Washington University in St.

Louis
Washington University Open Scholarship

Mechanical Engineering and Materials Science Mechanical Engineering & Materials Science
Independent Study

4-29-2018

Lane Change Strategy for Autonomous Vehicle

Shiyu Xing
Washington University in St. Louis

Mark Jakiela
Washington University in St. Louis

Follow this and additional works at: https://openscholarship.wustl.edu/mems500

Recommended Citation
Xing, Shiyu and Jakiela, Mark, "Lane Change Strategy for Autonomous Vehicle" (2018). Mechanical
Engineering and Materials Science Independent Study. 61.
https://openscholarship.wustl.edu/mems500/61

This Final Report is brought to you for free and open access by the Mechanical Engineering & Materials Science at
Washington University Open Scholarship. It has been accepted for inclusion in Mechanical Engineering and
Materials Science Independent Study by an authorized administrator of Washington University Open Scholarship.
For more information, please contact [email protected].
 Lane Change Strategy for
Autonomous Vehicle

By
Shiyu Xing

Supervised by
Mark Jakiela

Department of mechanical engineering & material science

Washington University in Saint Louis, March 2018
 Abstract

Abstract
Recently, people’s demand for smart vehicles continues to improve. As the core of

smart driving, driverless vehicle becomes the most concerned technology. Lane change,

the most common behavior in driverless situation, greatly affect the road efficiency. Fast

and safe lane change operations have very practical significance in reducing traffic

accidents. This paper uses driverless vehicle as research object, and the pathing planning

and pathing tracking for lane change situation are studied. An efficient path planning

method and trajectory tracking controller are designed and simulated. The main content

contains the three following aspects:

(1) A set of comprehensive lane change strategy is designed for different working

conditions. Then path planning for lane change is researched based on mass

point model and an efficient path planning method based on polynomial is

proposed and optimized.

(2) Kinematic model and 3 DOFs dynamic model of driverless vehicle based on

magic tire model are established using SIMULINK. Several simulation and test are

done to verify the rationality of the model.

(3) The trajectory -tracking control system based on PID controller is designed. Then

run simulation based on the model established and according to the results, the

trajectory -tracking control system can track the lane-changing path accurately

and analysis is made.

Key word: Driverless vehicle, Lane change, Path planning, Trajectory tracking control

I
 Category

Category
Abstract…………………………………………………………………………………………………………………………Ⅰ

Category…………………………………………………………………………………………………………………………Ⅱ

Chapter 1 Introduction………………………………………………………………………………………………… 1

1.1 Research background and significance………………………………………………………………… 1

1.2 Driverless vehicle overview…………………………………………………………………………………… 2

1.3 Key technologies of autonomous vehicle………………………………………………………………5

1.4 Research content……………………………………………………………………………………………………7

Chapter 2 Lane Change Strategy Design and Path Planning……………………………………9

2.1 Lane change strategy design…………………………………………………………………………………9

2.2 Lane change path planning………………………………………………………………………………… 11

2.3 Chapter summary………………………………………………………………………………………………… 18

Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation……………………19

3.1 Vehicle dynamic model…………………………………………………………………………………………19

3.2 Vehicle kinematic model……………………………………………………………………………………… 24

3.3 Simulation and analysis…………………………………………………………………………………………26

3.4 Chapter summary………………………………………………………………………………………………… 32

Chapter 4 Trajectory Tracking Control……………………………………………………………………… 33

4.1 Trajectory tracking PID controller………………………………………………………………………… 33

4.2 Simulation and result analysis……………………………………………………………………………… 34

4.3 Chapter summary………………………………………………………………………………………………… 37

Chapter 5 Summary and Outlook……………………………………………………………………………… 38

Reference……………………………………………………………………………………………………………………… 39

II
 Chapter 1 Introduction

Chapter 1 Introduction
1.1 Research background and significance

Vehicle is a machine that has changed the world and a machine with 100- year

history. As a kind of travel tool, vehicle brings great convenience to people and help them

save a lot of time. But while it is promoting the process of human civilization, the negative

impacts that cars bring us are growing. The increase of vehicle ownership brings a large

amount of traffic accidents, resulting in serious casualties and major economic loses. As

an essential part of vehicle usage, driving is full of danger and tedious. Drivers need to

pay attention to the surrounding environment and other vehicles all the time and then

make corresponding decision. Usually the decision depends on the drivers’ driving

experience and how concentrated the drivers are. In fact, the drivers’ judgement error is

the main reason for the vast majority of accidents, and about half of these accidents are

due to the drivers’ reaction delay.

With the rapid development of advanced technologies such as artificial intelligence

and pattern recognition, vehicle industry is undergoing dramatic changes. Car is no

longer a mere mechanical structure. It corporates many advanced scientific research to

improve the existing functions and increases the safety of the car. Some examples for this

are assistance driving and ABS system. Through these technologies, vehicles are getting

more and more intelligent. Driverless vehicles have become the technology of greatest

concern as the core of smart vehicles. Because some features like automatic road

identification, route design and vehicle body state adjustment free drivers from tedious

driving operations and make driving safer and easier. According to HIS, a consultancy for

automotive industry, the growth of autonomous vehicles is catching up with the electric

vehicles, and the average family will be able to enjoy the safe and convenient driving

experience by autonomous driving around 2025. Autonomous driving integrated the use

of many advanced technologies[1], such as active control, artificial intelligence and visual

analysis, so it is a measure of state research and industrial manufacturing.

Obstacle avoidance is an important feature of autonomous driving and lane change

is a frequent operation for autonomous vehicle. In the face of obstacles, the autonomous
1
 Chapter 1 Introduction

vehicle can obtain the obstacle and vehicle position information and select the best

obstacle avoidance method and routes, then control the speed and steering flexibly to

achieve a smoothie and safe driving. Therefore, in the process of obstacle avoidance, it is

necessary to realize the lateral and longitudinal control at the same time. In addition, the

smoothness, safety and speed of the obstacle avoidance process also need to be

considered. It is reported that there are about 1.2 million deaths caused by man-made

traffic accidents each year, most of them are caused by drivers’ negligence and fatigue.

Obstacle avoidance system can significantly reduce the drivers’ sense of tension and then

assist or replace the driver to take obstacle avoidance measures, thus minimize the

casualties caused by the accident.

In the context of booming electric vehicle and autonomous driving technology,

research on the obstacle avoidance strategy, especially for the lane change situation, has

theoretical and practical significance for the future development of obstacle avoidance

control system of self-driving vehicle.

1.2 Driverless vehicle overview

The superior performance of driverless technology over traditional vehicle in terms

of safety and accuracy has been fully demonstrated and its future development is also

generally accepted. First, driverless vehicle can reduce the number of man-made

accidents on a large scale, reduce the number of casualties and reduce the huge amount

of medical compensation and handling costs. Statistics show that unmanned technology

can reduce the number of traffic accidents by about 90%, and it can also significantly

reduce the car’s travel time and energy consumption, while reducing the number of

vehicles. Besides, autonomous vehicles also have the function of traffic congestion

avoidance, speed improvement and optimal path planning, which can greatly enhance

the efficiency of the road.

