autonomous driving vehicles on the road will gradually increase, and traffic environments that contain both autonomous driving vehicles and manually driven vehicles will become a common traffic scenario.

[0003] Lane changing is a common driving behavior on the road, but the risk of vehicle conflict increases during lane changing, which can easily lead to traffic accidents. Autonomous driving technology is expected to improve the safety of vehicles during lane changing. The lane changing algorithm of an autonomous vehicle usually has four levels: (1) the strategic planning layer, which is responsible for planning the vehicle's driving path during the trip and affects the vehicle's lane changing choice; (2) the tactical decision layer, which is responsible for making decisions on the vehicle's behavior, including the choice of following and lane changing, as well as the choice of acceleration and deceleration; (3) the trajectory planning layer, which generates a safe and reasonable lane changing trajectory based on the decision behavior, and needs to be optimized in real time according to the surrounding traffic environment; (4) the operation control layer, which commands the vehicle to control speed and direction along the trajectory based on the optimal trajectory generated by the trajectory planning layer. Lane changing decisions and trajectory planning belong to the tactical decision layer and trajectory planning layer, respectively.

[0004] There are many types of lane-changing decision models, including rule-based models, utility-calculation-based models, machine-learning-based models, and game-theory-based models. The rule-based model presupposes a scenario in which lanes need to be changed so that the vehicle status meets certain conditions and a lane change is required. However, fixed rules are usually difficult to adapt to complex and changing road environments; the utility-calculation-based model calculates the utility of behaviors such as lane changing and following, and takes the behavior with the highest utility value. This type of model is also difficult to give a general utility calculation method; the machine-learning-based model uses a large amount of data to train a machine-learning model that can determine whether to change lanes based on the current traffic status. This method is difficult to ensure the safety of the decision results; the game-theory-based method is to include the vehicle status in the current lane and the target lane in the decision-making process, and after considering multiple possible strategies, give the most appropriate behavior decision for the vehicle intending to change lanes, which can meet the lane-changing decision needs under complex traffic conditions. The lane-changing game model usually includes only two decision-makers: the vehicle intending to change lanes and the vehicle behind in the target lane. However, in a complex traffic environment with high traffic density, lane-changing behavior will be affected by more surrounding vehicles. Therefore, a multi-person game model that includes more decision-makers is more in line with the actual lane-changing process.

[0005] Lane-changing trajectory planning usually uses geometric curves as the basis for generating trajectories, among which polynomial curve-based trajectories are the most common, in addition to sine (cosine) curves, B-spline curves, trapezoidal curves and spirals. Since the traffic environment changes rapidly during the lane-changing process, the lane-changing trajectory needs to be frequently modified and optimized to ensure safety and efficiency. This technology is called dynamic trajectory planning.

[0006] The existing autonomous driving vehicle lane changing technology has the following difficulties that need to be solved:

[0007] 1. When an autonomous vehicle changes lanes, it is necessary to consider both lane change decision and lane change trajectory planning. If the trajectory planning is considered only after the decision is made, the vehicle may not be able to plan a smooth trajectory that conforms to the actual situation;

[0008] 2. Autonomous vehicles need to consider the motion state of the vehicle in front when making lane-changing decisions. If the vehicle in front is only considered to be moving at a constant speed, it may cause safety hazards during the lane-changing process in a complex traffic environment with high traffic density.

[0009] 3. When changing lanes, the autonomous vehicle needs to consider the possibility that the vehicle behind may overtake it due to lateral deviation. If only the vehicle in the target lane is considered, the autonomous vehicle will have a potential conflict with the vehicle behind it when it returns to the original lane due to the aggressiveness of the vehicle in the target lane.

[0010] Therefore, it is necessary to propose a method system that combines multi-vehicle game decision-making algorithm and dynamic trajectory planning to cope with lane-changing scenarios under mixed traffic flows of autonomous driving and manual driving, and to improve the lane-changing behavior quality of autonomous driving vehicles. Summary of the invention

[0011] The purpose of the present invention is to provide a lane-changing trajectory planning method and simulation test method for an autonomous driving vehicle based on multiplayer game for lane-changing scenarios under mixed traffic flows of autonomous driving and manual driving. The present invention has the following characteristics: 1. Taking into account the lane-changing decision and trajectory planning of autonomous driving vehicles; 2. Using the support vector regression model to predict the motion state of the front vehicle in the current lane and the front vehicle in the target lane within the required lane-changing time, reducing the safety hazards caused by the prior art treating the front vehicle as uniform motion; 3. Introducing the multiplayer game theory into the lane-changing trajectory planning method for autonomous driving vehicles, the autonomous driving vehicle can judge whether to continue to change lanes according to the motion state of itself and surrounding vehicles, and the multiplayer game model composed of more decision-making subjects can effectively consider the reaction of the target lane and the rear vehicle in the original lane during the lane-changing process, thereby improving the safety of lane-changing; 4. The lane-changing trajectory planning method for autonomous driving vehicles adopts a fifth-order polynomial to fit the vehicle lane-changing trajectory line, considers the vehicle collision constraint and the vehicle kinematic constraint in the optimization process, and plans the trajectory line that meets the vehicle kinematic constraint under different circumstances, which is more in line with the actual lane-changing process.

