A method for exploring the equilibrium strategy of interactive game between pedestrians and autonomous vehicles

Abstract

The present invention provides a method for exploring the equilibrium strategy of interactive game between pedestrians and autonomous driving vehicles, which belongs to the field of autonomous driving decision-making and traffic safety, including extracting the motion trajectory and traffic parameters of pedestrians and vehicles, constructing a following model and a vehicle expected speed function, constructing a human-vehicle game equilibrium model suitable for the incomplete state of mixed traffic flow information, and performing equilibrium solution on the human-vehicle game equilibrium model; using SUMO and Carla for joint simulation, analyzing the state and stability of the equilibrium point, and obtaining the "efficiency-safety" optimization scheme for the interactive game between pedestrians and autonomous driving vehicles in mixed traffic. Through the coupling modeling of the hybrid autonomous driving following model and the game model, the equilibrium point of the dynamic game model in the mixed traffic state is solved, solving the problem of shallow depth of theoretical modeling of the interactive game between hybrid autonomous driving and pedestrians in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving decision-making and traffic safety, and specifically to a method for exploring an interactive game equilibrium strategy between pedestrians and autonomous driving vehicles. Background Art

[0002] The present invention aims to study the conflict game problem of heterogeneous traffic flows of various traffic participants gathered at urban intersections when the penetration level of intelligent networked autonomous driving changes from low to high, especially the conflict problem between pedestrians crossing the street and intelligent vehicles (single-vehicle and multi-vehicle intelligence). Considering the autonomous driving traffic scenario, how to ensure the game equilibrium between pedestrians and autonomous driving vehicles at the same time, that is, under the general trend of intelligent networked autonomous driving, the relevant theories and solutions of traffic management and vehicle strategy formulation that take into account both safety and efficiency are urgent problems that need to be overcome.

[0003] Intelligent Connected Vehicle Technology Roadmap Figure 2 .0》Research and judgment shows that by 2025, the sales volume of PA and CA-class intelligent connected vehicles in my country will account for more than 50% of the total vehicle sales volume that year, and the C-V2X terminal new car assembly rate will reach 50%. In the future, the formation of mixed traffic flow of smart bicycles and intelligent connected vehicles is an inevitable trend under the development requirements of the automobile industry.

[0004] As the proportion of smart connected cars increases, although smart connected cars can perceive the information of surrounding pedestrians through computer vision and other means, and share information with other connected cars through big data platforms; however, since smart single vehicles can only obtain surrounding information through their own sensing devices, they cannot obtain roadside pedestrian traffic flow information in a timely manner. Therefore, in the case of smart single vehicles and multiple vehicles mixed driving, information is not shared, and it is difficult for pedestrians and vehicles to accurately identify and communicate information, which will increase the risk of conflict and potential efficiency loss (such as Figure 1 , 2 As shown in Figure 2, there are risks of interaction between people and vehicles at different intelligence levels. In addition, the risk of conflict between pedestrians and mixed-mode autonomous driving is constantly changing as the proportion of intelligent connected vehicles increases.

[0005] As the proportion of intelligent connected vehicles in the total number of intelligent vehicles increases from a small number to a large number, the penetration level of autonomous driving vehicles will also increase from low to high, and the phenomenon of mixed traffic flows with different intelligence will become increasingly prominent. At present, most of the relevant research focuses on the interactive game between pedestrians and a single traffic flow or the prediction and perception of pedestrians by autonomous driving vehicles. There are problems such as low depth of modeling of the interactive game between autonomous driving vehicles and mixed traffic flows and pedestrians, and insufficient revelation of the equilibrium game of human-vehicle interaction under this model.

[0006] Based on this, the present invention proposes a method for exploring the equilibrium strategy of the conflict game between pedestrians and hybrid intelligent vehicles, which is used to explore the safety issues between pedestrians and mixed traffic flows caused by incomplete information between hybrid intelligent vehicles at intersections. Summary of the invention

[0007] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a method for exploring the equilibrium strategy of the interactive game between pedestrians and autonomous driving vehicles. Through the coupling modeling of the hybrid autonomous driving following model and the game model, the autonomous driving following model is used to reflect the degree of influence of the degree of information non-sharing on the vehicle driving condition and is referenced to the dynamic game model. The dynamic game model is then used to reflect the dynamic feedback mechanism of the two parties in the game and the inherent characteristics of the behavior-risk two-way influence mechanism, so as to solve the equilibrium point of the dynamic game model under the mixed traffic state, and solve the problem of shallow depth of theoretical modeling of the hybrid autonomous driving and pedestrian interactive game in the prior art.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A method for exploring an interactive game equilibrium strategy between pedestrians and autonomous driving vehicles includes the following steps:

[0010] (1) Obtain a large number of videos of pedestrian-vehicle interaction scenes at intersections, and manually screen out video clips where driver-pedestrian interactions may occur in conflict areas; extract the motion parameters of pedestrians and vehicles, basic road parameters, and other relevant variable data that affect driving behavior and street crossing behavior from the screened video clips;

[0011] (2) constructing a car-following model, which includes a car-following model for a smart bicycle and a car-following model for a smart connected vehicle, and is used to obtain the acceleration of a smart bicycle or a smart connected vehicle in a mixed traffic state; constructing a vehicle expected speed function, and is used to obtain the vehicle expected speed;

[0012] (3) Construct a human-vehicle game equilibrium model suitable for the incomplete state of mixed traffic flow information. The human-vehicle game equilibrium model is embedded with a following model and a vehicle expected speed function. The vehicle acceleration and vehicle expected speed in the mixed traffic state obtained by the following model and the vehicle expected speed function are used as part of the explanatory variables and are included in the utility equation together with the key influencing factors affecting driving behavior and crossing behavior screened in step (1). The probability of the autonomous driving vehicle giving way and the probability of pedestrians crossing the street are expressed.

[0013] (4) constructing a solution model for the human-vehicle game equilibrium model in step (3), wherein the solution model is used to obtain the equilibrium point of the human-vehicle game equilibrium model in an incomplete state of mixed traffic flow information;

[0014] (5) The solution model in step (4) is applied to SUMO and Carla for joint simulation. The probability of pedestrians and vehicles adopting a specific strategy under single-point game and the iterative process of reaching the equilibrium point are deduced respectively. The stability of the equilibrium point is analyzed at the same time. By comprehensively considering the state and stability parameters of the equilibrium point, the "efficiency-safety" optimization scheme of the interactive game between pedestrians and autonomous driving vehicles in mixed traffic is obtained.

[0015] The smart bicycle following model is:

[0016] a K (t) = k[(γV K,i (Δx K,i (t-τ m ))+κV K,i+1 (Δx K,i+1 (t-τ c ))-v K (t)]+λ[βΔv K,i (t-τ m )] (1),

[0017] In the smart bike following model, there is no network data connection between smart bikes and smart connected cars. Only the front and rear radar sensors and machine vision provide perception information. Therefore, the acceleration expectation function of the smart bike following model is only related to the vehicle's own perception system, but has nothing to do with the mixed traffic state.

[0018] a K (t), v K (t) is the acceleration and speed of smart bike K at time t;

[0019] Δx K,i (t-τ m )=x i (t-τ m )-x K (t-τ m ) is the longitudinal position difference caused by the visual perception of the smart bicycle K to the adjacent front vehicle i; τ m Delay for machine vision perception;

[0020] Δv K,i (t-τ m )=v i (t-τ m )-v K (t-τ m ) is the longitudinal speed difference generated by the visual perception of the smart bicycle K to the adjacent front vehicle i;

[0021] Δx K,i+1 (t-τ c )=x K (t-τc )-x i+1 (t-τ c ) is the longitudinal position difference between the smart bicycle K and the adjacent following vehicle i+1 provided by the radar sensor;

[0022] τ c is the communication delay. In a smart bike, this item is the internal communication delay of the vehicle network. γ and β are the perception weights of autonomous driving. V K,i (.) represents the expected vehicle speed function generated by the interaction between the smart bike K and the adjacent front vehicle i, V K,i+1 (.) represents the expected vehicle speed function generated by the interaction between the smart vehicle K and the adjacent rear vehicle i+1;

[0023] κ is the influence of the front vehicle i on the autonomous driving decision, and k>0,λ>0 are sensitivity parameters.

[0024] The steps to build the intelligent connected car following model are as follows:

[0025] Assume that in the mixed traffic state of intelligent networked vehicles and intelligent bicycles, the intelligent networked vehicles are arranged in order from the front to the back as 0, 1, 2, ... j-1, j, j+1, ..., and intelligent bicycles are mixed in them at will;

[0026] Case 1: Assuming that the preceding vehicle of intelligent connected vehicle j is intelligent connected vehicle j-1, the following model of intelligent connected vehicles is:

[0027]

[0028] a j (t), v j (t) is the acceleration and speed of the intelligent connected vehicle j at time t;

[0029] Δx j,j-n (t-τ c )=x j-n (t-τ c )-x j (t-τ c ) is the longitudinal position difference between the intelligent networked vehicle j and the intelligent networked vehicle jn in front of it provided by the communication network;

[0030] Δv j,j-n (t-τ c )=v j-n (t-τ c )-v j (t-τ c ) is the longitudinal speed difference between the intelligent networked vehicle j and the intelligent networked vehicle jn in front of it provided by the communication network;

