Vehicle passing planning method and system for non-signal intersection

Abstract

The invention provides a vehicle passing planning method and system for a no-signal intersection, and relates to the field of traffic route planning, the method is based on a partially observable hidden Markov chain decision-making model, and modeling is carried out on an intersection vehicle passing order and an advancing effect thereof; an observation matrix, a state matrix, a state transition probability model and a target optimal (average shortest vehicle passing time) calculation expression of vehicle passing are constructed, the motion behavior of a single individual passing vehicle at an intersection is modeled, and the interpretability of a planning decision is improved; in addition, an optimal group target motion time sequence chain is searched based on a cost (passing time) calculation method of a Monte Carlo tree model, so that an optimal vehicle real-time passing plan is obtained, and the uncertainty of time consumption of intersection vehicle passing plan calculation is reduced within predictable algorithm time complexity.

Description

Technical Field

[0001] The present invention belongs to the technical field of traffic route planning, and particularly relates to a vehicle passing planning method and system for a signal-free intersection. Background Art

[0002] Currently, the connected vehicle autonomous driving system will utilize the roadside perception information system (such as edge computing perception devices), roadside guidance and control information (such as traffic lights), and in-vehicle devices (such as OBU) to obtain the global traffic information of road traffic in real time. With the continuous improvement and development of the connected autonomous driving infrastructure, the information of road traffic is becoming more and more complete, and the roadside guidance and control information has gradually become a limiting factor for road traffic efficiency.

[0003] The passing of vehicles at a signal-free intersection based on roadside control means that the roadside perception devices are used to perceive or the surrounding road traffic information is obtained through V2X communication, and the passing of vehicles is decision-planned according to the global traffic information. The roadside can perform lane-level decision planning on the vehicles in the same lane according to the lane level, or perform vehicle-level decision planning for the vehicles. The autonomous driving vehicles pass safely according to the decision planning information of the Road Side Sub-system (RSS) and improve the passing efficiency. Obviously, for vehicle-level decision planning, the granularity of the target research object is finer, and it is more likely to output a passing plan for the intersection with higher efficiency and higher safety. Therefore, the vehicle passing planning at a signal-free intersection is a new challenge for the path planning technology in the connected vehicle autonomous driving system.

[0004] In the prior art, a vehicle control technology for a signal-free intersection based on deep reinforcement learning is provided. The distributed reinforcement learning algorithm is used to train the passing instance data of the vehicles at the intersection, and a vehicle passing strategy feature model is obtained. A new reward function is proposed, and then in the scenario of any proportion of autonomous driving vehicles, a passing order plan with high efficiency is output for the vehicle passing. However, this solution lacks the mathematical logic modeling of the motion behavior of a single vehicle. Therefore, in terms of strategy optimization, the only solution that can be adopted is usually to increase the instance data. However, the increase in instance data will lead to a large computational amount of the overall model and is extremely time-consuming. Summary of the Invention

[0005] To solve the above technical problems, the present invention proposes a vehicle passing planning method and system for a signal-free intersection.

[0006] The first aspect of the present invention discloses a vehicle passing planning method for a signal-free intersection; the method includes:

[0007] Step S1: Based on the road topological attributes and the geometric features of the lane-level map at intersections, model the vehicle trajectories at road intersections and extract key trajectory feature points. Among them, the road topological attributes include going straight, turning left, turning right, and doing a U-turn.

[0008] Step S2: Based on the key trajectory feature points and road topological attributes, construct a vehicle motion behavior model using a partially observable hidden Markov chain model.

[0009] Among them, the vehicle motion behavior model is used to characterize the traffic order and travel effect of vehicles at intersections.

[0010] The vehicle motion behavior model includes an observation matrix, a state matrix, a state transition probability model, a feedback model, and an objective optimal calculation expression.

[0011] Step S3: Solve the vehicle motion behavior model based on the Monte Carlo tree model, search for the optimal group target motion time sequence chain, and thus obtain the optimal real-time vehicle traffic plan.

[0012] According to the method of the first aspect of the present invention, in the step S1, the calculation method of the key feature points of the right-turn lane is as follows:

[0013] Connect the midpoints of the right-turn lane and the target lane of the right-turn traffic flow to obtain LC-2_LC-4, and the point on the edge of the central intersection area between the right-turn lane and the target lane of the right-turn traffic flow with the largest projection distance from the line segment LC-2_LC-4 is taken as Q'.

[0014] Extend half of the lane width from point Q' to the center point O of the intersection to obtain point Q. The three points LC-2, Q, and LC-4 are used to construct an arc, and interpolation is performed from it to obtain the key feature points of the right-turn lane.