[0012] The technical solution of the present invention is:

[0013] A lane-changing trajectory planning method for an autonomous driving vehicle based on multiplayer game, characterized in that it comprises the following steps:

[0014] S1. Obtain vehicle information of the autonomous driving vehicle and its lane and vehicle information of the target lane;

[0015] S2, the autonomous driving vehicle generates a lane-changing intention, and predicts the motion states of the preceding vehicle in the current lane and the preceding vehicle in the target lane within the required lane-changing time based on its own information and the acquired vehicle information, based on the support vector regression model, and determines whether the lane-changing conditions are met based on the motion states. If so, it proceeds to S3-1, otherwise it returns to S1;

[0016] S3. Establish a game model to obtain the optimal lane-changing decision at the current moment:

[0017] S3-1. Establish a two-player game model between the autonomous driving vehicle (vehicle A) and the vehicle behind it in the target lane (vehicle B) to obtain the optimal lane-changing decision at the current moment:

[0018] Generate the utility function U of car A and car B payoff :

[0019]

[0020]

[0021] In the formula, are the game profit functions of vehicles A and B respectively. The subscript 2 indicates that the number of players is 2; a 0 is the initial state vehicle acceleration, a A is the lateral acceleration of the vehicle, i.e., the lane-changing decision variable; q A and q B are the aggressiveness coefficients of vehicles A and B respectively (the larger the aggressiveness coefficient, the more the driver tends to improve efficiency rather than safety when making decisions), β(q) is the cumulative distribution function of aggressiveness q, and 0≤β(q)≤1, U safety is the safety yield, U space is the space (efficiency) benefit, δ is the anti-collision parameter of the front vehicle, and f w is the penalty function;

[0022] According to the utility functions of car A and car B, we can solve the equilibrium solution of the following game:

[0023]

[0024] In the formula, x A is the decision of the autonomous driving vehicle A in the game process, x A*is the optimal lane-changing decision of autonomous vehicle A; x B is the decision of vehicle B in the target lane during the game process, x B* It is the optimal lane-changing decision for vehicle B in the target lane.

[0025] If the optimal game decision obtained by vehicle A is to execute lane change, then enter S4-1; if the optimal game decision obtained by vehicle A is to return to the original lane, then enter S4-2.

[0026] S3-2. When vehicle A chooses to return to the original lane, it may potentially conflict with vehicle C. A multiplayer game model is established between the autonomous driving vehicle (vehicle A) and the vehicle behind the target lane (vehicle B) and the vehicle behind the original lane (vehicle C) to obtain the optimal lane change decision at the current moment:

[0027] Generate the utility function U of car A, car B and car C payoff :

[0028]

[0029]

[0030]

[0031] In the formula, are the game profit function values ​​of vehicles A, B, and C respectively. The subscript 3 indicates that the number of players is 3; a B and a C Respectively represent the acceleration of vehicles B and C; q C is the aggressiveness coefficient of vehicle C; O n is the overtaking expectation parameter of the rear vehicle in the original lane (the higher the overtaking expectation parameter, the more the rear vehicle tends to accelerate to overtake the front vehicle rather than slow down to give way to the front vehicle);

[0032] According to the utility functions of car A, car B and car C, we can solve the equilibrium solution of the following game:

[0033]

[0034] In the formula, x C is the decision of car C in the game process, x C* is the optimal lane-changing decision for car C.

[0035] If the optimal game decision obtained by vehicle A is to re-execute the lane change, then enter S4-3; if the optimal game decision obtained by vehicle A is to return to the original lane, then enter S4-2.