[0031] Δx j,i+1 (t-τc )=x j (t-τ c )-x j+1 (t-τ c ) is the distance between the intelligent connected vehicle j and the adjacent following vehicle i+1 provided by the radar sensor;

[0032] Δx j,j-1 (t-τ c )=x j-1 (t-τ c )-x j (t-τ c ) is the distance between the intelligent connected vehicle j and the intelligent connected vehicle j-1 provided by the radar sensor;

[0033] V j,j-n (.) represents the vehicle expected speed function generated by the interaction between intelligent connected vehicle j and intelligent connected vehicle jn;

[0034] τ c is the communication delay; k>0,λ>0 are sensitivity parameters; p is the vehicle expectation function coefficient, w j,j-n The influence weights of different preceding vehicles on the intelligent decision-making mechanism are:

[0035]

[0036] Case 2: Assuming that the adjacent preceding vehicle of smart connected vehicle j is smart single vehicle K, the following model of smart connected vehicles is:

[0037]

[0038] a j (t),v j (t) is the acceleration and speed of the intelligent connected vehicle j at time t;

[0039] Δx j,j-n (t-τ c )=x j-n (t-τ c )-x j (t-τ c ) is the longitudinal position difference between the autonomous driving vehicle j and the front intelligent connected vehicle jn provided by the communication network;

[0040] Δv j,j-n (t-τ c )=v j-n (t-τ c )-v j (t-τ c ) is the longitudinal speed difference between the autonomous driving vehicle j and the front intelligent connected vehicle jn provided by the communication network;

[0041] Δx j,i+1 (t-τ c )=x j (t-τ c )-x i+1 (t-τ c ) is the longitudinal position difference between vehicle j and the adjacent following vehicle i+1 provided by the radar sensor to the intelligent decision-making mechanism;

[0042] Δx j,K (.) is the longitudinal position difference generated by the machine vision perception of the intelligent connected vehicle j to the adjacent intelligent bicycle K in front;

[0043] Δv j,K (.) is the longitudinal speed difference generated by the machine vision perception of the intelligent connected vehicle j to the adjacent intelligent bicycle K in front;

[0044] V j,K (.) is the vehicle expected speed function generated by the interaction between the intelligent connected vehicle j and the intelligent single vehicle front vehicle K;

[0045] τ m is the machine vision perception delay; k>0,λ>0 are sensitivity parameters;γ,β are the autonomous driving perception weights; p is the vehicle expectation function coefficient, w j,j-n The influence weights of different preceding vehicles on the intelligent decision-making mechanism are:

[0046] V(Δx) is the desired speed function of the vehicle, which is represented by the following piecewise function:

[0047]

[0048] F(Δx) is a continuous monotonic function and can be expressed as:

[0049]

[0050] Δx st The distance difference between the person and the car to stop the car, Δx go v is the distance difference between the person and the car that the car is allowed to travel. max The vehicle is traveling at maximum speed.

[0051] The construction of a human-vehicle game equilibrium model suitable for an incomplete state of mixed traffic flow information includes:

[0052] (3.1) Define S as the choice set of game participants, S1 as the decision set of autonomous driving, and S2 as the choice set of pedestrians. It is stipulated that both parties in the game will take the most efficient decision during the interaction process, and assume that the other party also takes such a decision strategy, while considering the negative impact of risk changes on decision benefits;

[0053] (3.2) Substitute the movement trajectories and traffic parameters of pedestrians and vehicles extracted in step (1) into the following model and the vehicle's expected speed function to obtain the vehicle acceleration and vehicle's expected speed in the mixed traffic state; select the key influencing factors that affect driving behavior and street crossing behavior from the movement trajectories and traffic parameters of pedestrians and vehicles extracted in step (1), and incorporate these key influencing factors together with vehicle acceleration and vehicle's expected speed into the utility equation to obtain the driver and pedestrian decision benefit matrix, and use the Logit-QRE form to express the probability of the autonomous driving vehicle giving way and the probability of pedestrians crossing the street.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] 1. The present invention is based on a hybrid autonomous driving following model and combines it with quantum response equilibrium game theory to establish a dynamic game model of hybrid autonomous driving traffic flow and pedestrian intersections. The autonomous driving following model is used to reflect the degree of influence of information non-sharing on vehicle driving conditions. The dynamic game model is used to reflect the dynamic feedback mechanism of the two parties in the game and the inherent characteristics of the behavior-risk two-way influence mechanism. By combining the model and theory, the information non-sharing is reflected by the following model, thereby solving the equilibrium point of the dynamic game model under mixed traffic conditions, and solving the problem of shallow depth of modeling of hybrid autonomous driving and pedestrian interaction game theory in the prior art.

