A method for resolving conflicts between autonomous driving vehicles based on differential game decision modeling

Abstract

The present invention discloses a method for resolving conflicts of autonomous driving vehicles based on differential game decision modeling, aiming to improve the safety and efficiency of autonomous driving vehicle decision-making. The method first defines the driving aggressiveness of autonomous driving vehicles in achieving human-like driving and personalized decision-making at unsignaled intersections, and then proposes a collision risk assessment algorithm based on Gaussian potential field method and event trigger mechanism. By combining the differential game method to model the human-like interaction and decision-making of autonomous driving vehicles, the utility function and cost function are finally defined to solve the driving conflict problem of autonomous driving vehicles at multi-lane unsignaled intersections. This method of predicting collision risks by combining differential game decision modeling and Gaussian potential field method can effectively improve the decision-making, safety and efficiency of autonomous driving vehicles, and provide technical support for the development of autonomous driving technology and traffic safety technology.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology and traffic safety technology, and in particular to a method for resolving conflicts among autonomous driving vehicles based on differential game decision modeling. Background Art

[0002] At multi-lane intersections without signal control, autonomous vehicles must make independent decisions and deal with dynamic interactions with other vehicles. Traditional optimization methods, such as Monte Carlo Tree Search (MCTS) and multi-vehicle coordination strategies, are effective in some scenarios, but they are computationally complex and difficult to respond in real time when dealing with complex interactions and highly dynamic environments, especially when dealing with personalized driving conflicts of autonomous vehicles at multi-lane unsignaled intersections. The combination of differential game modeling methods and Gaussian potential field methods to predict collision risks provides a new idea for improving the computational efficiency of decision-making algorithms and ensuring driving safety. Summary of the invention

[0003] The purpose of the present invention is to provide a method for resolving conflicts of autonomous driving vehicles based on differential game decision modeling. The method first defines the driving aggressiveness of autonomous driving vehicles in achieving human-like driving and personalized decision-making at unsignaled intersections, and then proposes a collision risk assessment algorithm based on Gaussian potential field method and event trigger mechanism. The human-like interaction and decision-making of autonomous driving vehicles are modeled by combining differential game methods, and finally the driving conflict problem of autonomous driving vehicles at multi-lane unsignaled intersections is solved by defining utility functions and cost functions. In particular, the invention improves the computational efficiency of the decision algorithm by predicting collision risks and decision modeling.

[0004] To achieve the above functions, the present invention designs a method for resolving conflicts among autonomous driving vehicles based on differential game decision modeling. For autonomous driving vehicles, the following steps S1 to S5 are executed to complete the optimal driving path strategy selection of the autonomous driving vehicle at the intersection:

[0005] Step S1: establishing a driving aggressiveness level of the autonomous driving vehicle based on the dynamic behavior of the autonomous driving vehicle, including acceleration, deceleration, and steering operations;

[0006] Step S2: Based on the driving aggressiveness level of the autonomous driving vehicle, a Gaussian potential field model is constructed using the current position, speed and predicted path of the autonomous driving vehicle, and the collision risk potential field value of the autonomous driving vehicle is calculated according to the Gaussian potential field model to evaluate the collision risk of the autonomous driving vehicle at an unsignalized intersection and determine the potential risk level;

[0007] Step S3: formulating the driving path decision process of the autonomous driving vehicle through differential game theory, including solving Nash equilibrium and Stackelberg equilibrium, combining the benefits of different driving path decisions of the autonomous driving vehicle, analyzing and determining the optimal driving path strategy of each vehicle at the intersection;

[0008] Step S4: defining a utility function for each autonomous vehicle, which combines the revenue of the autonomous vehicle at the end of the trip and the operating cost during the entire trip to evaluate the benefits of different driving path decisions;

[0009] Step S5: Define the operating cost function of each autonomous driving vehicle. The operating cost function is combined with the collision risk potential field value of the autonomous driving vehicle to evaluate driving safety and passing efficiency, thereby ensuring driving safety while improving traffic efficiency and completing the optimal driving path strategy selection for the autonomous driving vehicle at the intersection.

[0010] Beneficial effects: Compared with the prior art, the advantages of the present invention include:

[0011] 1. The present invention combines the differential game modeling method and the Gaussian potential field method to predict the collision risk, and has high computational efficiency.