[0036] S4. Generate a lane-changing trajectory according to the obtained optimal lane-changing strategy, specifically:

[0037] S4-1. If the optimal lane-changing decision is to continue changing lanes to the target lane, an optimization model is established and a fifth-order polynomial is used to fit the vehicle's lane-changing trajectory. The vehicle motion state at the start and end of the lane-changing trajectory and the lane-changing travel time t if Solve to obtain the position, velocity and acceleration of the vehicle while traveling on the quintic polynomial trajectory:

[0038] The position of the vehicle at any time (time t) in the trajectory is:

[0039]

[0040]

[0041] In the formula, x it (t) is the horizontal position of the vehicle at time t, and are the horizontal velocity and acceleration of the vehicle at the time of trajectory planning (i), y it (t) is the lateral position of the vehicle at time t, y if is the lateral distance between the vehicle position at the time of trajectory planning (time i) and the center line of the target lane, t if is the lane change time, and are the lateral velocity and acceleration of the vehicle at the time of trajectory planning (time i), respectively.

[0042] By using different lane change travel times t if Substituting into the fifth-order polynomial trajectory planning equation, different trajectories can be planned. The purpose of the optimization model is to select a lane-changing trajectory that takes into account both efficiency and safety from multiple feasible trajectories and obtain the optimal lane-changing driving time t if The objective function of the optimization model is:

[0043]

[0044] The constraints are:

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051] t min≤t if ≤t max

[0052] In the formula, α 1 ,α 2 ,α 3 is the target parameter, which is used to describe the tendency towards efficiency and safety. and are the maximum horizontal and lateral velocities in the planned trajectory, and are the maximum horizontal and lateral accelerations in the planned trajectory, and are the maximum horizontal and lateral accelerations in the planned trajectory, is the maximum horizontal and lateral speed allowed, is the maximum permissible horizontal and lateral acceleration, x if , are the horizontal positions of the lane-changing vehicle (car A), the vehicle behind the target lane (car B), and the vehicle in front of the target lane (car D) at the end of the lane-changing trajectory of car A. are the horizontal speeds of the lane-changing vehicle (car A), the vehicle behind the target lane (car B), and the vehicle in front of the target lane (car D) at the end of the lane-changing trajectory of car A, t is the reaction time of the lagging vehicle, and b A , b B and b D are the maximum accelerations of cars A, B, and D during deceleration, t min With t max are the shortest and longest lane change times.

[0053] After the optimization is completed, the optimal lane-changing trajectory that takes into account both efficiency and safety is obtained, and the system enters S5.

[0054] S4-2. If the optimal lane-changing decision is to terminate the lane-changing and return to the original lane, then the trajectory of the vehicle returning to the original lane is generated according to the above-mentioned optimization method and the quintic polynomial trajectory planning method. The target lane during trajectory planning becomes the original lane, and the front and rear vehicles in the target lane become the front vehicle (car E) and rear vehicle (car C) in the original lane, and then enter S5.

[0055] S4-3. If the optimal lane-changing decision is to terminate and return to the original lane and change lanes again, then generate a vehicle lane-changing trajectory according to the above optimization method and the quintic polynomial trajectory planning method, and enter S5.

[0056] S5. Drive for a time interval according to the obtained trajectory and update the vehicle information of the autonomous driving vehicle and its lane. If the vehicle has reached the center of the target lane or has returned to the center of the original lane, end the current lane change trajectory planning. Otherwise, re-predict the motion status of the front vehicle in the current lane and the front vehicle in the target lane within the required lane change time. If the vehicle is in the process of changing lanes or re-changing lanes at this time, return to S3-1; if the vehicle is in the process of returning to the original lane at this time, return to S3-2.

[0057] The beneficial effects of the present invention are as follows: 1) taking into account both the lane changing decision and trajectory planning of the autonomous driving vehicle, it can avoid the defect that some existing technologies only consider the lane changing decision and are unable to plan a smooth trajectory line that conforms to the actual situation; 2) using the support vector regression model to predict the motion state of the front vehicle in the current lane and the front vehicle in the target lane within the required lane changing time, it can reduce the safety hazard caused by the prior art treating the front vehicle as uniform motion; 3) introducing the multiplayer game theory into the lane changing trajectory planning method of the autonomous driving vehicle, the autonomous driving vehicle can judge whether to continue changing lanes according to the motion state of itself and the surrounding vehicles, and can effectively consider the reaction of the target lane and the rear vehicle in the original lane during the lane changing process, thereby improving the lane changing safety; 4) the lane changing trajectory planning method of the autonomous driving vehicle adopts a quintic polynomial to fit the vehicle lane changing trajectory line, considers the vehicle collision constraints and the vehicle kinematic constraints in the optimization process, and plans a trajectory line that conforms to the vehicle kinematic constraints under different situations, which is more in line with the actual lane changing process. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a schematic diagram of the lane changing scenario of an autonomous driving vehicle.

