Coordinated traffic control method for intersection without lane constraint based on euler spiral

Abstract

The application relates to the field of automatic driving, in particular to a cooperative passing control method for a lane-constraint-free intersection based on an Euler spiral, which comprises the following steps: adding intelligent network-connected vehicles into an ordered registration set; extracting the intelligent network-connected vehicles from the ordered registration set, and establishing an optimal lane-constraint-free trajectory from an intersection import stop line to a target export lane; acquiring intelligent network-connected vehicle sensing data, and generating a state-action-reward tuple; inputting the state-action-reward tuple into a decision neural network and an evaluation neural network, and updating network parameters through double-network cooperative training; iteratively generating optimal control actions of each sub-trajectory segment, and combining the optimal control actions to form a complete passing trajectory of the intelligent network-connected vehicle. The above technical scheme can make the intelligent network-connected vehicles interact and share motion states, and cooperatively generate a smooth collision-free passing trajectory, thereby improving the driving flexibility of the intelligent network-connected vehicles in the intersection, and breaking through the limitations of the traditional passing mode based on fixed lane line constraints.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computers, traffic big data and automatic driving trajectory planning, and particularly relates to a lane-constraint-free intersection collaborative traffic control method based on Euler spiral. BACKGROUND

[0002] As a confluence and distribution node of multi-directional traffic flow, intersection is the core hub of urban road network. The traditional intersection management faces double challenges: on the one hand, the phase cycle length and phase switching sequence of traditional traffic lights cannot adapt to traffic flow fluctuations in real time; on the other hand, the reaction delay of human driving behavior exacerbates the contradiction between traffic efficiency and safety. With the development of intelligent networked vehicles, vehicles can realize 10-millisecond-level information interaction through V2X communication, which provides a technical basis for breaking through the traditional traffic mode.

[0003] At present, many researchers propose lane-constraint-free intersection intelligent networked vehicle collaborative traffic methods based on fixed lane lines, including adaptive traffic lights, traffic right allocation strategies based on space-time conflict points, etc. However, these methods cannot break through the fixed lane lines and can only solve the collision avoidance problem by slowing down or even stopping, resulting in increased energy consumption and reduced traffic efficiency. Therefore, there is an urgent need for a technical solution that supports multi-degree-of-freedom trajectory planning without lane constraints and can realize smooth turning control with continuous curvature, to solve the problems existing in the prior art. SUMMARY

[0004] The purpose of the present application is to provide a lane-constraint-free intersection collaborative traffic control method based on Euler spiral, which can generate lane-constraint-free trajectories for intelligent networked vehicles, enabling them to turn flexibly and avoid collisions within the intersection, thereby reducing energy consumption and improving intersection safety.

[0005] To solve the above technical problems, the present application adopts the following technical solutions:

[0006] 1. A lane-constraint-free intersection collaborative traffic control method based on Euler spiral, comprising the following steps:

[0007] R1: When an intelligent networked vehicle reaches a preset registration distance d from the geometric center point O of the intersection, reg add the intelligent networked vehicle to the ordered registration set L;

[0008] R2: Extract the intelligent networked vehicle i in L obtained in step R1 in order, and establish an optimal lane-constraint-free trajectory from the intersection entrance stop line to the target exit lane, the optimal lane-constraint-free trajectory consisting of Euler spiral;

[0009] R3: Acquire the perception data of the intelligent connected vehicle i, construct the state space S containing vehicle status and environmental information, define the action space A containing trajectory control parameters, construct the reward function space R containing trajectory quality indicators, and generate the state-action-reward tuple e.

[0010] R4: Input the tuple e into the decision neural network and the evaluation neural network. The former takes the state as input and outputs the desired action to control the Euler spiral parameters, while the latter evaluates the value of the current state after executing the desired action. The network parameters are updated through collaborative training of the two networks.

[0011] R5 iteratively generates optimized control actions for each sub-trajectory segment, which are then combined to form the complete travel trajectory of the intelligent connected vehicle i.