[0056] 2. The present invention is based on the joint simulation of SUMO and Carla, and uses a relatively mature solution to solve the probability of pedestrians and vehicles adopting different strategies and the iterative process of equilibrium points under single-point games, providing modeling ideas for interactive game models with different autonomous driving penetration levels. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 The figure is a schematic diagram of a conflict scenario between a smart bicycle and a pedestrian in the prior art.

[0058] Figure 2 This is a schematic diagram of a conflict scenario between an intelligent connected car and a pedestrian in the prior art.

[0059] Figure 3 It is a flow chart of the solution of the present invention.

[0060] Figure 4 The figure is a flow chart of the data acquisition steps of the present invention.

[0061] Figure 5 It is a schematic diagram of the hybrid autonomous driving scenario information interaction of the present invention.

[0062] Figure 6 It is a schematic diagram of the following state between intelligent connected vehicles of the present invention.

[0063] Figure 7It is a schematic diagram of the following state of the intelligent network-connected vehicle and the intelligent bicycle of the present invention. DETAILED DESCRIPTION

[0064] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0065] like Figure 3-7 As shown, the present invention discloses a method for exploring an interactive game equilibrium strategy between a pedestrian and an autonomous driving vehicle, comprising:

[0066] Step 1: Obtain experimental model data. Obtain a large number of videos of human-vehicle interaction scenes at intersections, and manually screen out the video clips that need to be intercepted (i.e., clips where vehicles and pedestrians may interact in the conflict area, especially when vehicles turn right and interact with pedestrians).

[0067] Step 2: Extraction of vehicle and pedestrian movement trajectories and traffic parameters. Figure 4 As shown in the figure, computer vision models such as YOLO X and R-CNN are used to collect trajectory data of pedestrians and vehicles in step one, and the motion parameters of pedestrians and vehicles, basic road parameters, and related variable data affecting driving behavior and crossing behavior are extracted; including the number of pedestrians crossing and waiting in the surrounding area within a certain period of time before and after the pedestrians cross the street, pedestrian flow density, crossing speed, etc.; as well as vehicle traffic volume, density, speed, headway, saturation, traffic capacity, headway and other parameters.

[0068] Step 3: Establishment of hybrid autonomous driving vehicle following model. Under mixed traffic flow, considering the changes of pedestrian cognition and vehicle strategy with the degree of information non-sharing, the autonomous driving following model (including smart bicycles and smart connected cars) is used to bring in the key data in step 2 and the unique settings of the vehicle model to obtain the motion parameters of smart bicycles and smart connected cars in mixed traffic. The construction steps are as follows:

[0069] Use the smart bicycle following model and the smart connected car following model respectively:

[0070] 1) Intelligent bicycle following model:

[0071] a K (t) = k[(γV K,i (Δx K,i (t-τ m ))+κV K,i+1 (Δx K,i+1 (t-τ c ))-v K (t)]+λ[βΔv K,i (t-τm )] (1)

[0072] In this model, there is no network data connection between smart bicycles and smart connected cars. Only the front and rear radar sensors and machine vision provide perception information. Therefore, the acceleration expectation function of the smart bicycle following model is only related to the vehicle's own perception system, but has nothing to do with the mixed traffic state. Figure 5 shown, but Figure 5 Only part of the mixed traffic status is shown in the figure. It does not matter whether the smart bicycle K is a smart bicycle or a smart connected car.

[0073] a K (t),v K (t) is the acceleration of smart bike K at time t (m / s 2 ), speed (m / s);

[0074] Δx K,i (t-τ m )=x i (t-τ m )-x K (t-τ m ) is the longitudinal position difference generated by the visual perception of the smart bicycle K to the adjacent front vehicle i;

[0075] Δv K,i (t-τ m )=v i (t-τ m )-v K (t-τ m ) is the longitudinal speed difference caused by the visual perception of the smart bicycle K to the adjacent front vehicle i; τ m Delay for machine vision perception;

[0076] Δx K,i+1 (t-τ c )=x K (t-τ c )-x i+1 (t-τ c ) is the longitudinal position difference between the smart bicycle K and the adjacent following vehicle i+1 provided by the radar sensor;

[0077] τ c is the communication delay. In a smart bike, this item is the internal communication delay of the vehicle network. γ and β are the perception weights of autonomous driving. K,i (.) represents the expected vehicle speed function generated by the interaction between the smart bike K and the adjacent front vehicle i, V K,i+1 (.) represents the expected vehicle speed function generated by the interaction between the smart bicycle K and the adjacent rear vehicle i+1, which is consistent with the symbolic physical meaning of formula (2).

[0078] κ is the influence of the front vehicle i on the autonomous driving decision. k>0,λ>0 are sensitivity parameters.