[0059] Figure 2 It is a schematic diagram of the vehicles participating in the lane-changing game and their driving intentions.

[0060] Figure 3 It is a schematic diagram of the vehicle lane changing strategy generated by the lane changing game.

[0061] Figure 4 The present invention relates to a method for planning lane-changing trajectories of an autonomous driving vehicle based on multi-player game and a flow chart of a simulation test method. DETAILED DESCRIPTION

[0062] The technical solution of the present invention is described below in conjunction with the accompanying drawings:

[0063] Step 1: The vehicle obtains the vehicle kinematic information such as the speed and acceleration of the vehicle and surrounding vehicles in real time through the vehicle hardware equipment. The position of the vehicle and surrounding vehicles on the road is as follows: Figure 1 shown.

[0064] Step 2: The autonomous driving vehicle generates a lane-changing intention and predicts the motion states of the leading vehicle in the own lane and the leading vehicle in the target lane within the required lane-changing time based on the support vector regression model according to its own information and the acquired vehicle information.

[0065] When constructing a prediction model for the motion state of the leading vehicle, the motion state of the leading vehicles in the two lanes within a few seconds before the prediction time can be used as the model input. Assuming that the leading vehicles in the two lanes do not make significant lateral displacements, only predicting the longitudinal motion state of the leading vehicle can satisfy the construction of the lane change simulation scenario. The specific prediction method is:

[0066] Based on the support vector regression model, the longitudinal speed of the autonomous vehicle in the previous few seconds is obtained. Predict the speed of the vehicle ahead at the next moment. During the lane changing process, predict the speed of the vehicle ahead in real time until the lane changing process is completed.

[0067] Among them, the support vector regression model is a machine learning method derived from the principle of support vector machine, which has the advantages of high accuracy and fast convergence in solving small sample regression problems. The specific principles of constructing the support vector regression model are:

[0068] Construct training data set D = {(x 1 ,y 1 ),(x 2 ,y 2 ),…,(x m ,y m )},y i ∈R.

[0069] Among them, x i is the vector of the speed of the preceding vehicle several seconds before time i, y i is the speed of the preceding vehicle at time i, and m is the amount of data in the training set.

[0070] The training goal of the support vector regression model is to solve a regression model f(x) = ω T x+b, which is a hyperplane analytic expression in geometric sense, ω=(ω 1 ;ω 2 ;…;ω d ) is the normal vector of the hyperplane, and b is the displacement term of the hyperplane. The solution goal is to make the deviation between the predicted value f(x) and the actual speed as small as possible, while separating the data in the training set as much as possible. The deviation between the predicted value and the actual value is called the loss function, which can be constructed as follows:

[0071]

[0072] Among them, l εis the loss function, z is the model prediction deviation (the difference between the predicted value and the true value), and ε is the sensitive threshold to the deviation.

[0073] The solution of the support vector regression model can be transformed into the following nonlinear optimization problem:

[0074]

[0075]

[0076] Where C is the regularization parameter, which is used to adjust the relative weights of the maximum margin objective and the minimum deviation objective; i and is a slack variable.

[0077] In order to facilitate the solution, it is necessary to transform the above nonlinear optimization problem into an easy-to-solve convex optimization problem. To this end, the Lagrangian function is first constructed:

[0078]

[0079]

[0080] Where α i , μ i , are all Lagrange multipliers.

[0081] The extreme point of the above Lagrangian function is the optimal solution to the original problem, and the extreme value of the Lagrangian function is only obtained when the gradient of each variable is zero. According to the extreme value solution of the Lagrangian function, the dual problem of the original optimization problem can be obtained:

[0082]

[0083]

[0084] Solve the optimal solution by dual problem

[0085] Based on the above solution process, the optimal parameters in the regression model can be expressed as:

[0086]

[0087]

[0088] At this point, the support vector regression model is solved. The longitudinal speed within a few seconds before the required prediction time is input into the support vector regression model, and the speed of the preceding vehicle at the prediction time can be obtained. Determine whether the lane change condition is met according to the motion state. If so, proceed to step 3-1, otherwise return to step 1.

[0089] Step 3: Based on the vehicle kinematic information obtained in step 1 and the motion state of the preceding vehicle predicted in step 2, a vehicle game model is established to obtain the optimal lane-changing decision at the current moment.