[0015] According to the method of the first aspect of the present invention, in the step S1, the calculation method of the key feature points of the straight-through lane is as follows:

[0016] Connect the midpoints of the straight-through lane and the target lane of the straight-through traffic flow to obtain LC-1_LC-5, and interpolation is performed from it to obtain the key feature points of the straight-through lane.

[0017] According to the method of the first aspect of the present invention, in the step S1, the calculation method of the key feature points of the left-turn lane is as follows:

[0018] Connect the midpoints of the left-turn lane and the target lane of the left-turn traffic flow to obtain LC-6_LC-7;

[0019] Project from the center point O of the intersection to LC-6_LC-7 to obtain the perpendicular point P. The point at one-third of the connection line between point O and point P close to point O is denoted as point C;

[0020] Construct an arc from the three points LC-6, C, and LC-7, and interpolate from it to obtain the key feature points of the right-turn lane.

[0021] According to the method of the first aspect of the present invention, in the step S1, the calculation method of the key feature points of the U-turn lane is as follows:

[0022] Connect the midpoints of the U-turn lane and the target lane of the U-turn traffic flow, that is, LC-6_LC-3. Take it as the axis of the ellipse and construct a semi-ellipse. The other semi-major axis takes 2 times the average vehicle length. Obtain the arc of the semi-ellipse and interpolate from it to obtain the key feature points of the U-turn lane.

[0023] According to the method of the first aspect of the present invention, in the step S2, the vehicle motion behavior model is expressed as: (A, S, O, T, R), where A, S, and O respectively represent the action space, state space, and observation space; T represents the state transition probability model, and R represents the feedback model.

[0024] According to the method of the first aspect of the present invention, the expression of the action space a is:

[0025] a=(a v , a l )

[0026] Among them, a v represents the acceleration; a l represents the angular velocity;

[0027] The expression of the state space X t is:

[0028] X t =(X 0,t , X 1,t , X 2,t , …… X n,t ), where X i,t =(S i , V i,t , P i,t )

[0029] Among them, X i,t is the state of the i-th vehicle at time t, i is a natural number from 0 to n, S i is the distance from the previous moment, V i,t is the speed, and P i,t is the selected path;

[0030] The expression of the observation space:

[0031] O t =(O 1,t , O 2,t , …… O n,t), where O i,t =(x i,t , y i,t , v i,t )

[0032] where O i,t is the position and speed of the i-th vehicle at time t, x i,t , y i,t are the x and y coordinates of the i-th vehicle at time t, and v i,t is the speed of the i-th vehicle at time t;

[0033] The expression of the state transition probability model is:

[0034] X i,t+1 ={t|X i,t}

[0035] where S i,t+1 =S i,t +V i,t *t S i,t+1 , S i,t The positions all fall on lane P i,t ;

[0036] The expression of the feedback model is:

[0037] R = R c +R g +R v +R m +R r

[0038] where That is, if the distance between vehicles is less than the safe distance, a penalty will be imposed;

[0039] indicates that if the target is reached, a reward will be given;

[0040] R v =-(V - v ref ) 2 , indicating the deviation between the vehicle speed and the reference speed;

[0041] R g = 300y - 10x, indicating a positive correlation with Y and a negative correlation with x, and the parameters are set according to experience.

[0042] indicates that if it is in reverse, a penalty will be imposed;

[0043] The target optimal calculation is expressed as:

[0044]

[0045] In the second aspect of the present invention, a vehicle passing planning system for a signal-free intersection is disclosed; the system includes:

[0046] A first processing module, configured to model the vehicle trajectories at a road intersection based on the road topological attributes and the geometric features of the lane-level map of the intersection, and extract key trajectory feature points; wherein, the road topological attributes include straight, left turn, right turn, and U-turn.

[0047] A second processing module, configured to construct a vehicle motion behavior model by using a partially observable hidden Markov chain model based on the key trajectory feature points and the road topological attributes.

[0048] Wherein, the vehicle motion behavior model is used to characterize the passing order of vehicles at the intersection and their traveling effects.

[0049] The vehicle motion behavior model includes an observation matrix, a state matrix, a state transition probability model, a feedback model, and an objective optimal calculation expression.

[0050] A third processing module, configured to solve the vehicle motion behavior model based on the Monte Carlo tree model, search for the optimal group target motion time sequence chain, and thus obtain the optimal real-time vehicle passing plan.

[0051] In the third aspect of the present invention, an electronic device is disclosed. The electronic device includes a memory and a processor. When the processor executes a computer program stored in the memory, the steps in any one of the vehicle passing planning methods for a signal-free intersection in the first aspect of the present disclosure are implemented.