[0012] Specifically, step R2 is as follows:

[0013] R2-1 extracts the intelligent connected vehicle i from L sequentially and obtains the coordinates of the current lane stop line. Target exit lane coordinates

[0014] R2-2 constructs an optimizable laneless trajectory, which is formed by connecting n Euler spiral sub-trajectories end to end. The expression for the j-th curve segment is: in Let represent the initial point of intelligent connected vehicle i in the j-th sub-trajectory. Let represent the initial steering angle of intelligent connected vehicle i at the starting point of the j-th sub-trajectory, and s represent the arc length from the starting point to the current coordinate. This represents the linear curvature variation of the j-th sub-trajectory segment. Represents the curvature at the initial point. The parameter representing the rate of change of curvature. This represents the x and y coordinates of intelligent connected vehicle i at the end of the j-th sub-trajectory (which is also the starting point of the (j+1)-th sub-trajectory). Let j be the total arc length of the j-th sub-trajectory;

[0015] Specifically, step R3 is as follows:

[0016] When the R3-1 intelligent connected vehicle i reaches the starting point of the j-th sub-trajectory segment, it obtains the initial point of the current sub-trajectory segment. curvature at the initial point Intelligent connected vehicle i current speed The state is formed by the endpoint coordinates T of each sub-trajectory segment of all intelligent connected vehicles preceding vehicle i in the ordered registration set.

[0017] R3-2 Intelligent Connected Vehicle iExecution Action The arc length extended from the starting point is... the end point of the jth sub-trajectory segment (the start point of the (j+1)th sub-trajectory) and linearly changes the speed according to constant acceleration wherein represents the speed change amount when reaching the end point of the jth sub-trajectory segment;

[0018] R3-3 when the intelligent connected vehicle i reaches the start point of the (j+1)th sub-trajectory segment, acquires the start point coordinate of the current sub-trajectory segment and the curvature at the initial point the current speed of the intelligent connected vehicle i the end point coordinates T of all the sub-trajectory segments of all the intelligent connected vehicles before i in the ordered registration set, to form the state

[0019] R3-4 calculates the reward of the intelligent connected vehicle i from the start point to the end point of the jth sub-trajectory segment wherein represents the length of the jth sub-trajectory segment and the target lane, represents the sailing time of the intelligent connected vehicle i in the jth sub-trajectory segment, the collision penalty of the intelligent connected vehicle i with other vehicles in the jth sub-trajectory segment, [a, b, h] are weight parameters;

[0020] R3-5 acquires the sampling data through steps R3-1 to R3-4 and inputs them into step R4.

[0021] wherein, step R4 is specifically:

[0022] R4-1 inputs into the decision neural network, and outputs the decision action and inputs into the evaluation neural network, and outputs the value of executing the action in the state wherein the decision neural network and the evaluation neural network are both composed of several hidden layers and an output layer, each hidden layer contains multiple neurons, and the output of each neuron can be represented as n represents the number of neurons, [w1, w2,..., w n ] represents the weight parameter, [b1, b2,..., b n ] represents the bias term, [x1, x2,..., x n ] represents the output of the neurons of the previous hidden layer, f(·) represents the activation function, the output layer of the decision neural network is composed of , and the output layer of the evaluation neural network is composed of ReLU(z) = max(0, z);

[0023] R4-2 inputs ​Input into the evaluation neural network, output in state Performing action The value

[0024] R4-3 calculates the loss value And update the evaluation neural network weight parameters by gradient descent method, wherein g(·) represents the mean square error function, and γ represents the discount rate;

[0025] R4-4 inputs Into the decision neural network, output decision action Again input Into the evaluation neural network, output in state Performing action The value

[0026] R4-5 calculates the decision loss value Update the weight parameters of the decision neural network by the gradient descent method.