[0090] Step 3-1: Establish a two-player game model between the autonomous driving vehicle (vehicle A) and the vehicle behind it in the target lane (vehicle B) to obtain the optimal lane-changing decision at the current moment:

[0091] The basic principle of the game theory model is to calculate the payoff function of each party in the game, that is, the payoff of the two vehicles needs to be accurately calculated. The payoff function of the lane-changing vehicle (car A) and the vehicle behind in the target lane (car B) can be calculated as:

[0092]

[0093]

[0094] In the formula, a 0 is the initial state; a A is the decision of vehicle A; q is the aggressiveness coefficient, which is expressed as the consideration weight of the vehicle for the same traffic conditions. The larger the value, the more the vehicle tends to improve efficiency when making decisions; For security benefits; is the space benefit; T cl is the time required to complete the lane change; SP is the safety parameter, which is determined by the headway; RP is the space parameter, which is determined by the headway; is a penalty function, where v desired is the expected driving speed, w 1 、w 2 is the penalty parameter, v and a are the current speed and acceleration of the vehicle; β(q) is the cumulative distribution function of the aggressiveness q, 0≤β(q)≤1, and δ is the anti-collision parameter of the front vehicle, which avoids collision with the front vehicle by reducing the benefit in the case of collision.

[0095] According to the utility function, the optimal decision of the autonomous driving vehicle and the vehicle behind in the target lane is obtained:

[0096]

[0097] If the optimal game decision obtained by vehicle A is to change lanes, then go to step 4-1; if the optimal game decision obtained by vehicle A is to return to the original lane, then go to step 4-2.

[0098] Step 3-2: When vehicle A chooses to return to the original lane, it may potentially conflict with vehicle C. A multiplayer game model is established between the autonomous driving vehicle (vehicle A) and the vehicle behind the target lane (vehicle B) and the vehicle behind the original lane (vehicle C) to obtain the optimal lane change decision at the current moment:

[0099] The profit functions of the lane-changing vehicle (vehicle A), the vehicle behind the target lane (vehicle B), and the vehicle behind the original lane (vehicle C) can be calculated as follows:

[0100]

[0101]

[0102]

[0103] In the formula, are the game profit function values ​​of vehicles A, B, and C respectively. The subscript 3 indicates that the number of players is 3; a B and a C Respectively represent the acceleration of vehicles B and C; q C is the aggressiveness coefficient of vehicle C; O n is the expected overtaking parameter of the rear vehicle in the original lane (the higher the expected overtaking parameter, the more inclined the rear vehicle is to accelerate to overtake the front vehicle rather than slow down to give way to the front vehicle), which can be calculated by the following method:

[0104] When vehicle A chooses to return to the original lane (such as Figure 2 "Purpose A" in the figure), the lateral displacement of vehicle A (such as Figure 2 As shown in the figure, the rear vehicle C in the original lane has a desire to overtake (as shown in the figure Figure 2 In the multiplayer game, the game with the vehicle behind the original lane needs to consider the overtaking expectation of the vehicle behind. That is, when the vehicle changes lanes, the lane-changing vehicle deviates from the original lane, forming a certain width of lateral clearance, which will trigger the overtaking expectation of the driver of the following vehicle. The overtaking expectation of the following vehicle can be calculated according to the following formula:

[0105] G(△v n,n+1 (t),△v n,n+2 (t))=(1-P n )△v n,n+1 (t)+P n △v n,n+2 (t)

[0106] V[△X n,n+1 (t),△X n,n+1 v (t),△X n,n+2 (t)]=

[0107] V[(1-P n )((1-O n )△X n,n+1 (t)+O n △X n,n+1v (t)+P n ((1-O n )△X n,n+1 (t)+O n △X n,n+2 (t))]

[0108]

[0109] Where n is the serial number of the vehicle in the convoy, which refers to the following vehicle C in the original lane; G(.) is the stimulus-response function in the following model; △X n,n+1 (t) = X n+1 (t)-X n (t) is the headway between vehicle n and vehicle n+1 (vehicle A) at time t, is the headway between vehicle n and the virtual preceding vehicle n+1 at time t, △X n,n+2 (t) = X n+2 (t)-X n (t) is the headway between vehicle n and vehicle n+2 (the preceding vehicle of the autonomous driving vehicle, i.e., vehicle E) at time t; V is the speed optimization equation; is the differential of the speed of vehicle n at time t; P n is the ratio of the lateral displacement of the autonomous driving vehicle to the minimum passable clearance; △v n,n+1 (t) = v n+1 (t)-v n (t), △v n,n+2 (t) = v n+2 (t)-v n (t) are the speed differences between vehicle n+1, vehicle n+2 and vehicle n;

[0110] According to the above formula, the overtaking expectation of the following vehicle is:

[0111]

[0112] Among them, O n is the overtaking expectation parameter; α is the driver's sensitivity coefficient to speed difference, and κ is the sensitivity coefficient to respond to the stimulus G(.).