[0012] 2. The differential modeling method of the present invention utilizes utility functions and cost functions to make safe and efficient decisions for autonomous vehicles.

[0013] 3. Compared with the traditional decision-making method, the present invention defines the driving aggressiveness of the autonomous driving car to achieve human-like driving and personalized decision-making at an unsignalized intersection, ensures the driver's personalized driving preference, and can adapt to more diverse driving characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flowchart of a method for resolving conflicts of autonomous driving vehicles based on differential game decision modeling provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0016] An embodiment of the present invention provides a method for resolving conflicts of autonomous driving vehicles based on differential game decision modeling, for autonomous driving vehicles, referring to Figure 1 , execute the following steps S1 to S5 to complete the optimal driving path strategy selection of the autonomous driving vehicle at the intersection:

[0017] Step S1: establishing a driving aggressiveness level of the autonomous driving vehicle based on the dynamic behavior of the autonomous driving vehicle, including acceleration, deceleration, and steering operations, and determining the driving aggressiveness level in a numerical range of 0 to 1 by comprehensively considering various factors;

[0018] The driving aggressiveness level of an autonomous vehicle is as follows:

[0019]

[0020] Where κ is the driving aggressiveness level, α, β, and γ are weight coefficients, a(t) is the instantaneous acceleration of the autonomous vehicle at time t, a(t′) is the derivative of the instantaneous acceleration of the autonomous vehicle at time t with respect to time t, d(t) is the instantaneous deceleration of the autonomous vehicle at time t, and θ is k is the steering angle at the kth time step, t is the time period currently considered, and n is the length of the steering angle sequence.

[0021] Step S2: Based on the driving aggressiveness level of the autonomous driving vehicle, a Gaussian potential field model is constructed using the current position, speed and predicted path of the autonomous driving vehicle, and the collision risk potential field value of the autonomous driving vehicle is calculated according to the Gaussian potential field model to evaluate the collision risk of the autonomous driving vehicle at an unsignalized intersection and determine the potential risk level;

[0022] The Gaussian potential field model constructed in step S2 is as follows:

[0023]

[0024]

[0025]

[0026] In the formula, Γ i is the collision risk potential field value of autonomous driving vehicle i, Λ i is the amplitude coefficient of the potential field, a i is the acceleration of the autonomous driving vehicle i, x is the current position vector of the vehicle, and x ci is the steering center position vector of the autonomous driving vehicle i, ∑ i is the covariance matrix describing the position uncertainty of autonomous vehicle i, σ x represents the position uncertainty in the x direction, σ y represents the position uncertainty in the y direction, v i is the velocity vector of autonomous vehicle i, σ v is the standard deviation of velocity, is a fixed value, κ i is the driving aggressiveness level of the autonomous vehicle i.

[0027] Step S3: formulating the driving path decision process of the autonomous driving vehicle through differential game theory, including solving Nash equilibrium and Stackelberg equilibrium, combining the benefits of different driving path decisions of the autonomous driving vehicle, analyzing and determining the optimal driving path strategy of each vehicle at the intersection;

[0028] The specific steps of the process of making a driving path decision for the autonomous driving vehicle in step S3 are as follows:

[0029] Step S3.1: Establish the dynamic system model as follows:

[0030]

[0031]

[0032] f i1 (x i ,u i )=v i (t)

[0033]

[0034]

[0035] f i4 (x i ,u i )=f(v i (t),δ i (t),μ,…)

[0036] …

[0037] g ij1 (x i ,x j )=ψ(∥∥p i -p j ∥∥)

[0038] g ij2 (v i ,v j )=φ(v i -v j )

[0039] g ij3 (δ i ,δ j )=χ(δ i ,δ j )

[0040] …

[0041]

[0042]