[0052] In the fourth aspect of the present invention, a computer-readable storage medium is disclosed. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps in any one of the vehicle passing planning methods for a signal-free intersection in the first aspect of the present disclosure are implemented.

[0053] The solution proposed by the present invention has the following technical effects: Based on the partially observable hidden Markov chain decision model, the passing order of vehicles at the intersection and their traveling effects are modeled, and an observation matrix, a state matrix, a state transition probability model, and an objective optimal (average shortest vehicle passing time) calculation expression for vehicle passing are constructed. It is a modeling of the motion behavior of individual vehicles passing through the intersection, which improves the interpretability of planning decisions; in addition, based on the cost (passing time) calculation method of the Monte Carlo tree model, the optimal group target motion time sequence chain is searched, and thus the optimal real-time vehicle passing plan is obtained, reducing the uncertainty of the calculation time-consuming for vehicle passing planning at the intersection within the foreseeable algorithm time complexity.

[0054] In summary, by constructing a vehicle passing model at intersections and based on the expected navigation directions of vehicles (left turn, straight, right turn, U-turn), vehicle positions, and restricted speeds in the intersection area, the lane-level map matching technology for autonomous driving at intersections in signal-free areas enables real-time, reliable, and highly fault-tolerant lane-level map matching for the perception and positioning results of autonomous driving and vehicle-road cooperation, thus greatly improving the data quality of the fusion positioning results. This provides a data foundation for fully exploring the rich semantic information of high-precision maps and developing and expanding the utility of positioning information. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0056] Figure 1 FIG. is a flowchart of a vehicle passing planning method for a signal-free intersection according to an embodiment of the present invention;

[0057] Figure 2 FIG. is a flowchart of a method for constructing a CTP feature point trajectory according to an embodiment of the present invention;

[0058] Figure 3 FIG. is a flowchart of an algorithm solution for MCTS according to an embodiment of the present invention;

[0059] Figure 4 FIG. is an overall framework diagram of a vehicle passing planning method for a signal-free intersection according to an embodiment of the present invention;

[0060] Figure 5 FIG. is a structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0062] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first image may be referred to as a second image, and similarly, a second image may be referred to as a first image. Both the first image and the second image are images, but they are not the same image.

[0063] With the continuous improvement and development of the connected autonomous driving infrastructure, the information on road traffic is becoming more and more complete, and the roadside guidance and control information has gradually become a restrictive factor for road traffic efficiency. Therefore, the vehicle passing plan at unsignalized intersections is a new challenge for the path planning technology in the connected vehicle autonomous driving system. Therefore, the present invention provides a vehicle passing plan method and system for unsignalized intersections.

[0064] Please refer to Figure 1 , which is a flowchart of a vehicle passing plan method for unsignalized intersections disclosed in the first aspect of the present invention; the method includes:

[0065] Step S1, based on the road topological attributes and the geometric features of the lane-level map at the intersection, model the vehicle trajectories at the road intersection and extract the key trajectory CTP feature points; wherein, the road topological attributes include straight, left turn, right turn, and U-turn.

[0066] In the step S1, before elaborating on the method for constructing the CTP feature point trajectory, first define the element composition and structure of the lane-level map.

[0067] According to the definition of the map data standard in the "Content of High-Level Autonomous Driving Data Interaction Based on Vehicle-Road Collaboration" standard, in this method, the elements and their composition structures in the lane-level map are defined as follows. For details, please refer to Figure 2 .

[0068] Intersection: It includes the road sections flowing towards the center point of the intersection and the central intersection area. In the figure, the intersection includes all areas.

[0069] Road section: The vehicle road flowing towards the intersection area. An intersection has 3 to N road sections, and each road section has 1 to N lanes. In the figure, the road sections are marked with black solid lines.

[0070] Lane: The drivable area with a clear driving direction. In the figure, the rectangular area within two black solid lines marks the lane.

[0071] Central intersection area: The terminal intersection area of all road sections, without a clear fixed vehicle driving trajectory. In the figure, it is marked with a dotted area.

[0072] Generate the CTP feature point trajectory based on the geometric features of the lane-level map at intersections and the road traffic topological attributes, such as going straight, turning left, turning right, U-turn, etc., so that the trajectories of all relevant associated lanes are correlated.

[0073] As Figure 2 shown, the calculation of the key feature points is carried out by constructing the geometric features of the intersection area. First, connect the road green belts in opposite directions and set the intersection point as the center of the intersection.