Since the 1970s, some developed countries, such as the United States, Britain, Japan

and Germany, have started their research on driverless vehicle. From IT companies to car

manufactures, governments and organizations, it’s a growing consensus that

autonomous vehicles will be the future of the entire automotive industry. A lot of money

2
 Chapter 1 Introduction

and energy have been invested in related research and breakthroughs have been made

both in terms of feasibility and practicality.

The United States has achieved a high-level research on the key technologies of

unmanned vehicles. The most famous one is driverless car developed by Google, as

shown in Figure 1.1. Google Driverless Car is a fully autonomous driving car developed

by Google X Labs and it can be started, traveled and stopped without driver. This project

is led by Sebastian Trong, the inventor of Google Street View and the director of the

artificial intelligence lab of Stanford. These vehicles use cameras, radar sensors and laser

range finders to “see” other traffic conditions and use detailed maps to navigate the road

ahead. Just set the destination, the vehicle will automatically plan the optimal route. When

the vehicle is traveling according to the planned route, it will upload the route and traffic

to the central data processing center, which will further improve the map and make the

route planning more accurate. Allegedly, recorded a total mileage data of Google

driverless cars has reached 700,000 miles.

Fig 1.1 Google Driverless Car

As soon as the British government announced the new rules that UAVs would be

legally available, the British company RDM Group could not wait to release the prototype

Lutz Pathfinder, the country’s first unmanned vehicle, as shown in Figure 1.2. Lutz

Pathfinder is designed to help passengers, shoppers and the elderly travel short distances.

It can only carry two passengers with top speed of 15mph and battery duration of 8 hours.

The Mobile Robotics Group research team at Oxford University designed a full range of

image capture systems for unmanned vehicles, including 22 sensors, laser imaging
3
 Chapter 1 Introduction

components, panoramic cameras and radar positioning systems.

Fig 1.2 Lutz Pathfinder

France INRIA company spent ten years developing autonomous vehicle “Cycab”,

which looks like a future golf carts, as shown in Figure 1.3. Ordinary GPS system accuracy

can only reach a few meters, while “Cyber” is equipped with a special GPS system called

“real-time motion GPS”, its accuracy is up to 1 cm. This driverless car is equipped with

laser sensors that act as “eyes”, which helps the car Avoid obstacle along the way. It is

also equipped with twin cameras to follow signposts. People can even drive their cars

through their cell phones and every unmanned vehicle can communicate with each other

over the internet. This means that information sharing can be done between such

driverless cars, so that multiple driverless cars can make up a team and run in small

intervals. This car can also get real traffic information through the traffic network to

prevent traffic jams and it will automatically issue a warning to remind past pedestrians’

attention.

Fig 1.3 Cycab

4
 Chapter 1 Introduction

In China, there are also several companies that are worth our attention. The most

well-known one is Baidu, which has been developing driverless technology since 2013,

shown in Fig 1.4. In December 2015, it announced an establishment of an autonomous

driving business unit, and also announced a development plan of “three-year business,

five-year volume production”. In the round-trip test in April 2016, Baidu’s self-driving

vehicles successfully achieved the following complex driving actions such as deceleration,

lane change, overtaking, up-and-down ramping, and turning around. It also completed

the switching between different road scenarios like entering express way and driving out

of express way. The maximum speed was up to 100km/h in the test.

Fig 1.4 Baidu autonomous vehicle

Unmanned driving technology is the ultimate manifestation of intelligence. With the

continuous advancement of high technologies such as cloud computing, artificial

intelligence, modern sensing, information fusion and automatic control, the future

development speed of driverless car will accelerate.

1.3 Key technologies of autonomous vehicle

1.3.1 Environmental perception technology

Environmental perception technology is equivalent to the eyes and ears of

unmanned vehicles. Driverless vehicles identify surrounding environmental information

support for their decision making through this technology. Environmental perception

includes the perception of the position and attitude of driverless cars and the perception

5
 Chapter 1 Introduction

of the surrounding environments. A single sensor can measure only one aspect of the

measured object, which is unable to meet demands. Therefore, multiple sensors must be

used to measure one or more features of a measured object at the same time. After the

measured data is processed by computer, a useful signal is extracted.

The position and attitude information of the driverless vehicle mainly includes the

speed, acceleration, inclination, position, etc. This type of information is easy to measure

and is mainly measured with sensors such as drive motors, electronic compasses, tilt

sensors and gyroscopes.

Perception of the surrounding environments is realized mainly by active ranging

sensors such as radars, supplemented by passive ranging sensors. Because the

combination of laser, radar, ultrasonic and other active ranging sensors can meet the

needs of the task under complex and harsh conditions. Besides, the most important thing

is good real-time performance.

1.3.2 Navigation and positioning technology

The driverless car’s navigation module is used to determine the location, which is the

support for the mission and path planning of the unmanned vehicle. Navigation can be

divided into autonomous navigation and network navigation.

Autonomous navigation technology can complete the navigation task only with the

help of positioning assistance. Autonomous navigation technology stores geolocation

data locally and all calculations are done at the terminal. This can be done in any situation,

but the limited computing resources of autonomous navigation devices result in poor

computing power and sometimes fail to provide accurate and real-time navigation

services. Existing autonomous navigation can be divided into three categories: relative

positioning, absolute positioning and combination positioning. Relative positioning

mainly relies on the odometer, gyroscope and other internal sensors. It measures the

displacement of unmanned vehicles relative to the initial position to determine the

current location. Absolute positioning mainly uses navigation beacons, active or passive

identification, map matching or global positioning system positioning. The combination

positioning is the combination of the previous two, which can make up for the lack of

6
 Chapter 1 Introduction

single method.

Network navigation can carry out information exchange anytime and anywhere

through wireless network and traffic information center. Mobile devices are connected to

a direct server via a mobile communication network. The server performs map storage

and complex calculations and other function and users can download map data from the

server. Advantage of network navigation is that there are no storage capacity constraints,

strong computing power and it’s able to store any elaborate map, and map is always up-

to-date.

1.3.3 Path planning technology

Path planning is a bridge for information perception and intelligent control of

driverless vehicles and it is also the basis for autonomous driving. The mission of path

planning is to find a path from the initial state to the target state according to certain

evaluation criteria in an environment with obstacles. Path planning can be divided into

global path planning and local path planning. In the case of a known map, the global

plan uses known local information such as obstacles and road boundaries to determine

the feasible optimal route, which combines the optimization and feedback mechanisms

well. Local planning is based on the driving area of global planning under the guidance

of the sensor. It determines the driving trajectory of the road ahead that vehicle needs.

Global planning applies to circumstances with known environment. Local planning

applies to situations where environment is unknown.

