Content uploaded by Hui Chen
Author content
All content in this area was uploaded by Hui Chen on Jul 16, 2014
Content may be subject to copyright.
A Behavior-Based Path Planning for Autonomous
Vehicle
CaiJing Xiu and Hui Chen
School of Automotive Studies, Tongji University, 1239 Siping Road,
Shanghai 200092, China
xcj2_11@163.com
hui-chen@tongji.edu.cn
Abstract. The path planning is one of the key issues of autonomous vehicle. In
this paper, a behavior-based method is used for path planning. This method can
manage complexity environment and is easy to design and test behaviors. An
autonomous electric vehicle is tested in a driving course including the behaviors
of road following, turning, obstacle avoidance and emergency braking. The test
results show that the path planning algorithm can run in a real-time, while the
unknown obstacles can be avoided and the autonomous vehicle can be driven
toward the destination in a smooth path and continuous motion.
Keywords: Autonomous vehicle, Driver behavior characteristics, Path planning,
Behavior-based
1 Introduction
Autonomous vehicle becomes a hot research topic due to the demand of the intelligent
transportation systems (ITS) and the military applications in the recent decades, while
the path planning is one of the key issues of the vehicle. The path planning can be
divided into the global path planning or navigation with fully known environmental
information and the local path planning with uncertain environmental information.
Goal-based path planning should avoid obstacles in a dynamic environment and reach
the destination automatically. In the study of path planning, a top-down traditional
method consists of the modules of sensing, modeling, planning and action. This
method is difficult to establish an accurate map of environment model, so it is difficult
to cope with real-time dynamic environment especially urban environment [1]. The
traditional method based on artificial potential field (APF) is improved for its
disadvantages such as local minimum [2][3][4], but APF does not consider some
constraints or regulations of road and traffic, such as stop line and intersection. It
means that APF can not be used to deal with all traffic in urban environment.
Intelligent control methods, such as fuzzy control [5][6], neural networks[7],
swarm[8][9]and so on, are applied in the study of path planning, but most of them are
remained in simulation study or used in specific unstructured environment.
Supported by Tongji University, Shanghai 200092, hina, TRW Automotive R&D
(Shanghai) Co., Ltd. and Science and Technology Commission of Shanghai
Municipality(07dz05901)
It can be seen that the current path planning methods have the disadvantages mainly in
real-time performance, universal property and flexibility for different traffic conditions,
especially for urban environment. In consideration of the above issues, the
behavior-based path planning is applied for autonomous vehicle in this paper. The
behavior is decomposed into a series of relatively independent small behaviors and the
behavior decision decides the behavior of autonomous vehicle by the environmental
characteristics [10][11].
2 Architecture of Autonomous Vehicle
In the process of driving, the structure of the driver's brain activity can be divided into
four layers: destination planning, behavior decision, behavior planning and operation
control [12], as showed in Fig.1. Destination planning layer is the first activity of the
driver to be carried out according to current vehicle status (including the location and
pose), driving destinations and pre-stored road map. It is global path planning of
autonomous vehicle related to all possible routes from the current status to destination
location considering some constrains such as shortcut, optimal time and so on. A series
of sub-goals are outputted from this layer [12]. Behavior decision layer is carried out
based on the environment characteristics, including vehicle status, traffic rules and
other information in the memory in order to determine the vehicle's driving behavior.
Behavior plan layer is the planning of driving trajectory and vehicle speed according to
driving behavior which is outputted by behavior decision layer. Operational control
layer is the operation of vehicle driving, braking, steering,that istracking vehicle
trajectory which is given by behavior planning layer.
Fig.1. Driver's Thinking Layer
From the view point of controlling, the vehicle is the controlled object, the driver is the
sensor and controller, while the road and traffic regulations are the constraints of the
system in the process of driving. Driving safety is the basic requirements of the system.
From the above analysis, it can be found that the design target of an autonomous
vehicle is to replace the driver by the sensors and controllers, and to meet the
requirements of various constraints and conditions.
The ultimate goal of autonomous vehicles is to replace the driver from the closed-loop
control system of "driver - vehicle - road". Autonomous vehicle control system showed
in Fig.2 consists of four parts in this article: path planning, vehicle control that is the
trajectory and speed tracking, vehicle and environment. The path planning of this
closed-loop control system is studied in this paper.
Fig.2. Closed-Loop Control System of Autonomous Vehicle
3 Design of Path Planning Controller
The behavior-based planning method is applied to manipulate the unmanned vehicles
in uncertain environmental information. Behavior decision is made at each sampling
time according to real-time sensor data, the results of global and local path planning,
the distance and relative velocity between the autonomous vehicle and obstacles.
Autonomous vehicle behaviors are classified into road following, turning, obstacle
avoidance, emergency braking, ACC, merging into traffic flow, intersection crossing
and so on. In this paper, the first four behaviors of road following, turning, obstacle
avoidance and emergency braking are discussed.