[0074] The calculation method of the key feature points of the right-turn lane is as follows:

[0075] Connect the midpoints of the right-turn lane and the target lane of the right-turn traffic flow to obtain LC-2_LC-4, and the point on the edge between the right-turn lane and the target lane of the right-turn traffic flow in the central intersection passing area with the maximum projection distance from the line segment LC-2_LC-4 is taken as Q';

[0076] Extend half of the lane width from point Q' to the center point O of the intersection to obtain point Q; Three points LC-2, Q, and LC-4 are used to construct an arc, and interpolation is performed from it to obtain the key feature points of the right-turn lane.

[0077] The calculation method of the key feature points of the straight-through lane is as follows:

[0078] Connect the midpoints of the straight-through lane and the target lane of the straight-through traffic flow to obtain LC-1_LC-5, and interpolation is performed from it to obtain the key feature points of the straight-through lane.

[0079] The calculation method of the key feature points of the left-turn lane is as follows:

[0080] Connect the midpoints of the left-turn lane and the target lane of the left-turn traffic flow to obtain LC-6_LC-7;

[0081] Project from the center point O of the intersection to LC-6_LC-7 to obtain the perpendicular point P. The point at one-third of the connection line between point O and point P close to point O is denoted as point C;

[0082] Three points LC-6, C, and LC-7 are used to construct an arc, and interpolation is performed from it to obtain the key feature points of the right-turn lane.

[0083] The calculation method of the key feature points of the U-turn lane is as follows:

[0084] Connect the midpoints of the U-turn lane and the target lane of the U-turn traffic flow, that is, LC-6_LC-3. Take it as the axis of the ellipse and construct a semi-ellipse. The other semi-major axis takes 2 times the average vehicle length. Obtain the arc of the semi-ellipse and perform interpolation from it to obtain the key feature points of the U-turn lane.

[0085] Step S2: Based on the key trajectory feature points and road topological attributes, a partially observable hidden Markov chain model is used to construct a vehicle motion behavior model;

[0086] Among them, the vehicle motion behavior model is used to characterize the traffic order and driving effect of vehicles at intersections;

[0087] The vehicle motion behavior model includes an observation matrix, a state matrix, a state transition probability model, a feedback model, and an objective optimal calculation expression; this step models the motion behavior of individual vehicles passing through intersections, improving the interpretability of planning decisions.

[0088] The partially observable hidden Markov chain model (POMDP model) is usually a seven-tuple (A, S, O, T, R, PO, σ), where A, S, and O represent the action space, state space, and observation space respectively, T represents the transition model, R represents the feedback model Reward, PO represents the observation model under partially observable conditions, and σ is the attenuation factor. Next, the specific values in the seven-tuple are described respectively:

[0089] A. State Space

[0090] At the state space, explicit states and implicit states are specifically emphasized, but there doesn't seem to be an intuitive manifestation in the following. Anyway, the state space can be defined as:

[0091] X t =(X 0,t , X 1,t , X 2,t , …… X n,t ), where X i,t =(S i , V i,t , P i,t )

[0092] Among them, X i,t is the state of the i-th vehicle at time t, i is a natural number from 0 to n, S i is the distance from the previous moment, V i,t is the speed, and P i,t is the selected path;

[0093] Here, the benefits of the Frenet frame coordinate system are reflected. To describe the state of a vehicle on a reference line, only the distance S and the speed V in the tangent direction are needed. Among them, P represents the selected path, that is, the above-mentioned candidate path; of course, in addition to its own vehicle, the state space also includes oncoming vehicles. The distance S and speed V can be described in the same system, but P represents the intention judgment of turning left, turning right, or going straight, that is, the driving trajectory of oncoming vehicles is predicted based on our side.

[0094] B. Observation Space Model:

[0095] O t =(O 1,t , O 2,t , …… O n,t ), where O i,t =(x i,t , y i,t , v i,t )

[0096] Among them, O i,t is the position and speed of the i-th vehicle at time t, x i,t , y i,t are the x and y coordinates of the i-th vehicle at time t, and v i,t is the speed of the i-th vehicle at time t; the observation can only observe the approximate x, y coordinates and speed of oncoming vehicles.

[0097] C. Action Space

[0098] The action space includes two things: acceleration and sharp turn instruction, which actually correspond to the setting of the candidate path. When going straight, an acceleration with a gradient of 1 m / s 2 , belonging to the acceleration in [-4 m / s 2 , 4 m / s 2 , is used. When entering a curve, when there is a sharp turn instruction, it turns sharply at a fixed angular velocity. Therefore, the sharp turn instruction is a boolean variable.