1.3.4 Control technology

The main function of the control module is to make the estimation based on the

information obtained by the sensing system, and then make decisions on the next action

and then control the vehicle. Control system constantly adjust the steering and wheel

angle and speed based on the deviation between the current position and desired path

until reaching the destination.

1.4 Research content

This paper uses autonomous vehicle as research object. The research studies path

planning and trajectory tracking control in the obstacle avoidance conditions, especially

7
 Chapter 1 Introduction

lane changing obstacle avoidance process. The main research includes three aspects:

(1) A set of comprehensive obstacle avoidance strategy is designed for different

working conditions. Safe distance between vehicle is studied to avoid accidents. Then

path planning for lane change is researched based on mass point model and an efficient

path planning method based on polynomial is proposed.

(2) Kinematic model and 3 DOFs dynamic model of driverless vehicle based on magic

tire model are established using SIMULINK and MATLAB. Several simulations and tests

are done to verify the rationality of the model.

(3) The trajectory -tracking control system based on PID controller is designed. Then

run simulation based on the model established and according to the results, the trajectory

-tracking control system can track the lane-changing path accurately and analysis is

made.

8
 Chapter 2 Lane Change Strategy Design and Path Planning

Chapter 2 Lane Change Strategy Design and Path Planning

Obstacle avoidance, especially lane change, is a common driving behavior in

driverless situations. The autonomous vehicle can select the optimal obstacle avoidance

method based on the perceived obstacles and the location of the vehicle and control the

speed and steering flexibly to achieve a safe and stable driving. Successful lane change is

related to whether the lane change strategy is scientific. At present, path planning has

been the hotspot of research, and many research results have been obtained. Some

theories have been verified on real vehicles. However, in actual driving process, the

existing optimization process of the lane change planning is complicated and not

conductive to real-time calculation. Therefore, it is more practical to design an effective

and flexibly calculated lane change trajectory planning method.

This chapter first introduces the characteristics of various lane changing process, and

designs a comprehensive lane changing control strategy for different lane changing

conditions encountered during driving. Then according to the specific lane change

conditions, the path planning is studied.

2.1 Lane change strategy design

According to the different lane changing environment, the lane change behavior can

be divided into two situations, which are mandatory lane change and discretionary lane

change. Mandatory is the normal driving process in which the vehicle has to change lane.

The key point of such behavior is that there is a latest lane change point that the

unmanned vehicle must complete lane change before travelling to that point, which

usually occurs at intersections, lane merging, diversion and obstacles, as shown in Fig 2.1.

The discretionary lane change is the lane changing behavior of unmanned vehicles, the

purpose of which is to reach the driving target and surpass the front vehicle. This behavior

is mainly to complete the purpose of driving to reduce driving time, which usually occurs

in the situation where the vehicle behind travels faster than the front vehicle and the

driving conditions in the adjacent lane are more superior.

9
 Chapter 2 Lane Change Strategy Design and Path Planning

Fig 2.1 Mandatory lane change

When vehicles make lane change decisions, there may be three kinds of situations:

stop, follow and lane change. Stop is implemented by braking system to gradually

decelerate the vehicle to a stationary state. Stop is a safe and effective method when

there is a mandatory lane change for unmanned vehicles, but the actual situation cannot

meet the necessary conditions for lane change. Following the front car means the car

behind follow the front car to continue driving in the original lane. When autonomous

vehicles are under discretionary lane change situation, but the reality cannot meet the

necessary conditions for lane change, follow the car is a viable method. While driving

continues, control system will maintain the original state and wait for the opportunity to

change lanes. If there is enough lane change space on the target lane and the distance

to the front car then the autonomous vehicle will be able to change lanes.

In this paper, a lane changing strategy is designed improve traffic efficiency and

ensure safety. The flow chart of decision-making is shown in Figure 2.2. When a driverless

vehicle encounters a mandatory lane change, the vehicle will change lanes if the target

lane has sufficient lane change space, otherwise, the vehicle will only be able to stop and

wait for the next lane change opportunity. When driverless vehicle change lane in order

to avoid collisions with the front car or other driving purpose, which is discretionary lane

change, if the target lane has enough lane space and the distance to the front vehicle is

greater than the safe distance, then vehicle will change lane; otherwise, the vehicle needs

to continue to follow the original lane until the new lane changing opportunities occur.

10
 Chapter 2 Lane Change Strategy Design and Path Planning

Lane Change

N Y
Discretionary

Y N
Enough space
N Y in target lane
Enough space
in target lane
Y N
Distance >
Safe distance

Stop Change lane Follow Follow

Fig 2.2 Decision making strategy

Autonomous vehicles make the path planning based on the decision of lane change

strategy. Stop and following front car mainly based on the judgement of lane change

conditions and lane space. When lane change is viable, a reasonable path will be planned.

2.2 Lane change path planning

Lane change is a very common behavior. Smooth transition to an adjacent lane is a

basic requirement of autonomous driving. Especially when vehicles encounter obstacles

that block the original route, the importance of lane change movements is quite

prominent. It is common sense that the lane changing movements are related to the

continuity of autonomous driving. Choosing the appropriate lane change path is

important for the smoothness of the transition.

A variety of reasonable lane change trajectories have been studied, refined and

validated. Sine lane change trajectory design is simple and easy to adjust, but the initial

lateral acceleration is not zero, which means a lateral load will be applied on passengers

immediately. This kind of impact damages the travel experiences a lot. Although the arc

trajectory minimizes the time consumption of the lane change process, but there is step

change in the lateral acceleration, which will increase the difficulty of lateral control.
11
 Chapter 2 Lane Change Strategy Design and Path Planning

Trapezoidal acceleration lane change trajectory can continuously change, but it is

necessary to set planning parameters, so the flexibility is poor. At present, the polynomial

trajectory[2] used in the research of lane change does not have the defects above. The

displacement, velocity and acceleration of the planned polynomial function are smooth.

The path planning is carried out according to the actual driving conditions, and lane

change is stable and easy to control. This section will design the path based on

polynomial and analyze feasible methods for generating the path under obstacle

constraints conditions.

2.2.1 Path planning method based on polynomial function

Lane change diagram is shown in Figure 2.3, assuming the initial state of the vehicle

at 𝑡0 is 𝑝0 = [𝑋0 , 𝑋̇0 , 𝑋̈0 , 𝑌0 , 𝑌̇0 , 𝑌̈0 ]. The six state variables indicate the longitudinal and lateral

displacement, velocity and acceleration respectively. The status at end time 𝑡𝑓 is then 𝑝𝑓 =

[𝑋𝑓 , 𝑋̇𝑓 , 𝑋̈𝑓 , 𝑌𝑓 , 𝑌𝑓̇ , 𝑌𝑓̈ ].