[0079] V(Δx) is the desired speed function of the vehicle, which is represented by the following piecewise function:

[0080]

[0081] F(Δx) is a continuous monotonic function and can be expressed as:

[0082]

[0083] Δx is the longitudinal position difference transmitted by the sensor, Δx st The distance difference required to stop the car, Δx go is the distance difference allowed for the car to travel, v max The vehicle speed is no more than the maximum speed of the current traffic speed limit (v max Appropriate adjustments will be made in different traffic scenarios). In the smart bike following model, Δx is Δx K,i (t-τ m ), is a variable that is a function of the desired vehicle speed.

[0084] 2) When intelligent networked vehicles and intelligent bicycles are mixed, the intelligent networked vehicles are arranged in order from the front to the back as 0, 1, 2, ... j-1, j, j+1, ..., and intelligent bicycles are mixed in them at will;

[0085] Case 1: Assume that the preceding vehicle of smart connected vehicle j is smart connected vehicle j-1 (e.g. Figure 6 As shown, Figure 6 It also only illustrates some mixed traffic states in this case. For example, the adjacent following vehicle of smart connected car j can be a smart connected car or a smart bicycle. Then the following model of the smart connected car is:

[0086]

[0087] a j (t),v j (t) is the acceleration of the intelligent connected vehicle j at time t (m / s 2 ), speed (m / s);

[0088] Δx j,j-n (t-τ c )=x j-n (t-τ c )-x j (t-τ c ) is the longitudinal position difference between the autonomous driving vehicle j and the intelligent connected front vehicle jn provided by the communication network;

[0089] Δv j,j-n (t-τ c )=v j-n (t-τ c )-v j (t-τ c ) is the longitudinal speed difference between the autonomous driving vehicle j and the intelligent connected front vehicle jn provided by the communication network;

[0090] Δx j,i+1 (t-τ c )=x j (t-τ c )-x j+1 (t-τ c ) is the distance between vehicle j and its adjacent following vehicle i+1 provided by the radar sensor to the intelligent decision-making mechanism;

[0091] Δx j,j-1 (t-τ c )=x j-1 (t-τ c )-x j (t-τ c ) is the distance between vehicle j and the adjacent preceding vehicle j-1 provided by the radar sensor to the intelligent decision-making mechanism;

[0092] V j,j-n (.) represents the expected vehicle speed function generated by the interaction between the intelligent connected vehicle j and the preceding intelligent connected vehicle jn, τ c is the communication delay; k>0,λ>0 are sensitivity parameters.

[0093] p is the vehicle expectation function coefficient, w j,j-n The influence weights of different preceding vehicles on the intelligent decision-making mechanism are:

[0094]

[0095] Case 2: Assuming that the preceding vehicle of smart connected vehicle j is smart bicycle K (e.g. Figure 7 As shown, Figure 7 It also only illustrates some mixed traffic states in this case. For example, the adjacent following vehicle of smart connected car j can be a smart connected car or a smart bicycle. Then the following model of the smart connected car is:

[0096]

[0097] a j (t),v j (t) is the acceleration of the intelligent connected vehicle j at time t (m / s 2 ), speed (m / s);

[0098] Δx j,j-n(t-τ c )=x j-n (t-τ c )-x j (t-τ c ) is the longitudinal position difference between the autonomous driving vehicle j and the front intelligent connected vehicle jn provided by the communication network;

[0099] Δv j,j-n (t-τ c )=v j-n (t-τ c )-v j (t-τ c ) is the longitudinal speed difference between the autonomous driving vehicle j and the front intelligent connected vehicle jn provided by the communication network;

[0100] Δx j,i+1 (t-τ c )=x j (t-τ c )-x i+1 (t-τ c ) is the longitudinal position difference between the intelligent connected vehicle j and its adjacent following vehicle i+1 provided by the radar sensor;

[0101] Δx j,K (.) is the longitudinal position difference generated by the machine vision perception of the intelligent connected vehicle j to the adjacent intelligent bicycle K;

[0102] Δv j,K (.) is the longitudinal speed difference generated by the machine vision perception of the intelligent connected vehicle j to the adjacent intelligent bicycle K;

[0103] V j,K (.) is the expected vehicle speed function generated by the interaction between the intelligent connected vehicle j and the intelligent single vehicle K,

[0104] τ m is the machine vision perception delay; k>0,λ>0 are sensitivity parameters;γ,β are the autonomous driving perception weights; p is the vehicle expectation function coefficient, w j,j-n is the influence weight of different preceding vehicles on the intelligent decision-making mechanism, see formula (5);

[0105] From equations (1), (4), (5) and the physical quantities in them, it can be seen that the motion parameters of the vehicle during car-following will change with the degree of vehicle interconnection information. That is, the degree of information non-sharing will affect the car-following state of each intelligent connected car and the decision-making influence during car-following.

