CN115246394A

CN115246394A - Intelligent vehicle overtaking obstacle avoidance method

Info

Publication number: CN115246394A
Application number: CN202110451976.0A
Authority: CN
Inventors: 徐元利; 魏增娜
Original assignee: Tianjin University of Science and Technology
Current assignee: Tianjin University of Science and Technology
Priority date: 2021-04-26
Filing date: 2021-04-26
Publication date: 2022-10-28

Abstract

The invention relates to an intelligent vehicle overtaking obstacle avoidance method, which mainly utilizes advanced sensor technology to enhance the sensing capability of a vehicle to a driving environment, feeds real-time information such as vehicle speed, position and the like acquired by a sensing system back to the system, judges and analyzes potential safety hidden dangers according to comprehensive information of road conditions and traffic flows, calculates the acquired information through a specific algorithm to obtain an optimal obstacle avoidance line, automatically adopts measures such as alarm prompting, braking or steering to assist and control the vehicle to actively avoid obstacles in an emergency situation, and ensures that the vehicle safely, efficiently and stably drives. An intelligent vehicle obstacle avoidance principle adopting an artificial potential field method outputs a stable, comfortable and safe track through an algorithm, and the track is handed to a control module to be executed, so that overtaking obstacle avoidance is finally completed.

Description

Intelligent vehicle overtaking obstacle avoidance method

The technical field is as follows:

the invention relates to the field of overtaking obstacle avoidance processing, and the overtaking obstacle avoidance of a vehicle is completed through a specific algorithm.

Background art:

obstacle avoidance of the intelligent vehicle is mainly operated through path planning, obtained environment information is fed back, and an optimal processing result is calculated. Road planning techniques can be divided into global planning and local planning according to environmental road information and surrounding conditions. In the driving process, on the premise of obtaining a comprehensive road environment, according to preset driving conditions and set standards, a vehicle drives on a safe track on a road, and the requirement of avoiding danger can be met; the driving environment and the movement process have a lot of changes, the speed and the position information of the obstacle during the road driving process are calculated and obtained through unknown local path planning, namely vehicle information fed back by a sensor, and a reasonable path is calculated in a short time through a specific algorithm to avoid the action of the obstacle. Grids, visualizations, and free space are the primary methods of global planning.

The raster method is a map modeling method. Basically, the vehicle environment is divided into units expressed in squares of the same size as a certain resolution, and expressed by a quadtree method. And finally, acquiring the planned path by using a searching method. The selection of the mesh size has a great influence on the transparency of the environment information in the mesh map, and thus it is not easy to optimize the path. The visual method first describes the points of the vehicle and then displays an irregular pattern of obstacle sizes. Then, the start point, the vertex, and the end point of each obstacle are connected so as not to interfere with each other, and the shortest path from the start point to the end point is found. Although the visualization method can constitute the shortest path, the search time is long and there is no flexibility. The free space method constructs a free space by using a predefined basic form, expresses the free space as a graph, and explores a graph structure. In order to solve the flexible visualization method problem, a free space planning method is designed, and a heuristic search algorithm a is used for generating an optimal path from the beginning to the end. However, the difficulty of the algorithm and the number of obstacles are related, which means that the number of obstacles increases drastically. Therefore, it is not suitable for use in an environment with a plurality of obstacles.

Genetic algorithms, neural network algorithms and management methods for local trajectory planning are widely used. The local path planning algorithm based on the genetic algorithm is provided in Li Qing, a biological evolution process is modeled mainly through modes of coding, migration, mutation and the like, and compatible functions are created through coding, mutation, intersection and the like. Because of the consideration of the speed of the obstacle and the risk of collision, a dynamic environment model of the vehicle and the target obstacle is developed and a fuzzy neural network algorithm is introduced to avoid the obstacle.

The invention content is as follows:

the existing routing algorithm has the optimal capability in the aspects of calculating and solving large-scale modeling problems, so that a routing strategy scheme with reliability, flexibility and comprehensiveness is developed. The initial state and the final state of the polynomial may be used to determine the lateral and longitudinal directions of the vehicle according to the state, position and time of the vehicle. In the process of spatial data acquisition, the limitation of vehicle characteristics of an executable area is mainly considered, and the efficiency and the space of the algorithm are improved, so that a better effect is obtained. The front end and the periphery of the vehicle are provided with cameras for recording the surrounding environment of the vehicle and conducting the surrounding environment information. A distance measuring device and a speed sensor are added for detecting the distance between the vehicle and the target obstacle and the speed of the vehicle and the obstacle. And transmitting the obtained information to a processor, and calculating and planning by using the algorithm to obtain an optimal obstacle avoidance route and an optimal operation method. The system comprises a part of the algorithm module, a whole plan prediction module, a routing module and a high-precision map position input result as an output result, the algorithm outputs a smooth, comfortable and safe track, and the track control execution module is used for finishing the final vehicle overtaking and obstacle avoiding functions.