Y p𝒇

X p𝟎
0

Fig 2.3 Lane change diagram

The initial state can be measured and estimated by various sensors, while the final

state can be estimate based on the driving demand. From the vector, it is easy to know

that the lateral and longitudinal states have 6 constraints respectively. Therefore, a

quantic polynomial path with 6 parameters can be selected to define the longitudinal and

lateral path planning and relation is given as follows[3]:

5

X(𝑡) = ∑ 𝑎𝑖 × 𝑡 𝑖 (2-1)
𝑖=0

5
(2-2)
Y(𝑡) = ∑ 𝑏𝑖 × 𝑡 𝑖
𝑖=0

12
 Chapter 2 Lane Change Strategy Design and Path Planning

Polynomial coefficient can be expressed in vectors:

𝒂= [𝑎0 , 𝑎1 , 𝑎2 , 𝑎3 , 𝑎4 , 𝑎5 ]𝑇 , 𝒃= [𝑏0 , 𝑏1 , 𝑏2 , 𝑏3 , 𝑏4 , 𝑏5 ]𝑇 (2-3)

The vectors are calculated as follows:

𝐚 = 𝑇 −1 ∙ 𝑞𝑥 , 𝐛 = 𝑇 −1 ∙ 𝑞𝑦 (2-4)

Where,

𝑇
𝑞𝑥 = [𝑋0 , 𝑋̇0 , 𝑋̈0 , 𝑋𝑓 , 𝑋𝑓̇ , 𝑋𝑓̈ ] (2-5)

𝑇 (2-6)
𝑞𝑦 = [𝑌0 , 𝑌0̇ , 𝑌0̈ , 𝑌𝑓 , 𝑌𝑓̇ , 𝑌𝑓̈ ]

1 𝑡0 𝑡02 𝑡03 𝑡04 𝑡05

0 1 2𝑡0 3𝑡02 4𝑡03 5𝑡04
0 0 2 6𝑡0 12𝑡02 20𝑡03 (2-7)
T= 1 𝑡𝑓 𝑡𝑓2 𝑡𝑓3 𝑡𝑓4 𝑡𝑓5
0 1 2𝑡𝑓 3𝑡𝑓2 4𝑡𝑓3 5𝑡𝑓4
[0 0 2 6𝑡𝑓 12𝑡𝑓2 20𝑡𝑓3 ]

If the time interval, the initial state and final state can be known, the polynomial

coefficient can be calculated and then the ideal path can be obtained. Here’s how to

design a lane changing path using the above polynomial path planning approach.

To reduce the difficulty of the implementation of control, autonomous vehicles

normally choose a constant speed when driving forward. Assume the car moves to

position (0,0) at constant speed 10m/s at t0, and there is a static obstacle 50m in front of

it. Then it is necessary to make the lane change. First, estimate the end position. The

lateral end position is generally the width of the road, which is 3.5m. Longitudinal end

position is then 50m. Since the constant speed can reduce the difficulty of the control of

the autonomous vehicle, we try to make a constant speed planning in the longitudinal

movement. Assume that the final longitudinal velocity is the same as the initial velocity.

Both are 10m/s, then it takes 50/10=5s to accomplish the lane change process. Then we

can get 𝑞𝑥 = [0,10,0,50,10,0] and 𝑞𝑦 = [0,0,0,3.5,0,0]. Using equation (2-4), we can get

the longitudinal and lateral state as shown in Figure 2.4. It can be seen from the planning

results that the path planned can reach the desired position accurately and the changes

of speed and acceleration are relatively smooth, which is very beneficial to the control

system for trajectory tracking.

13
 Chapter 2 Lane Change Strategy Design and Path Planning

Fig 2.4 Longitudinal and lateral state

As can be seen from Fig 2.4, the planned motion is longitudinal constant. This

polynomial lane change, which is planned by selecting the appropriate lane change time

and longitudinal displacement at the end, is called a longitudinal constant velocity

polynomial lane change. This kind of lane change design can reduce the difficulty of

longitudinal control, and the output of each lateral state quantity has regularity, which

facilitates the rapid calculation.

2.2.2 Constraints analysis and optimization

The research in this section mainly focuses on the situation with obstacles. The

situation of discretionary lane change is relatively simple, so we will not discuss too much

in this section. The purpose of the lane change is usually to avoid the obstacles ahead,

but the lane change process may also lead to collision, therefore it is necessary to set

reasonable constraints to limit the lane change process. Meanwhile, the dynamic

characteristics of the vehicle must also be taken into account, which ensure the planned

lane change trajectory is executable. Defining the appropriate conditions for obstacle

avoidance is an important step in completing the lane change planning. The following

sections define the external and internal constraints.

14
 Chapter 2 Lane Change Strategy Design and Path Planning

A (t=c)

X 𝐀 (t=0) B

Fig 2.5 Constraint analysis

The collision that occurs during the lane changing mainly refers to the situation when

the vehicle approaches the obstacle and fails to fully enter the adjacent lane, which results

in contact and collision with the obstacle. In order to avoid collision, the first constraint

can be defined as follows:

When the front end of car A reaches the end of the obstacle B, the lateral distance

between the two should be greater than the width of the obstacle, otherwise collision

occurs.

Because the general heading angle is relatively small, while the lateral displacement

of the car is much smaller than the longitudinal displacement, it can be assumed that the

angle between the car and longitudinal axis of the road is approximately zero. The

diagram of critical collision is shown in Fig 2.5. After time tc, the longitudinal displacement

of car A is L and the lateral displacement is yc, and the front end of car A and end of

obstacle are close to collinear state. Assume the width of the obstacle, which is another

vehicle in most of the cases, is w. Then the first constraint can be represented by 𝑦𝑐 > 𝑤.

If the current speed is V, then the critical time tc can be decided by L/V.

In the case of unreasonable lane change route planning, it is possible that a collision

may occur during the lane changing process. Based on the research on the polynomial

function, the lateral displacement at same longitudinal displacement can be increased by

reducing the longitudinal final displacement. In this way, collision will be avoided. On the

basis of the planned routes in Fig 2.4, reduce the longitudinal displacement to 40m and

the lane change time to 4s. Fig 2.6 shows the results of comparison of displacements

before adjustment and after adjustment. Obviously, there is a significant increase in the

lateral displacement at the same longitudinal displacement.

15
 Chapter 2 Lane Change Strategy Design and Path Planning

Fig 2.6 Before & after Comparison

But the adjustment also causes an increase in the maximum lateral acceleration. Too

large lateral acceleration will result in more difficulty in control and bad riding comfort,

so the second constraint is defined as follows:

The lateral acceleration of the vehicle should be controlled within a reasonable range.

In this paper, the lateral acceleration ay is between -2 m/s2 and -2 m/s2.

The vehicle will have tendency to roll under the action of lateral acceleration, which

not only affects the comfort of riding but also poses a great threat to driving safety. Take

both comfort and safety into consideration, assume the maximum lateral acceleration is

2 m/s2. According to Fig 2.4, we find that the maximum lateral acceleration occurs at

about a quarter of the time consumption of whole lane change process. According to the

calculation, the error of this estimation is merely 2%. Thus, we can approximate that the

lateral acceleration corresponding to the 1/4 lane change time is the maximum lateral

acceleration. This method can improve the efficiency of calculation. When the maximum

lateral acceleration exceeds the constraint limit, the lane change need to be adjusted. The

maximum lateral acceleration can be reduced by increasing the longitudinal displacement

in the end. Again, on the basis of planned route in Fig 2.4. increase the longitudinal

displacement to 60m, Figure 2.7 shows the comparison result of before and after

adjustment.