[0113] In a multiplayer game, participating vehicles will select the strategy that maximizes their benefits as the implementation strategy. The lane-changing decision of the lane-changing vehicle at this moment can be obtained according to the following formula:

[0114]

[0115] In the formula, x C is the decision of car C in the game process, x C* is the optimal lane-changing decision for car C.

[0116] If the optimal game decision obtained by vehicle A is to re-execute the lane change, then go to step 4-3; if the optimal game decision obtained by vehicle A is to return to the original lane, then go to step 4-2.

[0117] Step 4: Generate a lane-changing trajectory based on the optimal lane-changing strategy. There are three types of trajectory lines that may be generated, such as Figure 3 As shown, they correspond to steps 4-1 to 4-3, specifically:

[0118] Step 4-1: If the optimal lane-changing decision is to continue changing lanes to the target lane, an optimization model is established and a fifth-order polynomial is used to fit the vehicle's lane-changing trajectory. The vehicle motion state at the start and end of the lane-changing trajectory and the lane-changing travel time t if Solve to obtain the position, velocity and acceleration of the vehicle while traveling on the quintic polynomial trajectory:

[0119] When fitting the lane-changing trajectory of a vehicle, a temporary coordinate system is established with the current position of the lane-changing vehicle as the coordinate origin, and the horizontal and vertical coordinates, horizontal and vertical velocities, and horizontal and vertical accelerations of the lane-changing vehicle are re-determined based on this.

[0120] Furthermore, in step 5, a fifth-order polynomial is used to fit the lane-changing trajectory of the vehicle as follows:

[0121]

[0122] y it (t) = k i0 +k i1 t+k i2 t 2 +k i3 t 3 +k i4 t 4 +k i5 t 5

[0123] In the formula, x it (t) is the horizontal position of the vehicle at any time (time t) in the trajectory planned at time i, y it (t) is the lateral position of the vehicle at any time (time t) in the trajectory planned at time i, and are the horizontal velocity and acceleration of the vehicle at the time of trajectory planning (i), k i0 ,k i1 ,k i2 ,k i3 ,k i4 ,k i5is the coefficient of the quintic polynomial. The position, velocity and acceleration of the vehicle during the updating state can be obtained according to the vehicle motion state at the start and end of the lane change trajectory:

[0124] The position of the vehicle at any time (time t) in the trajectory is:

[0125]

[0126]

[0127] In the formula, y if is the lateral distance between the vehicle position at the time of trajectory planning (i) and the center line of the target lane, t if is the lane change time, and are the lateral velocity and acceleration of the vehicle at the time of trajectory planning (time i), respectively.

[0128] By using different lane change travel times t if Substituting into the fifth-order polynomial trajectory planning equation, different trajectories can be planned. The purpose of the optimization model is to select a lane-changing trajectory that takes into account both efficiency and safety from multiple feasible trajectories and obtain the optimal lane-changing driving time t if The objective function of the optimization model is:

[0129]

[0130] The constraints are:

[0131]

[0132]

[0133]

[0134]

[0135]

[0136]

[0137] t min ≤t if ≤t max

[0138] In the formula, α 1 ,α 2 ,α 3 is the target parameter, which is used to describe the tendency towards efficiency and safety. and are the maximum horizontal and lateral velocities in the planned trajectory, and are the maximum horizontal and lateral accelerations in the planned trajectory, and are the maximum horizontal and lateral accelerations in the planned trajectory, is the maximum horizontal and lateral speed allowed, is the maximum permissible horizontal and lateral acceleration, x if , are the horizontal positions of the lane-changing vehicle (car A), the vehicle behind the target lane (car B), and the vehicle in front of the target lane (car D) at the end of the lane-changing trajectory of car A. are the horizontal speeds of the lane-changing vehicle (car A), the vehicle behind the target lane (car B), and the vehicle in front of the target lane (car D) at the end of the lane-changing trajectory of car A, t is the reaction time of the lagging vehicle, and b A , b B and b D are the maximum accelerations of cars A, B, and D during deceleration, respectively. D is the length of vehicle D. min With t max are the shortest and longest lane change times.

[0139] The objective function of the above optimization model can take into account both lane changing efficiency and lane changing safety. if The smaller it is, the higher the lane changing efficiency is; the maximum lateral speed in the second and third planning trajectories and the maximum lateral acceleration The smaller it is, the higher the lane change safety is. The objective function uses parameter α 1 ,a 2 ,α 3 To describe the preference for efficiency and safety.