The technical scheme of the invention is as follows:

when an automobile meets an obstacle needing to detour, firstly updating a safety area beside a road in real time according to the state of the obstacle, and then calculating the starting condition of the automobile, namely the current state of the automobile; collecting the final state of the next moment, and incorporating the collected information data of the initial state and the final state of the vehicle into a polynomial for polynomial coupling to form a polynomial track; and finally, selecting a vehicle overtaking track which accords with the dynamic performance of the vehicle, is not influenced by collision risk and ensures the optimal comfort.

The algorithm firstly generates a smooth reference line, namely a road center line, through a polynomial fitting curve based on a starting point and an end point of a GPS, and generates a track in real time for vehicle obstacle avoidance according to the dynamics of a vehicle and the path planning in the driving process of the vehicle, wherein the track comprises local point information, speed and acceleration information.

It is contemplated that the driver may have many ways to deal with the same obstacles. Thus, the first stage of the algorithm consists in taking enough trajectories and providing as many choices as possible; the second stage calculates the position and physical constraint of the barrier in each track; the third step is that the optimal track of the lowest price is displayed in real time in the periodic detection process; and finally, selecting the final track to be output.

Description of the drawings:

FIG. 1 FIG. 2 is a flow chart of a control strategy;

FIG. 3 is a flow chart of the algorithm;

FIG. 4 Frenet coordinate system;

the specific implementation mode is as follows:

1. trajectory sampling

After the camera is installed on the automobile, the camera is used for recording the surrounding environment and inputting information, which is the first step to be carried out by the vehicle for obstacle avoidance. The obstacle avoidance control is actually a planning problem of a motion track to determine the current state of the vehicle, and the speed detection device and the distance measurement device are used for carrying out speed detection and distance detection on the target obstacle and the vehicle and outputting results. The planning of the motion trail is a three-dimensional space optimization problem, and comprises a two-dimensional plane and time. In order to solve the dimension problem, the motion planning is divided into path planning and speed planning, wherein the motion planning is solved on an X-Y plane, the speed planning is solved on an S-T plane, and an S-axis is a path planned in advance.

A coordinate system based on the lateral and longitudinal directions of the lane lines is used. Simulating the driving habits of human beings and generating the track generation problem in the coordinate system, firstly, as a smooth reference line shown in fig. 4, the coordinate points of the vehicle are projected on the reference line to obtain a projected point o on the reference line. The path length from the starting point of the reference line to the projection point is the longitudinal offset of the car in the free coordinate system, denoted by s. The distance from the projection point to the vehicle position is denoted by 1 as the lateral offset of the vehicle in the freset coordinate system. Since the reference line is sufficiently smooth, first and second derivatives of lateral and longitudinal offsets in the free-space coordinate system can be calculated from the direction, velocity and acceleration of the vehicle.

In the figure, s' are respectively expressed as longitudinal distance, longitudinal speed and longitudinal acceleration; l, l' are respectively expressed as a lateral offset, a yaw angle, and a yaw rate.

2. Trajectory generation

It is known that the starting state of the smart car at time t0 is such that, in order to create a trajectory, the final state at time t1 is sampled in a coordinate system, and a fifth order polynomial is used to fit the final and starting states. The quintic polynomial has the advantages of simple structure, easy coefficient solving and capability of being adjusted according to sampling points and first and second derivatives. The quintic curve fitting formula is:

S(t)＝a ₀ +a ₁ t+a ₂ t ² +a ₃ t ³ +a ₄ t ⁴ +a ₅ t ⁵ (1)

the longitudinal trajectory s (t) is obtained by the conditions of the start state and the end state. The transverse trajectory l(s) can be obtained in the same way:

after the transverse and longitudinal tracks are represented by quintic polynomials, the tracks are synthesized into two dimensions. And (4) giving a time T, calculating the longitudinal deviation and the transverse deviation of the time T, and restoring the longitudinal deviation and the transverse deviation into track points in a two-dimensional plane through a datum line. And obtaining a series of track points through a series of time points, and finally forming a complete track.

To obtain multiple trajectories that can select the safest obstacle in the presence of an obstacle, a discrete spatial solution based on the latiz method is used. Because of the dispersion in a dynamic environment, both time and space are considered. The starting point defines the vehicle state vector [ x y θ va ] in the grid static space such that the vehicle satisfies the starting and ending boundary constraints of the path. The spread is made at intervals based on actual road width and vehicle parameters, and the lateral trajectory takes into account obstacle information and constraints on vehicle dynamics. In a longitudinal trajectory, the selection of points cannot be too dense. Being too dense wastes computational resources and becomes unstable when using a fifth order polynomial. Therefore, the selection can be carried out according to the current speed and the road condition, and all parameters can be formed by an algorithm model by combining the characteristics of the vehicle and the road, so that the method has high flexibility. And finally, performing two-dimensional synthesis on all longitudinal tracks and all transverse tracks to obtain a series of planning tracks.