We can clearly see that increasing the lane change time can reduce the maximum

lateral acceleration, but this will also reduce the lateral displacement

16
 Chapter 2 Lane Change Strategy Design and Path Planning

Fig 2.7 Before & after Comparison

On one hand, we hope to reduce the final longitudinal displacement to avoid

obstacles and enhance safety, on the other hand, we also hope to increase the final

longitudinal displacement to reduce the maximum lateral acceleration and improve the

riding comfort. So, we need to take both of them into consideration, select the best

longitudinal displacement at the end, so that the lane change process is safe and

comfortable, and of course the constraints discussed previously must be satisfied. Now
[4]
we have an optimization problem here.

The design variable of this problem is the final longitudinal displacement. Assume

the distance between the vehicle and obstacle is d. Then usually the range of the final

longitudinal displacement is [0.7d, 1.3d], which is the design space. Our object is to

enhance the safety as well as riding comfort. Since we have two different objectives, we

add them linearly with different weight to transfer the problem from a multiple objective

problem to a single objective problem. The constraints are introduced previously. Use the

MATLAB genetic algorithm toolbox to optimize the problem under different conditions,

and the results are as follows:

Table 2.1 Optimization results

Current speed Distance to the obstacle Optimal final longitudinal displacement

10 m/s 30 m 32.28 m

20 m/s 40 m no optimal found

30 m/s 80 m 94.15 m
17
 Chapter 2 Lane Change Strategy Design and Path Planning

It can be noticed that no optimal final displacement can be found under some

situations, because the speed is too high for the distance to the obstacle, which means

only a little time is given to the lane change process and this will lead to a very high lateral

acceleration which probably exceed 2m/s2. For situation like this, the sensor need to

detect obstacles earlier and the brake system should engage to decelerate the vehicle in

the lane change process, and this will not be discussed in this paper.

2.3 Chapter summary

The chapter discussed the design of a comprehensive and safe lane change strategy.

First, characteristics of several different operations such as stop, following and lane

change are analyzed. Then according to the specific condition and space, lane change

strategy will be selected flexibly.

When studying the path planning, the longitudinal velocity is assumed to be constant.

Use the polynomial to design the path and it works pretty well. We also figure out that

both maximum lateral acceleration and lateral displacement are related to the final

longitudinal displacement. The relation is studied, and optimization is made to enhance

the safety and riding comfort during the lane change process.

18
 Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation

Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation

As an extremely complex non-linear model, a car can be regarded as a system

composed of a car body and four wheels. In the last chapter of study, we take the car as

a mass point and we designed the lane change path by studying its longitudinal and

lateral motion. However, in fact, in addition to the longitudinal and lateral translation, yaw,

roll and pitch are all very significant movement of vehicle. In order to study the path

design more accurately, we need to establish the dynamics and kinematic model[5].

In this chapter, a three-DOF vehicle dynamic model, a magic tire model and a

kinematic model are built according to the characteristics of the automobile, then

simulation framework is set up in SIMULINK and simulation results are analyzed.

3.1 Vehicle dynamic model

Many scholars have made a great deal of research on vehicle dynamic modeling,

and the most commonly used one is three-DOF vehicle dynamic model with longitudinal

velocity, lateral velocity and yaw velocity. This is exactly the model used in this chapter,

the model is simple, so it can reduce computational complexity. It can also reflect the

research purpose of this paper. To use this model, we make the following idealized

assumptions:

(1) Ignore the steering linkage mechanism, take the steering angle of the front wheel

as input directly.

(2) Assume the autonomous vehicle runs on a flat road and neglect the vertical

movement of the vehicle

(3) Assume the suspension system and vehicle are rigid, regardless of the effect of

the suspension movement.

(4) Ignore the impact of air resistance and rolling resistance of the tires

The model includes longitudinal, lateral and yaw motion of the body, which is shown

in Fig 3.1 and can be described by the following equation:

m𝑉𝑥̇ = m𝑉𝑦 𝜑̇ + 𝐹𝑥1 + 𝐹𝑥2 + 𝐹𝑥3 + 𝐹𝑥4 (3-1)

m𝑉𝑦̇ =−m𝑉𝑥 𝜑̇ + 𝐹𝑦1 + 𝐹𝑦2 + 𝐹𝑦3 + 𝐹𝑦4 (3-2)

𝐼𝑧 𝜑̈ = (𝐹𝑦1 + 𝐹𝑦2 )𝑏 − (𝐹𝑦3 + 𝐹𝑦4 )𝑎 + (−𝐹𝑥1 + 𝐹𝑥2 − 𝐹𝑥3 + 𝐹𝑥4 )𝑐 (3-3)

19
 Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation

Fy1
Ft1 Fl1
Ft3 v

Vy 𝛅
Fl3
Fx1

c
𝝋̇
Vx
a b

Ft2 Fl2

Fl4

Ft4
Fig 3.1 3-DOF vehicle dynamic model

Where, the 𝑉𝑥 and 𝑉𝑦 are the longitudinal and lateral velocity in the vehicle coordinate

system respectively. 𝑉𝑥̇ and 𝑉𝑦̇ are the change rate of the longitudinal and lateral

velocity rather than the acceleration in longitudinal and lateral direction, since the yaw

angle is always changing. The longitudinal and lateral acceleration can be calculated using

𝐹𝑥𝑖 and 𝐹𝑦𝑖 (longitudinal and lateral force on tire, i=1,2,3,4) respectively, and the

accelerations calculated are the components of absolute acceleration in the vehicle

coordinate system. Please think carefully about this. 𝜑̇ and 𝜑̈ are yaw velocity and yaw

acceleration respectively. m is the total mass of the vehicle and 𝐼𝑧 is the moment of

inertia. 𝑎 and 𝑏 are distance from the centroid to front axle and rear axle respectively.

𝑐 is the distance from the tire to the longitudinal axis of the vehicle. Longitudinal and

lateral force (𝐹𝑥𝑖 and 𝐹𝑦𝑖 ) on the tire can be calculated using tire longitudinal and lateral

force (𝐹𝑙𝑖 and 𝐹𝑡𝑖 ) as well as steering angel δ:

𝐹𝑥 = 𝐹𝑙 cos 𝛿 − 𝐹𝑡 sin 𝛿 (3-4)

𝐹𝑦 = 𝐹𝑙 sin 𝛿 + 𝐹𝑡 cos 𝛿 (3-5)

The tire longitudinal and lateral forces (𝐹𝑙 and 𝐹𝑡 ) can be obtained according to the

tire model described below.