[0106] Step 4: Modeling of quantum response game of human-vehicle interaction based on mixed traffic state with incomplete information. The following model of step 3 is embedded into the quantum response equilibrium game model to construct a human-vehicle game equilibrium model suitable for mixed traffic flow information with incomplete state:

[0107] S is defined as the choice set of game participants, S1 is the decision set of autonomous driving, and S2 is the choice set of pedestrians. It is stipulated that both parties in the game will take the most efficient decision during the interaction process, and assume that the other party will also adopt such a decision-making strategy, while considering the negative impact of risk changes on decision-making benefits.

[0108] Assume that the autonomous driving decision S1 = {yield, do not yield}, and the pedestrian decision S2 = {cross the street, do not cross the street}.

[0109] The data from step 2 and the key influencing factors from step 3 are incorporated into the establishment of the utility equation, which is defined as follows:

[0110] 1) The expected utility function of the behavior that a pedestrian can choose is expressed as:

[0111] EU 过街 =P 让行 ×aU+c1 (7)

[0112] EU 不过街 =bV+c2 (8)

[0113] 2) The expected utility function of the driver's choice of behavior is expressed as:

[0114] EU 让行 =dM+c3 (9)

[0115] EU 不让行 =(1-P 过街 )×eW+c4 (10)

[0116] Among them, EU is the utility when the driver or pedestrian adopts the relevant strategy, U and V are the explanatory variable vectors of pedestrians crossing the street and not crossing the street, M and W are the explanatory variable vectors of drivers giving way and not giving way. c1, c2, c3, c4 are constant terms, a, b, d, e are the vectors of coefficients, 0≤P 让行 ≤1, respectively, pedestrians’ expectations that drivers will give way, 0≤P 过街 ≤1 represents the driver’s expectation that pedestrians will cross the road.

[0117] The explanatory variables for pedestrians crossing the street, the explanatory variables for not crossing the street, the explanatory variables for drivers giving way, and the explanatory variables for drivers not giving way are screened out from the pedestrian and vehicle motion parameters, basic road parameters, other related variable data affecting driving behavior and crossing behavior extracted in step two, and the vehicle acceleration and expected vehicle speed in step three. The screening criteria are: the Pearson correlation coefficient between the variables is less than 0.5, and according to the Boostrap sampling t-test, the mean of the variable is at the 5% level; the screened key variables are brought into the corresponding expected utility function to obtain the benefit value, and then the driver and pedestrian decision benefit matrix is ​​obtained.

[0118] In order to associate the autonomous driving decision-making mechanism with the pedestrian learning cognitive change mechanism, the Logit-QRE form is proposed to express the probability of the autonomous driving vehicle giving way and the probability of pedestrians crossing the street:

[0119] The probability that the driver gives way is:

[0120]

[0121] The probability of a pedestrian crossing the street is:

[0122]

[0123] Step 5: Equilibrium point solution model construction. It is proposed to start from the judgment conditions of quantum response equilibrium and combine the strategy change trends of both parties in multiple games. The equilibrium point solution is as follows:

[0124] The model parameters are estimated using the maximum likelihood estimation method:

[0125] ΔEU crossing and ΔEU giving way are potential indices of pedestrian and driver decision making.

[0126] ΔEU 过街 =EU 过街 -EU 不过街 (13)

[0127] ΔEU 让行 =EU 让行 -EU 不让行 (14)

[0128] Then construct the log-likelihood function of pedestrian decision:

[0129]

[0130] Among them, a is the parameter coefficient of the utility equation for people crossing the street, S is the choice set of game participants, and y i =1 means the pedestrian chooses the action {cross the street}, y i = 0 means the pedestrian chooses the action {not to cross the street}, I{yi =1} represents the utility coefficient of pedestrian i when choosing a crossing strategy, I{y i =0} represents the utility coefficient of pedestrian i when he chooses the strategy of not crossing the street, is the cumulative distribution function of the logistic distribution.

[0131] Similarly, the log-likelihood function of autonomous driving decision can be constructed:

[0132]

[0133] Where b is the parameter coefficient of the autonomous driving yielding utility equation, y j =1 means the driver chooses to operate {yield}, y j =0 indicates that the driver chooses the operation {do not give way}.

[0134] Therefore, using μ to represent the vector of all model parameters, the log-likelihood function can be expressed as:

[0135] LL(μ; y; X) = LL 行人 (a; S, V) + LL 车辆 (b; S, W) (17)

[0136] y represents the decision action of the interaction, and X is the parameter set of the utility function;

[0137] The equilibrium point of each game is LL, LL 行人 and LL 车辆 are all in the optimal state. This is a multi-objective optimization problem, namely: argmax{LL(μ; y, X), LL 行人 (a; S, V), LL 车辆 (b; S, W)}.