[0099] a=(a v , a l )

[0100] Among them, a v represents acceleration; a l represents angular velocity;

[0101] D. Reward Model

[0102] The design of the reward model also affects the optimization process and even determines whether part of the calculation process converges and the convergence efficiency. This solution uses a superposition method to cumulatively calculate various different forms of rewards. The main model of the reward mechanism is as follows

[0103] R = R c + R g + R v + R m + R r

[0104] Among them, That is, if the distance between vehicles is less than the safe distance, a penalty will be imposed;

[0105] Indicates that if the target is achieved, a reward will be given;

[0106] R v = -(V - v ref ) 2 , indicating the deviation between the vehicle speed and the reference speed;

[0107] R g = 300y - 10x, indicating a positive correlation with Y and a negative correlation with x, and the parameters are set according to experience;

[0108] Indicates that if it is in reverse, a penalty will be imposed;

[0109] E. State Transition Probability Model Transition Model

[0110] In the simulation calculation, the state is mainly reflected in the vehicle trajectory. In this method, the trajectory is not completely calculated according to kinematic characteristics, but the positions of all vehicles at each simulation moment are output according to the simulation calculation. That is, all trajectory points are pre-interpolated and calculated, and it is ensured that all vehicles are at a certain trajectory point at any simulation moment. The difference is that after one simulation calculation, some vehicles obtain the right of way based on the traffic strategy and reach the next trajectory point.

[0111] X i,t = (S i,t , V i,t , P i,t )

[0112] X i,t+1 = {t|X i,t}

[0113] S i,t+1 = S i,t + V i,t * tS i,t+1 , S i,t The positions all fall on the P i,t lane

[0114] Based on the understanding of the model, the average spacing of trajectory points in the area with obvious lane markings is longer than that in the central intersection area by about 20%.

[0115] Considering the vehicle navigation relationship and the lane connectivity topological association relationship, P i,t and P i,t+1They are the same lane or must be adjacent lanes before and after. Therefore, when constructing the state transition, the computational complexity of the state space is greatly compressed.

[0116] F. Partially Observation Model under Partially Observable Conditions

[0117] The observation model OM under partially observable conditions is the observation function given by the state and the actions taken by the system.

[0118] The observation model describes the perception of all vehicles at the road intersection of the state space. Now assume that the position and speed of the ego vehicle are fully observable. The observation space is similar to the actual state space. In this method, the observation model can be described as follows:

[0119] The types, speeds, positions, lanes where they are located, and driving directions of all vehicles at the intersection can always be obtained;

[0120] For vehicles that have not entered the intersection, their states are unknown and are regarded as unobservable factors and are not included in the calculation at the current moment

[0121] The observed values of the attributes and motion states of all vehicles conform to a normal distribution under the real state. The parameters of the distribution depend on the perception system model.

[0122] G. σ Decay Factor

[0123] σ is called the decay factor, and its value is in [0, 1]. The larger the value, the greater the impact of the later reward on the currently selected action. σ is generally selected as 0.99 or 0.995. In the simulation, the value of σ can be adjusted according to the previous effects.

[0124] Optimal Constraint Conditions:

[0125]

[0126] The maximum constraint condition is that the average speed of all vehicles passing through the intersection is the largest. Under the condition that the lane length of the intersection is unchanged, the passing time is the shortest.

[0127] In this step S2, the vehicle motion behavior model is expressed as: (A, S, O, T, R).

[0128] Step S3, solve the vehicle motion behavior model based on the Monte Carlo tree model, search for the optimal group target motion time sequence chain, and thus obtain the optimal real-time vehicle passing plan.

[0129] Monte Carlo Tree Search is a classic tree search algorithm, which is extremely effective for solving game problems with such a large-scale search space because its core idea is to allocate resources to more promising branches, that is, to concentrate computing power on more valuable (extended, higher-priority branches).

[0130] The algorithm of MCTS is mainly divided into four steps: selection, expansion, simulation, and backpropagation. For details, please refer to Figure 3 .

[0131] STEP 1: Selection

[0132] Starting from the root node, recursively select the optimal child node until a leaf node is reached. The advantage of a node (Upper Confidence Bounds (UCB)) will be evaluated by the following formula.

[0133]

[0134] Vi: The average Value under this node. For example, for a good move, its Value is larger, and for a bad move, it is relatively smaller

[0135] c: A constant, usually 2, which is equivalent to the weight of the two expressions on both sides of the plus sign

[0136] N: The total number of explorations, that is, how many times all nodes have been explored

[0137] ni: The number of explorations of the current node

[0138] In this model, the depth of the child node depends on the total number of vehicles at the intersection. Based on the above UCB formula, the UCB values of all child nodes can be calculated, and the child node with the largest UCB value is selected for iteration.