20
 Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation

3.1.1 Tire model

The road will exert friction on the tire when vehicle is moving, while tire is contacting

with the road. The force will be transmitted to the car body through tire. The interaction

between the wheel and the road surface generates tire longitudinal force and tire lateral

force respectively. The longitudinal force can realize braking and driving functions of the

vehicle, while tire lateral force provides the steering function for the vehicle. Therefore,

tire model is the key to vehicle dynamics analysis.

Tire models can be divided into three kinds: the theoretical tire model, the experience

tire model and the physical tire model. The most widely used of the these is Magic

Formula proposed by Pacejka. It can express the tire characteristics under different driving

conditions, the formula parameters are fitted according to the experimental data, and the

tire longitudinal force and lateral force are solved by the slip ratio, slip angle and vertical

load on tire. The following formula can be used to represent the relation between the

force on tire and the related variables:

𝐹𝑙 = 𝑓𝑙 (𝛼, 𝑠, 𝐹𝑧 ), (3-6)

𝐹𝑡 = 𝑓𝑡 (𝛼, 𝑠, 𝐹𝑧 ) (3-7)

Solving process is shown as below:

Input
Output

Vertical load Fz
Tire model Tire longitudinal force Fl
Slip ratio s
Tire lateral force Ft
Slip angle α

Fig 3.2 Relation between input and output

During driving, acceleration will change the center of gravity and the weight

distribution, so vertical load of each tire can be expressed as follows:

𝑚𝑔𝑎 𝑚ℎ𝐶𝐺 𝑎𝑥 𝑚ℎ𝐶𝐺 𝑎𝑦 (3-8)
𝐹𝑧1 = − −
2(𝑎 + 𝑏) 2(𝑎 + 𝑏) 4𝑐
𝑚𝑔𝑎 𝑚ℎ𝐶𝐺 𝑎𝑥 𝑚ℎ𝐶𝐺 𝑎𝑦
𝐹𝑧2 = − + (3-9)
2(𝑎 + 𝑏) 2(𝑎 + 𝑏) 4𝑐

21
 Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation

𝑚𝑔𝑏 𝑚ℎ𝐶𝐺 𝑎𝑥 𝑚ℎ𝐶𝐺 𝑎𝑦

𝐹𝑧3 = + − (3-10)
2(𝑎 + 𝑏) 2(𝑎 + 𝑏) 4𝑐
𝑚𝑔𝑏 𝑚ℎ𝐶𝐺 𝑎𝑥 𝑚ℎ𝐶𝐺 𝑎𝑦 (3-11)
𝐹𝑧4 = + +
2(𝑎 + 𝑏) 2(𝑎 + 𝑏) 4𝑐
Where the 𝑚 is the mass of the vehicle. ℎ𝐶𝐺 is the centroid height. 𝑎𝑥 and 𝑎𝑦 are the

longitudinal and lateral acceleration in vehicle coordinate system respectively. Notice the

𝑎𝑥 and 𝑎𝑦 are calculated using vehicle longitudinal and lateral force (𝐹𝑥𝑖 and 𝐹𝑦𝑖 ) as

well as Newton’s second law, however they are not 𝑉𝑥̇ and 𝑉𝑦̇ . Please think carefully

about this.

The slip angle of each tire can be calculated using the formulas below:
(𝑉𝑦 + 𝑏𝜑̇ )
𝛼1 = tan−1 −𝛿 (3-12)
(𝑉𝑥 − 𝑐𝜑̇ )
(𝑉𝑦 + 𝑏𝜑̇ )
𝛼2 = tan−1 −𝛿 (3-13)
(𝑉𝑥 + 𝑐𝜑̇ )
𝑉𝑦 − 𝑎𝜑̇
𝛼3 = tan−1 (3-14)
𝑉𝑥 − 𝑐𝜑̇
𝑉𝑦 − 𝑎𝜑̇ (3-15)
𝛼4 = tan−1
𝑉𝑥 + 𝑐𝜑̇
The slip ratio of each tire can be expressed as follows:
𝜔∙𝑟
s= −1 (3-16)
𝑣𝑙

Where 𝑟 is the effective radius of the tire, and 𝜔 is the angular velocity of the wheel. 𝑣𝑙

is the tire longitudinal velocity (not vehicle longitudinal velocity 𝑉𝑥 ), and they are

calculated as follows:

𝑣𝑙1 = (𝑉𝑥 − 𝑐𝜑̇ ) cos 𝛿 + (𝑉𝑦 + 𝑏𝜑̇ ) sin 𝛿 (3-17)

𝑣𝑙2 = (𝑉𝑥 + 𝑐𝜑̇ ) cos 𝛿 + (𝑉𝑦 + 𝑏𝜑̇ ) sin 𝛿 (3-18)

𝑣𝑙3 = 𝑉𝑥 − 𝑐𝜑̇ (3-19)

𝑣𝑙4 = 𝑉𝑥 + 𝑐𝜑̇ (3-20)

According to the fitting parameters of the magic tire model, there is a linear region

in the relationship between the tire longitudinal force and slip ratio. And linear region

also exists in the relationship between the lateral force and slip angle. Therefore, the force

on tire can be represented using following formulas:

𝐹𝑙 = 𝐶𝑙 ∙ 𝑠 , 𝐹𝑡 = 𝐶𝑡 ∙ 𝛼 (3-21)

Where 𝐶𝑙 is the longitudinal stiffness and 𝐶𝑡 is the lateral stiffness. Research has shown

22
 Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation

that when the lateral acceleration is lower than 0.4g, the linear tire model has relatively

good fitting accuracy. This model can also reduce computational complexity.

According to the mathematical model introduced earlier, we can build a model with

SIMULINK and packaged then into different sub-systems. Assume the vehicle is moving

on a horizontal dry cement road, so the friction coefficient is constant. Neglect the

steering linkage mechanism, take the steering angle as input directly. The vehicle dynamic

model is shown as follows:

Fig 3.3 Vehicle dynamic SIMULINK model

Before we set up the input signal and run the simulation, considering that we may
23
 Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation

want to visualize how the vehicle move under the control of the steering input, we need

to build another sub-system, which is associated with vehicle kinematic modeling.

3.2 Vehicle kinematic model

When we want to take vehicle as an independent individual to carry out research,

the vehicle coordinate system is generally used, just like what we do when we build the

dynamic model. When it is necessary to consider the relative movement of the vehicle

and surrounding objects, a global inertia coordinate is required. The kinematic model of

a vehicle which is turning on a flat road is shown in the Fig 3.4.

𝛗
Y(t)

X
O
X(t)

Fig 3.4 Vehicle kinematic model

Where (𝑋(𝑡), 𝑌(𝑡)) is the coordinate of the midpoint of the rear axle in the global

coordinate system. φ is the yaw angle and δ is the steering angle. 𝑉𝑥 and 𝑉𝑦 are the

longitudinal and lateral velocity in the vehicle coordinate system respectively, just as

introduced dynamic model earlier. Here we should notice that the vehicle coordinate

system is not a coordinate system fixed on the vehicle body. It is, in essence, a static

coordinate system aligned with the vehicle longitudinal and lateral direction at every
24
 Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation

moment. Otherwise, the 𝑉𝑥 and 𝑉𝑦 will be both 0. l is the wheelbase of the vehicle.