[0138] Parameter Estimation - Expectation-Maximization Algorithm

[0139] Expectation maximization (EM) can be applied iteratively to solve the logit-QRE. 让行 and P 过街 can be considered as latent variables; for a pair of initial probabilities {P 过街，m ,P 让行,m}, generate μ by maximizing the entire log-likelihood function m :

[0140]

[0141] In the formula, m represents the number of iterations.

[0142] Step 6: Simulation. Combined with SUMO and Carla for joint simulation, the probability of pedestrians and vehicles adopting a specific strategy under single-point game and the iterative process of reaching the equilibrium point are deduced respectively. At the same time, the stability of the equilibrium point is analyzed, and the "efficiency-safety" optimization solution for the interactive game between pedestrians and autonomous driving vehicles in mixed traffic can be obtained.

[0143] What is described above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for exploring the equilibrium strategy of interactive game between pedestrians and autonomous driving vehicles, characterized in that: The following steps are involved: (1) Obtain a large number of videos of human-vehicle interaction scenes at intersections, and manually screen out video clips of possible interactions between drivers and pedestrians in conflict areas; Extract the motion parameters of pedestrians and vehicles, basic road parameters, and other relevant variable data that affect driving and crossing behaviors from the screened video clips; (2) constructing a car-following model, which includes a car-following model for a smart bicycle and a car-following model for a smart connected vehicle, and is used to obtain the acceleration of a smart bicycle or a smart connected vehicle in a mixed traffic state; constructing a vehicle expected speed function, and is used to obtain the vehicle expected speed; (3) Construct a human-vehicle game equilibrium model suitable for the incomplete state of mixed traffic flow information. The human-vehicle game equilibrium model is embedded with a car-following model and a vehicle expected speed function. The vehicle acceleration and vehicle expected speed in the mixed traffic state obtained by the car-following model and the vehicle expected speed function are used as part of the explanatory variables and are incorporated into the utility equation together with the key influencing factors affecting driving behavior and street crossing behavior screened out in step (1). The probability of the autonomous driving vehicle giving way and the probability of pedestrians crossing the street are expressed. (4) constructing a solution model for the human-vehicle game equilibrium model in step (3), wherein the solution model is used to obtain the equilibrium point of the human-vehicle game equilibrium model in an incomplete state of mixed traffic flow information; (5) Apply the solution model in step (4) to SUMO and Carla for joint simulation, and deduce the probability of pedestrians and vehicles adopting a specific strategy under single-point game, as well as the iterative process of reaching the equilibrium point. At the same time, analyze the stability of the equilibrium point. By comprehensively considering the state and stability parameters of the equilibrium point, the "efficiency-safety" optimization scheme of the interactive game between pedestrians and autonomous vehicles in mixed traffic is obtained; The smart bicycle following model is: a K (t)=k[(γV K,i (Δx K,i (t-t m ))+κV K,i+1 (Δx K,i+1 (t-t c ))-v K (t)]+λ[βΔv K,i (t-t m )] (1), In the smart bike following model, there is no network data connection between smart bikes and smart connected cars. Only the front and rear radar sensors and machine vision provide perception information. Therefore, the acceleration expectation function of the smart bike following model is only related to the vehicle's own perception system, but has nothing to do with the mixed traffic state. a K (t), v K (t) is the acceleration and speed of smart bike K at time t; Δx K,i (t-τ m )=x i (t-τ m )-x K (t-τ m ) is the longitudinal position difference caused by the visual perception of the smart bicycle K to the adjacent front vehicle i; τ m Delay for machine vision perception; Δv K,i (t-τ m )=v i (t-τ m )-v K (t-τ m ) is the longitudinal speed difference generated by the visual perception of the smart bicycle K to the adjacent front vehicle i; Δx K,i+1 (t-τ c )=x K (t-τ c )-x i+1 (t-τ c ) is the longitudinal position difference between the smart bicycle K and the adjacent following vehicle i+1 provided by the radar sensor; τ c is the communication delay. In a smart bike, this item is the internal communication delay of the vehicle network. γ and β are the perception weights of autonomous driving. V K,i (.) represents the expected vehicle speed function generated by the interaction between the smart bike K and the adjacent front vehicle i, V K,i+1 (.) represents the expected vehicle speed function generated by the interaction between the smart vehicle K and the adjacent rear vehicle i+1; κ is the influence of the front vehicle i on the autonomous driving decision, k and λ are sensitivity parameters, k>0, λ>0; The steps to build the intelligent connected car following model are as follows: Assume that in the mixed traffic state of intelligent