[0139] STEP 2: Expansion

[0140] If the current leaf node is not a terminal node, then create one or more child nodes and select one of them for expansion. When the aforementioned state transition is mentioned, P i,t and P i,t+1 are the same lane or must be adjacent front and back lanes. When performing expansion calculations, by judging the connectivity of the lanes, the width and depth of the expansion calculations can be greatly compressed, thereby simplifying the calculations and improving the overall efficiency of the algorithm.

[0141] STEP 3: Simulation

[0142] Starting from the expanded node, run a simulated output until the game ends. For example, starting from this expanded node, if the simulation is run ten times and the game is won nine times in the end, then the score of this expanded node will be relatively high; otherwise, it will be relatively low. Here is also the pseudocode for a simulation process:

[0143]

[0144] STEP 4: Backpropagation

[0145] Based on the result value value simulated in the third step, backpropagation is used to update the current action sequence. For all its parent nodes (all the nodes on the black line in the following figure), the exploration times of all of them are incremented by 1, and their values are also accumulated. Then the entire algorithm will repeat many times until the Monte Carlo tree can give the optimal operation strategy in the current state, that is, the vehicle passing order at the road intersection, that is, at the current moment, which lane's vehicle moves forward to the trajectory point of the next state. At this time, the vehicle safety distance and driving speed both meet the aforementioned constraint conditions.

[0146] Please refer to Figure 4 , for the overall framework diagram of the method, including:

[0147] Step 1: Initialization. Load the road map, set the speed limits for roads / lanes, the maximum acceleration and maximum turning angle of the vehicle, the vehicle safety distance (longitudinal), and the traffic flow level at the intersection (used to identify the maximum number of vehicles);

[0148] Step 2: According to the geometric characteristics of the lane-level map at the intersection and the road traffic topological attributes, such as going straight, turning left, turning right, U-turn, etc., generate the CTP feature point trajectories to associate the trajectories of all relevant lanes. Here, define the attributes and concepts of the map.

[0149] Step 3: Randomly generate vehicles newly entering the intersection. The motion state parameters of the vehicles include: vehicle speed, vehicle position, the lane where the vehicle is located, and the expected driving route trajectory (navigation route, left / straight / right driving direction, turn signal, including at least one). For the newly generated vehicles, the constraint conditions to be considered include: a. Do not have a position conflict with the existing vehicles, that is, maintain a safe distance; b. The total number of vehicles does not exceed the traffic flow level; c. On the same road section, only one vehicle is generated and randomly appears in a certain lane of this road section;

[0150] In this step, vehicles newly entering the intersection are randomly generated. The motion state parameters of the vehicles include: vehicle speed, vehicle position, lane where the vehicle is located, and expected driving route trajectory (navigation route, straight or left or right driving direction, turn signal, including at least one of them). For the newly generated vehicles, the constraint conditions to be considered include: a. There should be no position conflict with the existing vehicles, that is, a safe distance should be maintained; b. The total number of vehicles does not exceed the traffic flow level; c. On the same road section, only one vehicle is generated and randomly appears on a certain lane of this road section.

[0151] According to the above principles, a random function is designed to generate vehicles newly entering the intersection. The definition of the random function is as follows:

[0152] Property vehicle (Location,Speed,Road,Lane,Direction,Existance)

[0153] =RandFunction

[0154]

[0155] Step 4: Update the vehicle queue at the intersection, its status, and the overall running time expression;

[0156] Step 5: Based on the CTP feature points and road topology attributes, construct an observable hidden Markov chain model, and construct an observation space, a state transition probability model, and a feedback model;

[0157] Step 6: Based on the MCTS Monte Carlo tree structure, perform a depth search of the tree structure, traverse all traffic orders, and calculate the overall passing time under each order;

[0158] Step 7: Obtain the shortest passing time of the current state, and update the vehicle queue status at the next moment according to the corresponding state chain, including historical trajectory and planned route;

[0159] Step 8: Transfer to Step 3 and recalculate until the total simulation duration is reached, and end the simulation calculation.

[0160] Step 8: According to the historical trajectories of the vehicle queue in the simulation calculation, statistically calculate the overall average passing time, the three values (Max, Min, Mean) of the overall passing time of a single vehicle, and the driving trajectory of the vehicle.

[0161] The second aspect of the present invention discloses a vehicle passing planning system for an unsignalized intersection; the system includes:

[0162] The first processing module is configured to model the vehicle trajectories at a road intersection based on the road topological attributes and the geometric features of the lane-level map of the intersection, and extract key trajectory feature points; wherein, the road topological attributes include straight, left turn, right turn, and U-turn.

[0163] The second processing module is configured to construct a vehicle motion behavior model by using a partially observable hidden Markov chain model based on the key trajectory feature points and the road topological attributes.