According to the geometric relation, the vehicle kinematic model can be expressed using

the following formula:

𝑋(𝑡) = 𝑉𝑥 cos φ − 𝑉𝑦 sin φ (3-22)

𝑌(𝑡) = 𝑉𝑦 cos φ + 𝑉𝑥 sin φ (3-23)

Build a kinematic model subsystem in SIMULINK then connect it with the dynamic

model we build previously, which is shown in Fig 3.5.

Fig 3.5 Vehicle kinematic SIMULINK model

25
 Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation

3.3 Simulation and analysis

To verify the rationality of models built, we use different steering angle input to

simulate and observe the rationality of the state variable and trajectories of the vehicle.

In this section, we will run two simulations with different input steering angle signals, and

then analyze the state variable output like yaw velocity, slip ratio, slip angle lateral

acceleration and so on. Firstly, a ramp signal is used as input and then a sine wave signal.

And in both simulation the vehicle velocity is 36km/h.

3.3.1 Ramp input

The steering angle start with an initial value 0 which means no steering, and then it

begins increasing at 1s, finally it reaches 3.438 degrees and then keeps constant. The

input is shown as follow:

Vehicle state outputs are shown as follows:

As we can see, because the steering angle is 0 until 1s, the vehicle is moving in a

straight line in fact. Correspondingly, the yaw velocity and lateral acceleration are both 0

26
 Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation

until 1s. Once the vehicle begins turning, yaw velocity and lateral acceleration increases.

Since the steering angle is positive, which means the vehicle turns left, the yaw velocity

and lateral acceleration are both positive, which means the vehicle is rotates anti-

clockwise and the acceleration is to the left. When the steering angle reaches constant

value, yaw velocity and lateral acceleration also reaches a constant, since it’s a steady

state then.

Since the vehicle moves in a straight line until 1s with constant speed, no acceleration

causes the load redistribution. So, we can see same vertical load on tire 1 and 2, and same

vertical load on tire 3 and 4. Once steering angle is positive, the vehicle begins turning

left and there will be an acceleration to the left, which means the center of gravity of

vehicle will move to the right. So, the load on the right two tires, 2 and 4, increases. The

load on the left two tires, 1 and 3, decreases. Then all of them becomes steady state once

27
 Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation

the steering angle reaches constant.

The motion of the four wheels are pure rolling until 1s, so the slip ratio is 0. When

vehicle turns left, the right two wheels have a longer traveling curve than the left two

wheels according to Ackerman steering geometry. We didn’t build the model of

differential in this paper, so the rotational speed of the left and right half shaft are same,

which means the angular velocity of wheels cannot match the expected velocity for pure

rolling motion. Wheels on the right has a higher velocity than expected which leads to

the negatives slip ratio on tire 2 and 4. Wheels on the left has a smaller velocity than

expected which leads to the positive slip ratio on tire 1 and 3. Then all of them becomes

steady state once the steering angle reaches constant.

Slip angle of 4 tires are all 0 until 1s. When vehicle begins turning left, the slip angle

becomes negative to produce positive tire lateral force(to the left) which provide the

lateral acceleration to the left

28
 Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation

The trajectory is shown by the kinematic model. There isn’t steering angle until 1s so

the vehicle moves straightly, then steering angle occurs and vehicle begins truing left.

Finally, the steering angle reaches constant value and vehicle begins a circular motion

just as expected.

3.3.2 Sine input

The steering angle input is a sine wave with amplitude of 3.438 degrees and period

of 4π, which, in real world, is a driver turns left and right alternately. The input signal is

shown as follow:

Vehicle state output shown as follows:

Both yaw velocity and lateral acceleration are sine wave with period of 4π and their

phase difference with steering angle input are 0. When steering angle is positive, which

means vehicle turns left, both yaw velocity and lateral acceleration are positive, which

means the vehicle is rotating anti-clockwise and accelerate to the left. Things are opposite

when steering angle become negative.

29
 Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation

All vertical load versus time of 4 tires are sine wave with period of 4π. Vertical load

on tire 2 and 4 have same phase as steering angle input, while tire 1 and 3 have phase

difference of π. Because when vehicle turns left, centroid of gravity of vehicle moves to

right. It moves to left when vehicle makes a right turn. So, the vertical load on tire 2 and

4 always have the same phase as steering angle and increases with the increase of

steering angle, while load on tire 1 and 3 have an opposite phase and decrease with the

increase of steering.

The slip ratio of each tire is sine wave with period of 4π. Slip ratio of tire 1 and 3 have

same phase as steering angle input, while tire 2 and 4 have phase difference of π. Because

when steering angle is positive and vehicle turns left, tire 2 and 4 travel along a curve

than tire 1 and 3, so expected velocity of tire 2 and 4 are higher, however tire 1 and 2

have same rotational speed, which leads to tire 1 and 3 rotate faster than expected and

tire 2 and 4 rotate slower than expected. So, slip ratio is positive for tire 1 and 3 and it’s

negative for tire 2 and 4 when vehicle turns left. Things are opposite when steering angle
30
 Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation

is negative. We can also see that the absolute value of slip ratio increases with increase

of steering angle, this is because if steering angle increases, vehicle will make a sharper

turn and the turning radius of vehicle will get smaller, and smaller turning radius will lead

to a bigger difference between the expected velocity of tires on the right and tires on the

left. Imagine the turning radius is infinitely long, the vehicle is then very much like moving

straightly, and slip ratio is close to 0.

The slip angle of each tire is sine wave with period of 4π. And they all have a

difference of π. When vehicle turns left, negative slip angle will provide positive lateral

force on tire, which will lead to positive lateral acceleration. Things are opposite when

vehicle turns right.

Since the steering angle input is sine wave with 0 initial phase, so. in real world, the

vehicle turns left and right alternately. A series curves are shown in the trajectory as

expected. The vehicle moves at constant speed and steering is a sine wave, so the

trajectory changes regularly.

3.3.2 Conclusion
All the vehicle state variable output, including yaw velocity, lateral acceleration,
31
 Chapter 3 Vehicle Dynamic and Kinematic Modeling and Simulation

vertical load, slip ratio, slip angle and trajectory, are reasonable since their plot all

perfectly follow the vehicle dynamic and kinematic theory and their magnitude are also

reasonable.

3.4 Chapter summary

In this chapter, a 3-DOF vehicle dynamic model and a kinematic model are built to

have a more real research on more details of vehicle. The model is built in SIMULINK and

packaged into different subsystems. Assume the vehicle moves on a horizontal dry

cement road with constant speed, then simulation is done using two different steering

angle inputs. All the results are perfectly reasonable which means our model is rational

and can be used to study the controller in the next chapter.