[0140] The constraints of the above optimization model are obtained according to the following principles:

[0141] (1) The autonomous vehicle should avoid potential collisions with the vehicle in front and behind in the target lane during the lane change process. Each time the trajectory is updated, it is necessary to ensure that the lane-changing vehicle maintains a safe distance S from the vehicle in front (vehicle D) when it reaches the final position in the target lane. p The safety distance S from the vehicle behind (car B) l , that is, the horizontal position of vehicle A in the target lane should be within the range.

[0142] The following safety distance can be calculated based on the Gipps following model. The principle of the safety distance of the Gipps following model is that the rear vehicle should maintain a safe distance from the front vehicle, so that the front vehicle and the rear vehicle can stop without a rear-end collision in the event of an emergency brake of the front vehicle. The reaction time of the lagging vehicle is taken into account in the safety distance rule of the Gipps model. Based on the Gipps following model, the horizontal position constraint of the vehicle at the end of the lane change in the target lane is:

[0143]

[0144]

[0145] (2) During the lane-changing process, the horizontal and lateral speeds and accelerations of the vehicle should not exceed the permitted range, and the lane-changing time should not be too long or too short, that is:

[0146]

[0147]

[0148]

[0149]

[0150] t min ≤t if ≤t max

[0151] After the optimization is completed, the optimal lane-changing trajectory that takes into account both efficiency and safety is obtained, and the process proceeds to step 5.

[0152] Step 4-2: If the optimal lane-changing decision is to terminate the lane-changing and return to the original lane, then generate a trajectory for the vehicle to return to the original lane according to the above optimization method and the quintic polynomial trajectory planning method. The target lane during trajectory planning becomes the original lane, and the front and rear vehicles in the target lane become the front vehicle (car E) and rear vehicle (car C) in the original lane, and proceed to step 5.

[0153] Step 4-3: If the optimal lane-changing decision is to terminate the vehicle and return to the original lane and change lanes again, generate the vehicle's lane-changing trajectory according to the above optimization method and the quintic polynomial trajectory planning method, and proceed to step 5.

[0154] Step 5: Drive for a time interval according to the obtained trajectory. The time interval can be flexibly adjusted according to the vehicle controller and sensor settings, but should be set to 0.1 seconds or less. Update the vehicle information of the autonomous driving vehicle and its lane. If the vehicle has reached the center of the target lane or has returned to the center of the original lane, end the current lane change trajectory planning. Otherwise, re-predict the motion state of the front vehicle in the current lane and the front vehicle in the target lane within the required lane change time. If the vehicle is in the process of changing lanes or re-changing lanes at this time, return to step 3-1; if the vehicle is in the process of returning to the original lane at this time, return to step 3-2.