3.1 Road following
When there is no obstacle or the distance between autonomous vehicle and obstacle is
greater than a safe distance, the road following behavior is triggered.
The control algorithm of road following behavior is based on preview following
theory [13] combined with PID control. The control model is shown in Fig.3.
Fig.3. Road Following Control Model
In this control model, V is the velocity of autonomous vehicle, (X,Y) is the current
coordinates of autonomous vehicle, they are the input of controller system. The
preview distance is given by the equation: S = K * V, while K is a proportional
coefficient and V is the velocity of autonomous vehicle. Steering angle is calculated
according to the lateral deviation between vehicle current coordinates and preview
point coordinates. The vehicle constrain module is the vehicle dynamics and
kinematics constraints such as the maximum steering angle and steering angular
velocity, etc.
3.2 Turning behavior
When the autonomous vehicle encounters an intersection, turning behavior is
triggered. Because of the vehicle dynamics and kinematics constraints, the curvature
of tracking of autonomous vehicle must be bounded and continuous, and the
derivative of curvature is also bounded. Therefore, behavioral strategy of turning is
calculated by polar polynomial of Nelson [14] [15].
General Nelson polarity polynomial expression is:
()
2345
012345
raaaaaa
ϕ
ϕϕ ϕ ϕ ϕ
=+++++
(1)
r is the polar radius,
ϕ
is the polar angle, i
a(i=1,2……)is polynomial coefficients,
they will be given in the following equation (4).
The constraint of location, slope '
r and curvature
κ
are:
,0,0 0
,0,0
rRr
rRr
κϕ
κϕ
′
=== =
⎧
⎨′
=== =Φ
⎩ (2)
As the curvature of the path is:
()
()
22
3
22
2
2rrrr
rr
κϕ
′′′
+−
=′
+
dr
rd
ϕ
′=
2
2
dr
rd
ϕ
′′ =
(3)
From equations of (1) (2) (3), the equation (4) can be obtained.
0
1
2
3
42
5
0
2
2
0
aR
a
R
a
R
a
R
a
a
=
⎧
⎪=
⎪
⎪=
⎪
⎪
⎨=−
⎪
Φ
⎪
⎪=
⎪Φ
⎪=
⎩ (4)
To substitute the parameters of equation (1)by equation (4), the expression will be:
()
23 4
2
122
rR
ϕϕ ϕ
ϕ
⎛⎞
=+−+
⎜⎟
ΦΦ
⎝⎠
(5)
Fig.4. Turning Control Model
The control model is shown in Figure4. R is maximum turning radius when turns,
road angle (the Φ in the formula) is the angle between the current road and the next
driving road of autonomous vehicle, LOOR is an sign from global path planning that
inform turn right or left. Angle calculation subsystem calculated front wheel angle by
equation: angle=arctan(L/r), L is the length of autonomous vehicle, slip angle is
ignored in this equation.
3. 3 Obstacle Avoidance
When the distance between autonomous vehicle and the obstacle is less than safe
distance, the obstacle avoiding behavior is triggered. Obstacle avoidance behavior is
achieved by an improved artificial potential field method.
In the structured environment, the road boundary can be considered as a repulsive
force field. There is no target point defined in the traditional artificial potential field
method. In this paper, the distance between autonomous vehicle and target point is
considered in the repulsive potential field [2][3].
Gravitational potential function is follows:
2
at goal
1
U(X)= k(X-X )
2 (6)
K is the location gain coefficient, goal
(X-X )is the distance between autonomous
vehicle and target point, the corresponding attractive potential field is the negative
gradient of target point.
()-( )
at
g
oal
FX kXX=−
uv (7)
Improved repulsive potential field function is as following:
2
re goal o
oo
o
111 111
U (X)= ( - )+ ( - )(X-X )
22
0 >
η
ηρρ
ρρ ρρ
ρ
ρ
≤
(8)
η
is the location gain coefficient of repulsion,
ρ
is the distance between autonomous
vehicle and obstacle, o
ρ
is the safety threshold of autonomous vehicle. When the
distance of autonomous vehicle and obstacle is greater than the safety threshold, and
the repulsive function is zero. Repulsion function is obtained by following equation:
2
goal o
22
oo
o
11 1 11 1
(X)= ( - ) + ( - ) (X-X )
0 >
re
F
η
ηρρ
ρρ ρ ρρ ρ
ρ
ρ
≤
uv
(9)
The force of the autonomous vehicle is:
sum at re
FF F=+
∑
uv uv u u v (10)
Fig.5. Sensor Fusion
Figure5 is the input of APF method,(Xob,Yob) is the obstacle coordinate that is given
by processing of radar data;(Xgo,Ygo) is the goal point coordinate that is given by
processing of camera data;(X,Y) is the autonomous vehicle coordinate that is given by
GPS. goal
(X-X ) and
ρ
will be known with the formula of the Length of the line
between two points. Then the Fre and Fat are ensured, because the input of
autonomous vehicle must be steering angle, according to the orientation of
autonomous vehicle, Fre and Fat will be divided into two pitch Fx and Fy, and the
steering angle is calculated by the following formula: angle=
arctan( / )Fy Fx
∑∑. Obstacle avoidance control Model is shown in Fig.6.