32
 Chapter 4 Trajectory Tracking Control

Chapter 4 Trajectory Tracking Control

After getting the feasible lane changing route based on polynomial, it is necessary

to choose a control method to track the lane change path. PID controller is widely accept

in many practical applications. It’s independent of the established system model and the

control parameter can be obtained by trial and error method.

4.1 Trajectory tracking PID controller

Today’s closed-loop automation technology is based on the concept of feedback to

reduce the uncertainty. The elements of feedback theory include three parts:

measurement, comparison and implementation. The key to the measurement is the

difference between the actual value of the control variable and the expected value, which

is used to correct the system response and perform regulation control. In engineering

practice, the most widely used regulator control law is proportional, integral and

differential control, known as PID regulation

PID controller is common feedback loop components in industrial control

applications. It consists of proportional unit P, integral unit I and differential unit D. PID

control is based on proportional control. Integral control eliminates steady-state errors

but may increase overshoot. Differential control accelerates the response of large inertial

systems and weakens overshoot trends.

When design the lane change path in chapter 2, we make vehicle move at constant

speed to reduce the difficulty of control. According to the path planning based on mass

point model, the longitudinal velocity is constant, there is no need to implement control

in longitudinal direction. The only thing need to be done is to control the lateral

Expected lateral Vehicle State

displacement PID Steering angle dynamic variable
Expected
Trajectory controller &
kinematic
model

Actual lateral displacement

Fig 4.1 Structure diagram of control system

33
 Chapter 4 Trajectory Tracking Control

displacement to make it close to the expected path. The structure diagram of control

system is shown in Fig 4.1.

According to the diagram of control system, build the controller using SIMULINK and

connect it with the vehicle model established earlier, which is shown in Fig 4.2. Control

parameter is obtained by trial and error method.

Fig 4.2 SIMULINK model of controller

4.2 Simulation and result analysis

Create a m document using MATLAB, which can integrate all the work in this paper.

Run the document and first the users will be asked to give two inputs, current speed of

the vehicle and distance to the obstacle and here we assume this information are already

obtained by sensors. Then the optimal final longitudinal displacement is calculated using

genetic algorithm. Finally, the document will call the SIMULINK model we have built

earlier and transfer the current speed to the vehicle model and transfer the optimal path

to the controller. The controller will keep the vehicle moving along the path by adjusting

the steering angle.

We assume the current speed of the vehicle is 54km/h and there is an obstacle 50m

in front of it. Run the m document then a window for user input shows up. The pop-up

window is shown in Fig 4.3.

34
 Chapter 4 Trajectory Tracking Control

Fig 4.3 Pop-up window

Set the current speed at 15 m/s (54km/h) and distance at 50m then click “OK”. The

result are as follows:

According to the optimization, the optimal final longitudinal displacement is 53.38m.

The actual trajectory and expected trajectory overlap each other, which means the

controller can help the vehicle track the path accurately. Let’s look at more details.

35
 Chapter 4 Trajectory Tracking Control

Actual lateral displacement, velocity and acceleration all overlap the expected curve.

Final lateral displacement is exactly 3.5m and lateral acceleration is less than 2m/s2, which

satisfy the constraint.

The actual longitudinal displacement overlaps the expected curve, but both velocity

and acceleration have deviation. This probably because we assume the vehicle is moving

with constant longitudinal velocity, so we didn’t complement any control in that direction.

Since the magnitude of deviation is relatively small, we can still say the control is result is

good.

Steering angle is output of the controller as well as the input of the vehicle dynamic

and kinematic model. The controller measures the difference between the expected

displacement and actual displacement then adjust the steering angle to make the vehicle
36
 Chapter 4 Trajectory Tracking Control

moves along the expected path. So steering angle is the result that we are most interested

in. As we can see, the steering angle that the controller adopts is very close to a sine wave,

which means the vehicle turns left first and then right, just as expected. We also notice

that the steering angle curve is quite smooth, which means the control process and the

change of steering angle is continuous without sudden change. The steering angle

worked out by the controller is quite feasible.

4.3 Chapter summary

In this chapter, a controller based on PID regulation is introduced and designed.

Then the controller is built in SIMULINK and connected with the vehicle model. A m

document which integrate all the work earlier is created and simulation is run. The result

shows that the controller controls the vehicle accurately to make it move along the

expected trajectory. Displacement, velocity and acceleration in both lateral and

longitudinal direction are as expected. The change of steering angle is continuous and

smooth. All the result indicate that the controller works pretty well, which means the path

designed is feasible.

37
 Chapter 5 Summary and Outlook

Chapter 5 Summary and outlook

5.1 Summary

In this paper our study focuses on obstacle avoidance, especially lane change

situation, for autonomous vehicle. First, a comprehensive obstacle avoidance strategy is

studied and lane change path is designed based on mass point model and optimized.

Then, to study more real detail of vehicle, we build a vehicle dynamic model as well as a

kinematic model. The dynamic model is 3-DOF model, and it includes subsystem like

magic tire model. Two tests is done and the rationality of the model is verified. Finally, a

controller based on PID regulation is designed, and according to the simulation result,

the controller works pretty well, and all the result are very close to expectation which

means our path designed is feasible.

5.2 Outlook

This paper proposes some feasible path planning and control methods for lane

change problem of autonomous vehicle. However, there are still many aspects which

needs improvement, which are as follows:

(1) The paper only considers the situation with one static obstacle and assume the

other lane is clear and open. More complicated situation is not studied, so more research

is necessary to complement the strategy and path design.

(2) The lane change trajectory is designed assuming the longitudinal speed is

constant when vehicle changes lane. But we do encounter some situations where the

longitudinal velocity control is necessary. For example, when we deal with the

optimization, there are some results without optimal because the distance to the obstacle

is too small for vehicle speed in that situation. Then we do need to implement longitudinal

control to slow down the vehicle to make the lane change doable.

(3) All the work in this paper is limited to theory analysis and simulation. Many

features like steering linkage mechanism, wind resistance and tire resistance are ignored.

So, it is necessary to further carry out testing and other work on real vehicles. Only then,

will this project have real application and promotion significance.

38
 Reference

Reference
[1] Mark Campbell, Magnus Egerstedt, Jonathan P.How and Richard M.Murray Autonomous

driving in urban environment: approaches, lessons and challenges[J]. Philosophical Transactions

of the Royal society, 2010(9): 4649-4672

[2] Iakovos Papadimitriou, Masayoshi Tomizuka. Fast lane changing computaions using

polynomials[J]. Proceeding of the American control conference, June4-6, 2003

[3] Li Wei, Wang Jing, Duan Jianmin. Lane changing trajectory planning of intelligent vehicles

based on polynomials[J]. Computer engineering and applications, 2012, 48(3): 242-245

[4] Jasbir S.Arora. Introduction to optimum design[M]. The university of Iowa, college of

engineering, Iowa city, Iowa

[5] Zhisheng Yu. The theory of antomobile[M]. Tsinghua University

[6] PID controller. https://en.wikipedia.org/wiki/PID_controller[Z]. Wikipedia, 2018

39