[0043] In the formula, is the rate of change of the state variables of the autonomous driving vehicle i over time, F i is a function of the dynamics of the autonomous driving vehicle i itself, G ij is the interaction dynamics between the autonomous driving vehicle i and other vehicles j, f ik is the kth element function of the dynamics of the autonomous vehicle i, k∈{1,2,…,m}, including the dynamics of the position f i1 , dynamics of speed i2 , dynamics of the steering angle i3 , dynamic f of friction i4 , v i (t) is the speed of vehicle i at time t, a i (t) is the acceleration of vehicle i at time t, F i (t) is the force applied by the accelerator to vehicle i at time t, m is the mass of the vehicle, is the steering angular velocity of vehicle i at time t, μ is the friction coefficient between the tire and the ground, g ijl is the lth element function of the interaction dynamics between the autonomous vehicle i and other vehicles j, l∈{1,2,…,N}, including the relative position effect g ij1 , speed matching affects g ij2 , steering effect g ij3 , ψ is a function of distance, p i and p j is the position vector of vehicle i and vehicle j, φ is a function of the speed difference, v i and v j is the velocity vector of vehicle i and vehicle j, χ is the function of the steering angle, ← i and δ j is the steering angle of vehicle i and vehicle j, x i and x j are the state vectors of autonomous driving vehicle i and other vehicles j, u i and u j are the control input vectors of autonomous driving vehicle i and other vehicles j, u i1 and u j1 is the acceleration control of the autonomous driving vehicle i and other vehicles j, u i2 and u j2 is the steering control of the autonomous driving vehicle i and other vehicles j, u iN and u jN are other control parameters of the autonomous driving vehicle i and other vehicles j;

[0044] Step S3.2: Solve the Nash equilibrium as follows:

[0045]

[0046] Constraints:

[0047] u j Given, and x i ,u i Satisfy system dynamics and constraints;

[0048] In the formula, is the optimal control strategy for autonomous vehicle i, U i is the utility function of autonomous vehicle i, x i is the state vector of autonomous vehicle i, u i is the control input vector of autonomous vehicle i, F i is a function of the dynamics of the autonomous vehicle i itself, G ij represents the interaction dynamics between autonomous vehicle i and other vehicles j, j≠i represents all other vehicles;

[0049] Step S3.3: Solve the Stackelberg equilibrium as follows:

[0050]

[0051]

[0052] Constraints:

[0053] x i ,u i For leaders, x j ,u j For followers, satisfying system dynamics and constraints;

[0054] In the formula, is the optimal control strategy of leader i, i.e., autonomous driving vehicle i, U i is the utility function of leader i, is the optimal control strategy of follower j, i.e. other vehicle j, U j is the utility function of follower j, x j is the state vector of follower j.

[0055] Step S4: defining a utility function for each autonomous vehicle, which combines the revenue of the autonomous vehicle at the end of the trip and the operating cost during the entire trip to evaluate the benefits of different driving path decisions;

[0056] The utility function U of each autonomous driving vehicle defined in step S4 i , that is, the utility function U in the Stackelberg equilibrium i As follows:

[0057]

[0058]

[0059] C i，utility (x i (t),u i (t)) = D i x i (t)+E i u i (t),w(t)=exp(-λt 2 )

[0060] Where U i (x i ,u i ) is the utility function of autonomous driving vehicle i, Φ i (x Ti ) indicates that at the end of the trip x Ti The profit function when C i，utility represents the operating cost function of autonomous vehicle i, w(t) is the time decay function, κ i is the driving aggressiveness level of the autonomous vehicle i, Q i，utility is the weight matrix for adjusting the impact of the state on the utility, which is used to quantify the deviation or utility loss between the vehicle state and the ideal state. i is the weight matrix of the control strategy’s impact on utility, D i and E i is the coefficient matrix in cost calculation, and λ is the time decay coefficient.

[0061] Step S5: Define the operating cost function of each autonomous driving vehicle. The operating cost function is combined with the collision risk potential field value of the autonomous driving vehicle to evaluate driving safety and passing efficiency, thereby ensuring driving safety while improving traffic efficiency and completing the optimal driving path strategy selection for the autonomous driving vehicle at the intersection.