networked vehicles and intelligent bicycles, the intelligent networked vehicles are arranged in order from the front to the back as 0, 1, 2, ... j-1, j, j+1, ..., and intelligent bicycles are mixed in them at will; Case 1: Assuming that the preceding vehicle of intelligent connected vehicle j is intelligent connected vehicle j-1, the following model of intelligent connected vehicles is: a j (t), v j (t) is the acceleration and speed of the intelligent connected vehicle j at time t; Δx j,j-n (t-τ c )=x j-n (t-τ c )-x j (t-τ c ) is the longitudinal position difference between the intelligent networked vehicle j and the intelligent networked vehicle jn in front thereof provided by the communication network; Δv j,j-n (t-τ c )=v j-n (t-τ c )-v j (t-τ c ) is the longitudinal speed difference between the intelligent networked vehicle j and the intelligent networked vehicle jn in front of it provided by the communication network; Δx j,i+1 (t-τ c )=x j (t-τ c )-x i+1 (t-τ c ) is the distance between the intelligent connected vehicle j and the adjacent following vehicle i+1 provided by the radar sensor; Δx j,j-1 (t-τ c )=x j-1 (t-τ c )-x j (t-τ c ) is the distance between the intelligent connected vehicle j and the intelligent connected vehicle j-1 provided by the radar sensor; V j,j-n (.) represents the vehicle expected speed function generated by the interaction between intelligent connected vehicle j and intelligent connected vehicle jn; τ c is the communication delay; k and λ are sensitivity parameters, k>0, λ>0; p is the vehicle expectation function coefficient, w j,j-n The influence weights of different preceding vehicles on the intelligent decision-making mechanism are: Case 2: Assuming that the adjacent preceding vehicle of smart connected vehicle j is smart single vehicle K, the following model of smart connected vehicles is: a j (t),v j (t) is the acceleration and speed of the intelligent connected vehicle j at time t; Δx j,j-n (t-τ c )=x j-n (t-τ c )-x j (t-τ c ) is the longitudinal position difference between the intelligent networked vehicle j and the preceding intelligent networked vehicle jn provided by the communication network; Δv j,j-n (t-τ c )=v j-n (t-τ c )-v j (t-τ c ) is the longitudinal speed difference between the intelligent connected vehicle j and the preceding intelligent connected vehicle jn provided by the communication network; Δx j,i+1 (t-τ c )=x j (t-τ c )-x i+1 (t-τ c ) is the longitudinal position difference between the intelligent connected vehicle j and the adjacent following vehicle i+1 provided by the radar sensor; Δx j,K (.) is the longitudinal position difference generated by the machine vision perception of the intelligent connected vehicle j to the adjacent intelligent bicycle K in front; Δv j,K (.) is the longitudinal speed difference generated by the machine vision perception of the intelligent connected vehicle j to the adjacent intelligent bicycle K in front; V j,K (.) is the vehicle expected speed function generated by the interaction between the intelligent connected vehicle j and the adjacent intelligent bicycle K in front; τ m is the machine vision perception delay; k and λ are sensitivity parameters, k>0, λ>0; γ, β are the autonomous driving perception weights; p is the vehicle expectation function coefficient, w j,j-n The influence weights of different preceding vehicles on the intelligent decision-making mechanism are: V(Δx) is the desired speed function of the vehicle, which is represented by the following piecewise function: F(Δx) is a continuous monotonic function, expressed as: Δx st The distance difference between the person and the car to stop the car, Δx go v is the distance difference between the person and the car that the car is allowed to travel. max The vehicle is traveling at maximum speed.

2. According to claim 1, a method for exploring the equilibrium strategy of interactive game between pedestrians and autonomous driving vehicles is characterized in that: The construction of a human-vehicle game equilibrium model suitable for an incomplete state of mixed traffic flow information includes: (3.1) Define S as the choice set of game participants, S1 as the decision set of autonomous driving, and S2 as the choice set of pedestrians. It is stipulated that both parties in the game will take the most efficient decision during the interaction process, and assume that the other party also takes such a decision strategy, while considering the negative impact of risk changes on decision benefits; (3.2) Substitute the motion trajectories and traffic parameters of pedestrians and vehicles extracted in step (1) into the car-following model and the vehicle's expected speed function to obtain the vehicle acceleration and vehicle's expected speed in the mixed traffic state; select the key influencing factors affecting driving behavior and street crossing behavior from the motion trajectories and traffic parameters of pedestrians and vehicles extracted in step (1), and incorporate these key influencing factors together with the vehicle acceleration and vehicle's expected speed into the utility equation to obtain the driver and pedestrian decision benefit matrix, and use the Logit-QRE form to express the probability of the autonomous driving vehicle giving way and the probability of pedestrians crossing the street.

Images