3. Loss function

A loss function is assigned to the planned trajectory, the loss function representing relative implementation possibilities of the intelligent vehicle along the trajectory. The calculation of the loss function can be divided into the calculation of the collision distance, the calculation of the smoothness of the trajectory and the calculation of the derivative of the longitudinal acceleration (jerk). The total loss function of the trajectory is a weighted sum of the three components.

minC _total (f(s))＝C _obstacle (f)+C _smooth (f)+C _jerk (f) (3)

In the formula, l = f(s) represents a functional relationship between the transverse direction and the longitudinal direction.

Based on the boundary distance d between the obstacle and the vehicle, the obstacle loss function is expressed as:

the formula C _mudge Is a monotonically decreasing function, d _c Is a minimum safe distance, and d _n Is to determine a range of values based on the scene, and C _collision At the cost of a collision, a larger value is typically taken to help detect a trajectory that is at risk of collision.

The smoothness function of the trajectory is expressed as:

the formula represents the course difference between the lane and the automobile and the derivative of the respective track of the automobile to the curvature and the curvature, and in the test, the weight value is adjusted according to the current speed to obtain the optimal track.

In this formula, f '(s) represents the difference in heading between the lane and the car, and f '(s) and f '(s) are the derivatives of the respective trajectories of the cars with respect to curvature and curvature, and in the experiment, k is adjusted according to the current speed ₁ ，k ₂ ，k ₃ And (4) obtaining an optimal track by using the weight value.

The path quadratic programming is an improvement of dynamic programming, and aims to obtain a smoother track by satisfying constraint conditions such as collision avoidance and road boundary, and considering optimization technology of comfort and vehicle controllability. And calculating the orbit with the lowest loss function value by quadratic programming to serve as the orbit executed by the intelligent automobile. Finally, an LQR transverse controller is designed, the control variable of the front wheel angle is obtained, and lane change for avoiding obstacles is completed.

In the process, a camera of the vehicle continuously inputs the surrounding vehicle conditions, a speed detector detects the movement speeds of the vehicle and objects around the vehicle, and calculation of an algorithm required by vehicle obstacle avoidance is continuously carried out in the period, so that support required by lane changing and obstacle avoidance is timely provided for the vehicle.

Claims

1. An intelligent vehicle overtaking obstacle avoidance method is characterized by comprising the following steps:

determining a current lane where a vehicle is located and a vehicle speed of the vehicle;

acquiring traffic data of the current lane and adjacent lanes, wherein the traffic data comprises lane information, objects existing on the lanes and relative speeds between the objects and the vehicle;

determining an object in front of the vehicle in the traffic data of the current lane as a first obstacle, and determining the vector speed of the first obstacle according to the traffic data of the current lane and the speed of the vehicle;

determining pre-control information of the vehicle according to the speed of the vehicle;

determining obstacle avoidance information according to the vector speed of the first obstacle, the vehicle speed of the vehicle and traffic data of the adjacent lanes;

and determining an obstacle avoidance strategy of the vehicle based on the pre-control information and the obstacle avoidance information.

2. The method of claim 1, wherein the determining the vector speed of the first obstacle from the traffic data for the current lane and the vehicle speed of the vehicle comprises:

extracting motion information of the first obstacle from traffic data of the current lane, the motion information including a relative speed and a relative direction between the first obstacle and the vehicle;

and determining the vector speed of the first obstacle according to the motion information of the first obstacle and the vehicle speed of the vehicle.

3. The intelligent obstacle avoidance system for automobiles of claim 2, wherein the distance measuring device is a laser distance measuring device or a radar distance measuring device.

4. A vehicle obstacle avoidance system, the system comprising:

the vehicle driving control device comprises a first determination module, a second determination module and a control module, wherein the first determination module is used for determining a current lane where a vehicle is located and the vehicle speed of the vehicle;

the traffic data acquisition module is used for acquiring traffic data of the current lane and the adjacent lanes, wherein the traffic data comprises lane information, objects existing on the lanes and relative speeds between the objects and the vehicle;

the second determination module is used for determining an object positioned in front of the vehicle in the traffic data of the current lane as a first obstacle, and determining the vector speed of the first obstacle according to the traffic data of the current lane and the vehicle speed of the vehicle;

the pre-control information determining module is used for determining the pre-control information of the vehicle according to the speed of the vehicle;

the collision information determining module is used for determining collision information according to the vector speed of the first obstacle, the vehicle speed of the vehicle and the traffic data of the adjacent lanes;

and the obstacle avoidance strategy determining module is used for determining an obstacle avoidance strategy of the vehicle based on the pre-control information and the collision information.