[0027] The technical scheme provided in the above technical scheme has the following beneficial effects compared with the prior art:

[0028] 1、The present application constructs a plurality of optimal sub-trajectories by adopting Euler spiral, and the curvature thereof changes continuously according to a linear rule, so that the intelligent networked vehicle can move smoothly in the intersection, and the problem of vehicle steering jitter caused by sudden change of curvature can be avoided, and the passenger comfort is significantly improved. At the same time, the multi-segment optimal sub-trajectory splicing breaks through the fixed lane mode of the traditional method, allowing the intelligent networked vehicle to plan a path flexibly in the intersection and realize collision avoidance between vehicles by accelerating and steering, fully utilizing the intersection space resources and improving the energy efficiency and driving safety.

[0029] 2、Based on the cooperative training mechanism of the decision network and the evaluation network, the vehicle sub-trajectory parameters And the dynamic optimization of speed change The decision network adopts tanh activation function to constrain the action output range, and the evaluation network uses ReLU to process nonlinear value evaluation, and the gradient of the double network is updated cooperatively to ensure the rapid convergence of the strategy. At the same time, through the training sample input mechanism of state-action-reward tuples, the intelligent networked vehicle can adapt to random traffic flow and interference, and realize the multi-objective optimization of traffic safety and traffic efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0030] Fig. 1 The flowchart of the intersection cooperative traffic control method based on Euler spiral in the embodiment of the present application is shown in the figure;

[0031] Fig. 2 An example of an optimal lane-free trajectory for reference of the present application;

[0032] Fig. 3 A diagram of the collaborative training mechanism of the decision network and the evaluation network of the present application; DETAILED DESCRIPTION

[0033] In order to make the purpose and advantages of the present application more clear and explicit, the present application is specifically described below in combination with embodiments. It should be understood that the following text is only used to describe one or several specific embodiments of the present application, and does not strictly limit the specific protection scope requested by the present application.

[0034] Next, the present application is analyzed and described in combination with the accompanying Figs. 1-3 , the present application is further analyzed and described.

[0035] As Fig. 1 shown, one embodiment of a lane-free intersection collaborative traffic control method based on Euler spiral in the embodiment of the present application has the following specific steps:

[0036] Step 1, when the intelligent connected vehicle reaches a preset registration distance d reg from the geometric center point O of the intersection, the intelligent connected vehicle is added to the ordered registration set L;

[0037] Step 2, the intelligent connected vehicle i in L obtained in step 1 is extracted in order, and an optimal lane-free trajectory from the intersection entrance stop line to the target exit lane is established, the optimal lane-free trajectory is composed of Euler spirals;

[0038] 2.1 The intelligent connected vehicle i in L is extracted in order, and the current lane stop line coordinates target exit lane coordinates

[0039] 2.2 An optimal lane-free trajectory is constructed, which is composed of n segments of Euler spiral sub-trajectories, as Fig. 2 shown, the jth curve expression is: wherein represents the initial point of the intelligent connected vehicle i in the jth sub-trajectory, represents the initial turning angle of the intelligent connected vehicle i at the starting point of the jth sub-trajectory, s represents the arc length from the starting point to the current coordinates, represents the curvature linear variation law of the jth sub-trajectory segment, represents the initial point curvature, represents the curvature change rate parameter, represents the horizontal and vertical coordinates of the intelligent connected vehicle i at the terminal point of the jth sub-trajectory (which is also the starting point of the j+1th sub-trajectory), is the total arc length of the jth sub-trajectory;

[0040] Step 3, obtaining intelligent connected vehicle i perception data, constructing state space S containing vehicle state and environment information, defining action space A containing trajectory control parameters, constructing reward function space R containing trajectory quality indicators, and generating state-action-reward tuple e;

[0041] 3.1 When the intelligent connected vehicle i drives to the starting point of the jth sub-trajectory segment, obtain the initial point of the current sub-trajectory segment and the initial point curvature The current speed of the intelligent connected vehicle i The endpoint coordinates T of each sub-trajectory segment of all intelligent connected vehicles before i in the ordered registration set form the state