[0062] The operating cost function of the autonomous driving vehicle defined in step S5 is as follows:

[0063] C i，operational =η s S i (x i ,u i )+η e E i (x i ,ui )

[0064]

[0065]

[0066] In the formula, C i，operational is the operating cost function of the autonomous driving vehicle i, η s is the safety weight coefficient, η e is the efficiency weight function, S i is the safety cost function of autonomous driving vehicle i, E i is the efficiency cost function of autonomous driving vehicle i, Γ i is the collision risk potential field value of autonomous driving vehicle i, x i (t) is the state vector of the autonomous driving vehicle i at time t, u i (t) is the control input vector of the autonomous driving vehicle i at time t, x goal is the target state of the autonomous driving vehicle i, u opt is the optimal control strategy of autonomous driving vehicle i, P i is the weight matrix of the quantized state, Q i , operational is a weighted matrix of control strategy deviations, used to weigh different factors related to operating costs.

[0067] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A method for resolving conflicts among autonomous driving vehicles based on differential game decision modeling, characterized in that: For the autonomous driving vehicle, the following steps S1 to S5 are executed to complete the optimal driving path strategy selection of the autonomous driving vehicle at the intersection: Step S1: establishing a driving aggressiveness level of the autonomous driving vehicle based on the dynamic behavior of the autonomous driving vehicle, including acceleration, deceleration, and steering operations; The driving aggressiveness level of the autonomous driving vehicle in step S1 is as follows: Where κ is the driving aggressiveness level, α, β, and γ are weight coefficients, a(t) is the instantaneous acceleration of the autonomous vehicle at time t, a(t′) is the derivative of the instantaneous acceleration of the autonomous vehicle at time t with respect to time t, d(t) is the instantaneous deceleration of the autonomous vehicle at time t, and θ is k is the steering angle at the kth time step, t is the time period currently considered, and n is the length of the steering angle sequence; Step S2: Based on the driving aggressiveness level of the autonomous driving vehicle, a Gaussian potential field model is constructed using the current position, speed and predicted path of the autonomous driving vehicle, and the collision risk potential field value of the autonomous driving vehicle is calculated according to the Gaussian potential field model to evaluate the collision risk of the autonomous driving vehicle at an unsignalized intersection and determine the potential risk level; Step S3: formulating the driving path decision process of the autonomous driving vehicle through differential game theory, including solving Nash equilibrium and Stackelberg equilibrium, combining the benefits of different driving path decisions of the autonomous driving vehicle, analyzing and determining the optimal driving path strategy of each vehicle at the intersection; Step S4: defining a utility function for each autonomous vehicle, which combines the revenue of the autonomous vehicle at the end of the trip and the operating cost during the entire trip to evaluate the benefits of different driving path decisions; Step S5: Define the operating cost function of each autonomous driving vehicle. The operating cost function is combined with the collision risk potential field value of the autonomous driving vehicle to evaluate driving safety and passing efficiency, thereby ensuring driving safety while improving traffic efficiency and completing the optimal driving path strategy selection for the autonomous driving vehicle at the intersection.

2. The method for resolving conflicts of autonomous driving vehicles based on differential game decision modeling according to claim 1, characterized in that: The Gaussian potential field model constructed in step S2 is as follows: In the formula, Γ i is the collision risk potential field value of autonomous driving vehicle i, Λ i is the amplitude coefficient of the potential field, a i is the acceleration of the autonomous driving vehicle i, x is the current position vector of the vehicle, and x ci is the steering center position vector of the autonomous driving vehicle i, ∑ i is the covariance matrix describing the position uncertainty of autonomous vehicle i, σ x represents the position uncertainty in the x direction, σ y represents the position uncertainty in the y direction, and the position uncertainty is expressed in the form of covariance; v i is the velocity vector of autonomous vehicle i, σ v is the standard deviation of velocity, is a fixed value, κ i is the driving aggressiveness level of the autonomous vehicle i.