Autonomous vehicle will avoid the obstacles when the front obstacle hinder the travel
of autonomous vehicle, and meet road constraint according to this method.
Fig.6. Avoid Obstacle Model
3.4 Emergency Braking Behavior
When the autonomous vehicle can not avoid collision by change the steering and
deceleration, emergency brake behavior is triggered.
4 Experiment
In order to verify the behavior-based path planning algorithm, a real vehicle
experiment is performed.
4.1 Experiment Condition
The autonomous electric vehicle shown in Fig.7 drives along a rectangular route
while a static or moving obstacle will emerge randomly. The vehicle speed is
controlled at 10km/h for this experiment. The environment perception sensors are
lidar, camera, GPS and so on.
Fig.7. Experiment Platform
4.2 Experiment Result
The trajectory of autonomous vehicle is shown in Figure 9. According to the safety
threshold, safety threshold is calculated as S = K * V, while the coefficient K is
related with the dynamics of vehicle and the communication speed of sensor, and is
set up as K = 8. When obstacles emerge, emergency braking behavior or collision
avoiding behavior will be carried out.
Fig.8. Experiment Tracking
5 Conclusions
In this paper, a behavior-based path planning method is applied for autonomous
electric vehicle. The experiment results show, 1) The applied path planning method is
effective for a real-time dynamic environment. 2) The behavior decision can perform a
correct behavior selection only with the information of vehicle status and surrounding
environment characteristics. The flexibility and real-time performance of the control
system is ensured. 3) Driving behaviors can be successfully decomposed into several
elemental behaviors. It shows the applied path planning method could be adapted to an
urban environment with complex driving behaviors. The further experiment studies
will be performed to verify the adaptability of urban driving condition.
References
1. YiBin Zhang,CaiHong Zhang, “The Study of Mobile Robot Path Planning in Uncertain
Enviroment”, Journal of Hangzhou Electronic Science and Technology University,2005
2. WANG Meng,WANG Xiao-rong, “Study of local path planning of mobile robot based on
improved artificial potential field method”, Computer Engineering and Design,2008
3. SHEN Kui-hua,TAN Yue-gang, “A Method Solving the Local Minimization Problem of
Artificial Potential Field”,Journal of Wuhan University of Technology,2005
4. David Clinton Conner,“sensor fusion,navigation,and control of autonomous vehicles”,
Virginia Polytechnic Institute and State University, Thesis for the degree of Master.
5. Neil Eugene Hodge and Mohamed B. Trabia,“steering fuzzy logic controller for an
autonomous vehicle”, Proceedings of the 1999 IEEE International Conference on Robotics
& Automation.
6. EI Hajjaji,Oulasine,“modeling human vehicle driving by fuzzy logic for standardized ISO
double lane change maneuver”, Proceedings of the 2001 IEEE Symposium on Robot and
Human Interactive Communication.
7. Istvan Engedy,Gabor Horvath ,“Artificial Neural Network based Mobile Robot
Navigation” 2009 IEEE International Symposium on Intelligent Signal Processing
8. Sheetal Doctor,Ganesh K.Venayagamoorthy,“Unmanned Vehicle Navigation Using Swarm
Intelligence”2004 IEEE
9. Alejandro Ramirez-Serrano,Giovanni C.Pettinaro,“Navigation of Unmanned Vehicles
Using a Swarm of Intelligent Dynamic Landmarks”2005 IEEE
10. D. Langer, J. K. Rosenblatt, and M. Hebert, “A Behavior-Based System For Off-Road
Navigation” The Robotics Institute Carnegie Mellon University
11. Tobias Gindele, Daniel Jagszent, “Design of the Planer of Team AnnieWAY’s
Autonomous Vehicle Used in the DARPA Urban Challenge 2007” 2008 IEEE Intelligent
Vehicles Symposium.
12. Hangen He, “The Study of the Improvement and Technology of Intelligent Vehicle”
Defense Technology university.
13. GAO Zhen-hai,GUAN Hsin,GUO Kong-hui.Driver directional control model and the
application in the research of intelligent vehicle[J].China Journal of Highway and
Transport,2000,13(3).
14. W. Nelson, Continuous-Curvature Paths for Autonomous Vehicles, IEEE Int. Conf. on
Robotics and Automation, May 1989, Scotsdale, Ariz.,vol.3, pp.1260-1264,1989
15. O. Pinchard, A. Liegeois and F. Pougnet, Generalized polar polynomials for vehicle path
generation with dynamic constraints, In Proc. of the IEEE Int. Conf. on Robotics and
Automation, volume 1, pages 915–920, Minneapolis, MN (US), April 1996