[0042] 3.2 Intelligent connected vehicle i performs action Extend the Euler spiral line from the starting point with an arc length of to the endpoint of the jth sub-trajectory segment (the starting point of the j+1th sub-trajectory), and linearly change the speed according to constant acceleration Where represents the speed change when reaching the endpoint of the jth sub-trajectory segment;

[0043] 3.3 When the intelligent connected vehicle i drives to the starting point of the j+1th sub-trajectory segment, obtain the starting point of the current sub-trajectory segment and the initial point curvature The current speed of the intelligent connected vehicle i The endpoint coordinates T of each sub-trajectory segment of all intelligent connected vehicles before i in the ordered registration set form the state

[0044] 3.4 Calculate the reward of the intelligent connected vehicle i from the starting point to the endpoint of the jth sub-trajectory segment Where represents the length of the jth sub-trajectory segment endpoint and the target lane, represents the navigation time of the intelligent connected vehicle i in the jth sub-trajectory segment, The collision penalty of the intelligent connected vehicle i with other vehicles in the jth sub-trajectory segment, [α, β, η] are weight parameters;

[0045] 3.5 Obtain sampling data through steps 3.1 to 3.4 and input to step 4.

[0046] Step 4, input tuple e into decision neural network and evaluation neural network, the former inputs state and outputs expected action to control Euler spiral line parameters, the latter evaluates the value of expected action after current state, update network parameters through double network cooperative training, as shown in Fig. 3 .

[0047] 4.1 will The input is fed into a decision neural network, which then outputs a decision action. Then The input is fed into the evaluation neural network, and the output is in the state. Next action value Both the decision neural network and the evaluation neural network consist of several hidden layers and one output layer. Each hidden layer contains multiple neurons, and the output of each neuron can be represented as... n represents the number of neurons, [w1, w2, ..., w n [b1, b2, ..., b] represents the weight parameters. n ] represents the bias term, [x1, x2, ..., x n ] represents the output of the neuron in the previous hidden layer, f(·) represents the activation function, and the output layer of the decision neural network is composed of The output layer of the evaluation neural network is composed of ReLU(z) = max(0,z);

[0048] 4.2 will The input is fed into the evaluation neural network, and the output is in the state. Next action value

[0049] 4.3 Calculate the loss value The evaluation neural network weight parameters are updated using the gradient descent method, where g(·) represents the mean squared error function and γ represents the discount rate.

[0050] 4.4 will The input is fed into a decision neural network, which then outputs a decision action. Then The input is fed into the evaluation neural network, and the output is in the state. Next action value

[0051] 4.5 Calculate the decision loss value The weight parameters of the decision neural network are updated using the gradient descent method.

[0052] Step 5: Iteratively generate optimized control actions for each sub-trajectory segment, and combine them to form the complete travel trajectory of the intelligent connected vehicle i.

[0053] The embodiments of the present application are described in detail above with reference to the accompanying drawings, but the present application is not limited to the above-described embodiments, and for those skilled in the art, after learning the contents described in the present application, several equivalent transformations and substitutions can be made without departing from the principles of the present application, and these equivalent transformations and substitutions should also be considered as belonging to the protection scope of the present application.