Claims

1. A lane-changing trajectory planning method for autonomous driving vehicles based on multiplayer games. It is characterized in that The following steps are involved: S1. Obtain vehicle information of the autonomous driving vehicle and its lane and vehicle information of the target lane; S2, the autonomous driving vehicle generates a lane-changing intention, and predicts the motion states of the preceding vehicle in the current lane and the preceding vehicle in the target lane within the required lane-changing time based on its own information and the acquired vehicle information, based on the support vector regression model, and determines whether the lane-changing conditions are met based on the motion states. If so, it proceeds to S3-1, otherwise it returns to S1; S3. Establish a game model to obtain the optimal lane-changing decision at the current moment: S3-1. Establish a two-player game model between the autonomous driving vehicle and the vehicle behind the target lane to obtain the optimal lane-changing decision at the current moment. Define the autonomous driving vehicle as vehicle A and the vehicle behind the target lane as vehicle B: Generate the utility function U of car A and car B payoff : In the formula, are the game profit functions of vehicles A and B respectively. The subscript 2 indicates that the number of players is 2; a 0 is the initial state vehicle acceleration, a A is the lateral acceleration of the vehicle, i.e., the lane-changing decision variable; q A and q B They are the aggressiveness coefficients of vehicles A and B, respectively. The definition of aggressiveness coefficient is: the larger the aggressiveness coefficient, the more the driver tends to improve efficiency rather than safety when making decisions; β(q) is the cumulative distribution function of aggressiveness q, and 0≤β(q)≤1, U safety is the safety yield, U space is the space benefit, δ is the anti-collision parameter of the front vehicle, and f w is the penalty function; According to the utility functions of car A and car B, we can solve the equilibrium solution of the following game: In the formula, x A is the decision of the autonomous driving vehicle A in the game process, x A* is the optimal lane-changing decision of autonomous vehicle A; x B is the decision of vehicle B in the target lane during the game process, x B* is the optimal lane-changing decision for vehicle B in the target lane; If the optimal game decision obtained by vehicle A is to change lanes, then enter S4-1; if the optimal game decision obtained by vehicle A is to return to the original lane, then enter S4-2; S3-2. When vehicle A chooses to return to the original lane, it may potentially conflict with vehicle C. A multiplayer game model is established between the autonomous driving vehicle and the vehicle behind the target lane and the vehicle behind the original lane to obtain the optimal lane-changing decision at the current moment. The vehicle behind the original lane is defined as vehicle C: Generate the utility function U of car A, car B and car C payoff : In the formula, are the game profit function values ​​of vehicles A, B, and C respectively. The subscript 3 indicates that the number of players is 3; a B and a C Represent the acceleration of vehicles B and C respectively; q C is the aggressiveness coefficient of vehicle C; n is the overtaking expectation parameter of the rear vehicle in the original lane. The definition of the overtaking expectation parameter is: the higher the overtaking expectation parameter, the more the rear vehicle tends to accelerate to overtake the front vehicle rather than slow down to give way to the front vehicle; According to the utility functions of car A, car B and car C, we can solve the equilibrium solution of the following game: In the formula, x C is the decision of car C in the game process, x C* is the optimal lane-changing decision of car C; If the optimal game decision obtained by vehicle A is to re-execute the lane change, then enter S4-3; if the optimal game decision obtained by vehicle A is to return to the original lane, then enter S4-2; S4. Generate a lane-changing trajectory according to the obtained optimal lane-changing strategy, specifically: S4-1. Establish an optimization model and use a fifth-order polynomial to fit the vehicle lane-changing trajectory. According to the vehicle motion state at the starting and ending points of the lane-changing trajectory and the lane-changing driving time t if Solve to obtain the position, velocity and acceleration of the vehicle while traveling on the quintic polynomial trajectory: The position of the vehicle at any time t in the trajectory is: In the formula, x it (t) is the horizontal position of the vehicle at time t, and are the horizontal velocity and acceleration of the vehicle at the time of trajectory planning, y it (t) is the lateral position of the vehicle at time t, y if is the lateral distance between the vehicle position at the time of trajectory planning and the center line of the target lane, t if is the lane change time, and are the lateral velocity and acceleration of the vehicle at the time of trajectory planning; By using different lane change travel times t if Substitute it into the quintic polynomial trajectory planning equation to plan different trajectories. The purpose of the optimization model is to select a lane-changing trajectory that takes into account both efficiency and safety from multiple feasible trajectories and obtain the optimal lane-changing driving time t if ; The objective function of the optimization model is: The constraints are: In the formula, α 1 ,α 2 ,α 3 is the target parameter, which is used to describe the tendency towards efficiency and safety. are the maximum horizontal and lateral velocities in the planned trajectory, and are the maximum horizontal and lateral accelerations in the planned trajectory, is the maximum horizontal and lateral speed allowed, is the maximum permissible horizontal and lateral acceleration, are the horizontal positions of cars A, B, and D at the end of the lane change trajectory of car A, where car D is the front car in the target lane. are the horizontal speeds of car A, car B and car D at the end of the lane-changing trajectory of car A, τ is the reaction time of the lagging vehicle, b A , b B and b D are the maximum accelerations of cars A, B, and D during deceleration, respectively. D is the length of vehicle D, t min With t max is the shortest and longest lane change time; After the optimization is completed, the optimal lane-changing trajectory that takes into account both efficiency and safety is obtained, and the process enters S5; S4-2, if the optimal lane-changing decision is to terminate the lane-changing and return to the original lane, the optimization model of S4-1 and the fifth-order polynomial trajectory planning method are used to generate a trajectory for the vehicle to return to the original lane. The target lane during trajectory planning becomes the original lane, and cars D and B become cars E and C, where car E is the front car in the original lane, and the process goes to S5; S4-3, if the optimal lane-changing decision is to terminate and return to the original lane and change lanes again, the S4-1 optimization model and the quintic polynomial trajectory planning method are used to generate the vehicle's lane-changing trajectory, and then enter S5; S5. Drive for a time interval according to the obtained trajectory and update the vehicle information of the autonomous driving vehicle and its lane. If the vehicle has reached the center of the target lane or has returned to the center of the original lane, end the current lane change trajectory planning. Otherwise, re-predict the motion status of the front vehicle in the current lane and the front vehicle in the target lane within the required lane change time. If the vehicle is in the process of changing lanes or re-changing lanes at this time, return to S3-1; if the vehicle is in the process of returning to the original lane at this time, return to S3-2.

Images