[0164] Wherein, the vehicle motion behavior model is used to characterize the traffic order and the traveling effect of vehicles at the intersection.

[0165] The vehicle motion behavior model includes an observation matrix, a state matrix, a state transition probability model, a feedback model, and an objective optimal calculation expression.

[0166] The third processing module is configured to solve the vehicle motion behavior model based on the Monte Carlo tree model, search for the optimal group target motion time sequence chain, and thus obtain the optimal real-time vehicle traffic plan.

[0167] A third aspect of the present invention discloses an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps in a vehicle traffic planning method for a signal-free intersection according to any one of the first aspects of the present disclosure are implemented.

[0168] Figure 5 FIG. is a structural diagram of an electronic device according to an embodiment of the present invention. As Figure 5 shown, the electronic device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Wherein, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the electronic device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be implemented through WIFI, a carrier network, near field communication (NFC), or other technologies. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen. The input device of the electronic device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad provided on the housing of the electronic device, or an external keyboard, touchpad, or mouse, etc.

[0169] Those skilled in the art can understand. Figure 5The structure shown is only a structural diagram of the part related to the technical solution of the present disclosure, and does not constitute a limitation on the electronic device to which the solution of this application is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0170] The fourth aspect of the present invention discloses a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps in a vehicle passing plan method for a signal-free intersection in any one of the first aspects of the present disclosure are implemented.

[0171] The solution proposed by the present invention has the following technical effects: Based on the partially observable hidden Markov chain decision model, the vehicle passing order and its traveling effect at the intersection are modeled, and the observation matrix, state matrix, state transition probability model, and target optimal (average shortest vehicle passing time) calculation expression for vehicle passing are constructed. It is a modeling of the motion behavior of individual vehicles passing through the intersection, which improves the interpretability of the planning decision-making; in addition, based on the cost (passing time) calculation method of the Monte Carlo tree model, the optimal group target motion time series chain is searched, so as to obtain the optimal real-time vehicle passing plan, and within the foreseeable algorithm time complexity, the uncertainty of the calculation time-consuming for vehicle passing planning at the intersection is reduced.

[0172] In summary, by constructing a vehicle passing model for the intersection and according to the expected navigation directions (left turn, straight, right turn, U-turn) of the vehicles, the vehicle positions, and the speed limits in the intersection area, the automatic driving lane-level map matching technology for intersections in signal-free areas is used for the perception and positioning results of automatic driving and vehicle-road cooperation, so as to realize real-time, reliable, and highly fault-tolerant lane-level map matching, thereby greatly improving the data quality of the fusion positioning results. This provides a data basis for fully exploring the rich semantic information of high-precision maps and developing and expanding the utility of positioning information.

[0173] Please note that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, all possible combinations of the technical features in the above embodiments are not described. However, as long as the combinations of these technical features do not conflict, they should be considered as the scope described in this specification. The above embodiments only represent several implementation manners of this application, and their descriptions are relatively specific and detailed, but they should not be construed as a limitation on the scope of the invention patent. It should be pointed out that for those of ordinary skill in the art, without departing from the concept of this application, several deformations and improvements can still be made, and these all belong to the protection scope of this application. Therefore, the protection scope of the patent of this application should be subject to the appended claims.

Claims

1. A vehicle passing planning method for signal - free intersections, characterized in that, the method includes: Step S1, based on the road topological attributes and the geometric features of the lane - level map of the intersection, model the vehicle trajectories at the road intersection and extract key trajectory feature points; wherein, the road topological attributes include straight - ahead, left - turn, right - turn, and U - turn; Step S2, based on the key trajectory feature points and the road topological attributes, construct a vehicle motion behavior model using a partially observable hidden Markov chain model; wherein, the vehicle motion behavior model is used to characterize the vehicle passing order and its traveling effect at the intersection; The vehicle motion behavior model includes an observation matrix, a state matrix, a state transition probability model, a feedback model, and an objective optimal calculation expression; Step S3, based on the Monte Carlo tree model, solve the vehicle motion behavior model, search for the optimal group target motion time - series chain, and thus obtain the optimal real - time vehicle passing plan.

2. The vehicle passing planning method for signal - free intersections according to claim 1, characterized in that, in the step S1, the calculation method of the key feature points of the right - turn lane is as follows: Connect the mid - points of the right - turn lane and the target lane of the right - turn traffic flow to obtain LC - 2_LC - 4, and the point on the edge between the right - turn lane and the target lane of the right - turn traffic flow in the central intersection passing area with the maximum projection distance from the line segment LC - 2_LC - 4 is taken as Q'; From the Q' point, extend half of the lane width towards the center point O of the intersection to obtain the Q point; The three points LC - 2, Q, and LC - 4 are used to construct an arc, and interpolation is performed therefrom to obtain the key feature points of the right - turn lane.