Claims

1. A lane-constrained intersection cooperative traffic control method based on Euler spirals, characterized in that, Includes the following steps: R1 When the intelligent connected vehicle reaches the preset registration distance from the geometric center point O of the intersection. At that time, intelligent connected vehicles are added to the ordered registration set L; R2 sequentially extracts the intelligent connected vehicle i from L obtained in step R1, and establishes an optimizable lane-free trajectory from the intersection entrance stop line to the target exit lane. The optimizable lane-free trajectory is composed of several Euler spiral line sub-trajectory segments connected end to end. Step R2 is as follows: R2-1 Extracts the intelligent connected vehicle i from L sequentially and obtains the coordinates of the current lane stop line. Target exit lane coordinates ; R2-2 constructs an optimizable laneless trajectory, which is formed by connecting n Euler spiral sub-trajectories end to end. The expression for the j-th curve segment is: ,in Let represent the initial point of intelligent connected vehicle i in the j-th sub-trajectory. Let represent the initial steering angle of intelligent connected vehicle i at the starting point of the j-th sub-trajectory. This represents the arc length from the starting point to the current coordinate. The curvature linear variation law representing the j-th sub-trajectory segment. Represents the curvature at the initial point. The parameter representing the rate of change of curvature. This represents the x and y coordinates of the intelligent connected vehicle i at the endpoint of the j-th sub-trajectory. Let j be the total arc length of the j-th sub-trajectory; R3 acquires the perception data of the intelligent connected vehicle i, constructs a state space S containing vehicle status and environmental information, constructs an action space A containing trajectory control parameters, constructs a reward function space R containing trajectory quality indicators, and generates a state-action-reward tuple e. Step R3 is as follows: When intelligent connected vehicle i reaches the starting point of the j-th sub-trajectory segment, it obtains the initial point of the current sub-trajectory segment. curvature at the initial point Intelligent connected vehicle i current speed The state is composed of the endpoint coordinates T of each sub-trajectory segment of all intelligent connected vehicles preceding i in the ordered registration set. ; R3-2 Intelligent Connected Vehicle i-Execution Action The arc length extended from the starting point is... The Euler spiral reaches the end of the j-th sub-trajectory segment and changes velocity linearly with constant acceleration. ,in This represents the change in velocity when reaching the end of the j-th sub-trajectory segment; When intelligent connected vehicle i reaches the starting point of the (j+1)th sub-trajectory segment, it obtains the starting point of the current sub-trajectory segment. curvature at the initial point Intelligent connected vehicle i current speed The state is composed of the endpoint coordinates T of each sub-trajectory segment of all intelligent connected vehicles preceding i in the ordered registration set. ; R3-4 Calculate the reward for intelligent connected vehicle i to reach the destination from the starting point of the j-th sub-trajectory segment. ,in Represents the length between the end point of the j-th sub-trajectory segment and the target lane. This represents the travel time of intelligent connected vehicle i in the j-th sub-trajectory segment. The intelligent connected vehicle i is penalized for colliding with other vehicles within sub-trajectory segment j. These are weight parameters; R3-5 Obtain sampling data through steps R3-1 to R3-4. And input it into step R4; R4 inputs tuple e into a decision neural network and an evaluation neural network. The former takes the state as input and outputs the desired action to control the Euler spiral parameters, while the latter evaluates the value of the desired action after the current state is executed. The network parameters are updated through collaborative training of the two networks. R5 iteratively generates optimized control actions for each sub-trajectory segment, which are then combined to form the complete travel trajectory of the intelligent connected vehicle i.

2. The lane-constrained intersection cooperative traffic control method based on Euler spirals according to claim 1, characterized in that, Step R4 is as follows: R4-1 will The input is fed into a decision neural network, which then outputs a decision action. Then The input is fed into the evaluation neural network, and the output is in the state. Next action value Both the decision neural network and the evaluation neural network consist of several hidden layers and one output layer. Each hidden layer contains multiple neurons, and the output of each neuron can be represented as... , where n represents the number of neurons. Represents the weight parameter. Represents the bias term. This represents the output of the neurons in the previous hidden layer. Representing the activation function, the output layer of the decision neural network consists of... The output layer of the evaluation neural network is composed of... constitute; R4-2 will The input is fed into the evaluation neural network, and the output is in the state. Next action value ; R4-3 Calculate the loss value The weight parameters of the evaluation neural network are updated using the gradient descent method. Represents the mean squared error function. Represents the discount rate; R4-4 will The input is fed into a decision neural network, which then outputs a decision action. Then The input is fed into the evaluation neural network, and the output is in the state. Next action value ; R4-5 Calculate the decision loss value The weight parameters of the decision neural network are updated using the gradient descent method.

Images