3. The method for resolving conflicts of autonomous driving vehicles based on differential game decision modeling according to claim 1, characterized in that: The specific steps of the process of making a driving path decision for the autonomous driving vehicle in step S3 are as follows: Step S3.1: Establish the dynamic system model as follows: f i1 (x i ,u i )=v i (t) f i4 (x i ,u i )=f(v i (t),δ i (t),μ,…) … g ij1 (x i ,x j )=ψ(∥∥p i -p j ∥∥) g ij2 (v i ,v j )=φ(v i -v j ) g ij3 (d i ,d j )=x(δ i ,d j ) … In the formula, is the rate of change of the state variables of the autonomous driving vehicle i over time, F i is a function of the dynamics of the autonomous driving vehicle i itself, G ij is the interaction dynamics between the autonomous driving vehicle i and other vehicles j, f ik is the kth element function of the dynamics of the autonomous vehicle i, k∈{1,2,…,m}, including the dynamics of the position f i1 , dynamics of speed i2 , dynamics of the steering angle i3 , dynamic f of friction i4 , v i (t) is the speed of vehicle i at time t, a i (t) is the acceleration of vehicle i at time t, F i (t) is the force applied by the accelerator to vehicle i at time t, m is the mass of the vehicle, is the steering angular velocity of vehicle i at time t, μ is the friction coefficient between the tire and the ground, g ijl is the lth element function of the interaction dynamics between the autonomous vehicle i and other vehicles j, l∈{1,2,…,N}, including the relative position effect g ij1 , speed matching affects g ij2 , steering effect g ij3 , ψ is a function of distance, p i and p j is the position vector of vehicle i and vehicle j, φ is a function of the speed difference, v i and v j is the velocity vector of vehicle i and vehicle j, χ is the function of the steering angle, δ i and δ j is the steering angle of vehicle i and vehicle j, x i and x j are the state vectors of autonomous driving vehicle i and other vehicles j, u i and u j are the control input vectors of autonomous driving vehicle i and other vehicles j, u i1 and u j1 is the acceleration control of the autonomous driving vehicle i and other vehicles j, u i2 and u j2 is the steering control of the autonomous driving vehicle i and other vehicles j, u iN and u jN are other control parameters of the autonomous driving vehicle i and other vehicles j; Step S3.2: Solve the Nash equilibrium as follows: Constraints: Given, and x i ,u i Satisfy system dynamics and constraints; In the formula, is the optimal control strategy for autonomous vehicle i, U i is the utility function of autonomous vehicle i, x i is the state vector of autonomous vehicle i, u i is the control input vector of the autonomous vehicle i, F i is a function of the dynamics of the autonomous vehicle i itself, G ij represents the interaction dynamics between autonomous vehicle i and other vehicles j, j≠i represents all other vehicles; Step S3.3: Solve the Stackelberg equilibrium as follows: Constraints: x i ,u i For leaders, x j ,u j For followers, satisfying system dynamics and constraints; In the formula, is the optimal control strategy of leader i, U i is the utility function of leader i, is the optimal control strategy of follower j, U j is the utility function of follower j, x j is the state vector of follower j.

4. The method for resolving conflicts of autonomous driving vehicles based on differential game decision modeling according to claim 3, characterized in that: The utility function of each autonomous driving vehicle defined in step S4 is as follows: C i , utility (x i (t),u i (t))=D i x i (t)+E i u i (t),w(t)=exp(-λt 2 ) Where U i (x i ,u i ) is the utility function of autonomous driving vehicle i, Φ i (x Ti ) indicates that at the end of the trip x Ti The profit function when C i , utility represents the operating cost function of autonomous driving vehicle i, w(t) is the time decay function, k i is the driving aggressiveness level of the autonomous vehicle i, Q i , utility is the weight matrix for adjusting the impact of the state on utility, R i is the weight matrix of the control strategy’s impact on utility, D i and E i is the coefficient matrix in cost calculation, and λ is the time decay coefficient.

5. The method for resolving conflicts of autonomous driving vehicles based on differential game decision modeling according to claim 1, characterized in that: The operating cost function of the autonomous driving vehicle defined in step S5 is as follows: C i , operational =the s S i (x i ,u i )+η e E i (x i ,u i ) In the formula, C i , operational is the operating cost function of the autonomous driving vehicle i, η s is the safety weight coefficient, η e is the efficiency weight function, S i is the safety cost function of autonomous driving vehicle i, E i is the efficiency cost function of autonomous driving vehicle i, Γ i is the collision risk potential field value of autonomous driving vehicle i, x i (t) is the state vector of the autonomous driving vehicle i at time t, u i (t) is the control input vector of the autonomous driving vehicle i at time t, x goal is the target state of the autonomous driving vehicle i, u opt is the optimal control strategy of autonomous driving vehicle i, P i is the weight matrix of the quantized state, Q i，operational is the weighting matrix of the control strategy deviation.

Images