3. The vehicle passing planning method for signal - free intersections according to claim 1, characterized in that, in the step S1, the calculation method of the key feature points of the straight - ahead lane is as follows: Connect the mid - points of the straight - ahead lane and the target lane of the straight - ahead traffic flow to obtain LC - 1_LC - 5, and interpolation is performed therefrom to obtain the key feature points of the straight - ahead lane.

4. The vehicle passing planning method for signal - free intersections according to claim 1, characterized in that, in the step S1, the calculation method of the key feature points of the left - turn lane is as follows: Connect the mid - points of the left - turn lane and the target lane of the left - turn traffic flow to obtain LC - 6_LC - 7; Project from the center point O of the intersection to LC - 6_LC - 7 to obtain the perpendicular point P. The point at one - third of the line segment connecting O and P closer to O is denoted as C; The three points LC - 6, C, and LC - 7 are used to construct an arc, and interpolation is performed therefrom to obtain the key feature points of the right - turn lane.

5. The vehicle passing planning method for signal - free intersections according to claim 1, characterized in that, in the step S1, the calculation method of the key feature points of the U - turn lane is as follows: Connect the mid - points of the U - turn lane and the target lane of the U - turn traffic flow, that is, LC - 6_LC - 3. Take it as the axis of the ellipse to construct a semi - ellipse, and the other semi - major axis takes 2 times the average vehicle length. Obtain the arc of the semi - ellipse, and interpolation is performed therefrom to obtain the key feature points of the U - turn lane.

6. The vehicle passing planning method for signal - free intersections according to claim 1, It is characterized in that In the step S2, the vehicle motion behavior model is expressed as: (A, S, O, T, R), where A, S, and O respectively represent the action space, state space, and observation space; T represents the state transition probability model, and R represents the feedback model.

7. A vehicle passing planning method for an unsignalized intersection according to claim 6, It is characterized in that Action space a The expression is: a = (a v , a l ) Among them, a v represents acceleration; a l represents angular velocity; State space X t The expression of which is: X t =(X 0,t , X 1,t , X 2,t , …… X n,t ), where X i,t =(S i , V i,t , P i,t ) Among them, X i,t is the state of the i-th vehicle at time t, where i is a natural number from 0 to n, S i is the distance from the previous moment, V i,t is the speed, P i,t is the selected path; The expression of the observation space: O t =(O 1,t , O 2,t , …… O n,t ), where O i,t =(x i,t , y i,t , v i,t ) Among them, O i,t is the position and speed of the i-th vehicle at time t, x i,t , y i,t are the x and y coordinates of the i-th vehicle at time t, v i,t is the speed of the i-th vehicle at time t; The expression of the state transition probability model is: X i,t+1 = {t|X i,t} where S i,t+1 = S i,t + V i,t * t S i,t+1 , S i,t The positions all fall on P i,t lane; The expression of the feedback model is: R = R c + R g + R v + R m + R r Among them, that is, if the distance between cars is less than the safe distance, a penalty will be imposed; Indicates that a reward will be given if the goal is achieved; R v = -(V - v ref ) 2 , representing the deviation between the vehicle speed and the reference speed; R g = 300y - 10x, indicating a positive correlation with Y and a negative correlation with x, and the parameters are set according to experience; Indicates that if it is a reverse, there will be a penalty; The target optimal calculation is expressed as:

8. A vehicle passing planning system for an unsignalized intersection, It is characterized in that The system includes: A first processing module, configured to model the vehicle trajectory at the road intersection and extract key trajectory feature points based on the road topological attributes and the geometric features of the lane-level map of the intersection; wherein, the road topological attributes include straight, left turn, right turn, and U-turn. A second processing module, configured to construct a vehicle motion behavior model by using a partially observable hidden Markov chain model based on the key trajectory feature points and the road topological attributes. Wherein, the vehicle motion behavior model is used to characterize the passing order of vehicles at the intersection and their traveling effects. The vehicle motion behavior model includes an observation matrix, a state matrix, a state transition probability model, a feedback model, and a target optimal calculation expression. A third processing module, configured to solve the vehicle motion behavior model based on the Monte Carlo tree model, search for the optimal group target motion time series chain, and thus obtain the optimal real-time vehicle passing plan.

9. An electronic device, It is characterized in that The electronic device includes a memory and a processor. When the processor executes the computer program stored in the memory, the steps in the vehicle passing planning method for an unsignalized intersection according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the steps in the vehicle passing planning method for an unsignalized intersection according to any one of claims 1 to 7 are implemented.

Images