Autonomous Vehicle Parking Using Artificial Intelligent Approach

Chen-Kui Lee
Department of Electrical Engineering, National Chung Hsing University, Taichung, Taiwan, R.O.C water73429@hotmail.com

Chun-Liang Lin
Department of Electrical Engineering, National Chung Hsing University, Taichung, Taiwan, R.O.C chunlin@nchu.edu.tw

Bing-Min Shiu
Department of Electrical Engineering, National Chung Hsing University, Taichung, Taiwan, R.O.C 49264032@ee.nchu.edu.tw
AbstractThis paper devotes to the design and implementation of a hybrid artificial intelligent control scheme for a car-like vehicle through the performance optimization of the task of car parking. The genetic algorithm is used to determine the feasible parking locations. The Petri-net is used to replace the traditional system flow chart and most importantly, to plan alternative parking routes especially in a global space. A fuzzy controller is utilized to drive the car along the optimal parking route. Keywords-Optimization, Genetic Algorithm, Petri net, Fuzzy Control, Parallel Parking

With regard to the problem of autonomous parallel parking, several factors must be considered so that some of the platform technology developed for mobile robot systems industry can be easily extended to parking applications. Hwang et a1[17] have proposed that Petri nets are well-suited to the modeling of concurrent systems and [18-20] propose many analytic techniques allowing Petri nets to verify the occurrence of potentially undesirable states. This paper presents an automatic parallel parking planning system for a car-like vehicle, with the intention of applying the research results of robot motion planning into real-world applications. Unlike traditional design methods, the proposed method focuses primarily on the path planning in a global parking region and it combines with Petri nets and genetic algorithms to present an optimal car parking scheme. The overall scheme solves the parallel parking problem and extends the problem to larger parking space. The presented strategy can be used in parking aid devices, and it has the potential to be integrated into automobiles. II. ARCHITECTURE OF PARKING SYSTEM

I.

INTRODUCTION

The study of motion autonomy for various vehicles has always been a significant point of interest for researchers globally. A classical approach for solving the motion problem is to apply Dubins curves [1] or Reed & Sheep curves [2]. Lafferiere and Sussmann [3] presented the motion plan for vehicles based on a constructive proof of controllability. Marry and Sastry [4] approached the problem by steering a nonholonomic system by means of sinusoids between arbitrary points. However, neither of them addressed the issue of obstacle avoidance. Paromtchik and Laugier [5] presented an approach where a vehicle was driven along a sinusoidal path in reverse direction. The parking space is scanned in advance before the vehicle reverses into it. Recently, intelligent control methods are used more extensively, such as fuzzy logic controller, neural network controller, and genetic algorithms. Yasunobu and Murai [6] proposed a human experience based fuzzy logic controller. Jenkin and Yuhas [7] introduced a simplified neural network controller trained on the basis of kinematics data. These methods have also been extended to applications in autonomous mobile robot control and yielded much success [816].

A. Problem Description Automatic parking is defined as an autonomous car from a traffic lane maneuvering into a parking lot to perform parallel, perpendicular or angle parking. The automatic parking aims to enhance the safety and comfort of drivers and passengers while driving in a restrained area where much attention and experience is required. The parking maneuver is achieved by means of coordinated control of the steering angle and speed, which takes the actual circumstances into account to ensure collision-free motion within available space.

D. Ready-For-Parking Space While an actual parking environment is considered, not all the trajectories are feasible. To avoid collision, the vehicle must be positioned inside a suitable region before it can perform the parking procedure. Therefore, the preliminary task is to determine a suitable region from two other constraints. One is the limitation of the maximum curvature given by

Figure 1. Car-like vehicle.

TABLE I.

DEFINITIONS OF VARIABLES

k ( x) =

y [1 + ( y) ]
3 2 2

kmax

(2)

(x f , y f )

Position of the center of front wheels Position of the center of rear wheels Orientation of the steering wheels with respect to the frame of the vehicle Angle between vehicle frame orientation and X-axis Wheelbase of vehicle Center of curvature Distance from point O to the midpoint of front wheels axle Curvature radius; Speed of front wheels Curvature of trajectory

( xr , yr )

l O

The other is the distance from the car to the obstacle point ( xob , yob ) , as illustrated in Fig 2. According to the vehicles dimensions as defined in Fig. 3, we can solve the values of ( xrr , yrr ) and ( x fr , y fr ) from the reference point ( xref , yref

) produced by the reference trajectory.

sin cos
sin cos

r v k

W xrr = ( xref L2 cos ) + 2 y = ( y L sin ) W ref 2 rr 2

(3)

B. Modeling of theCcar-LikeVehicle Consider the kinematical model of the car-like vehicle shown in Fig. 1, where the rear wheels are fixed parallel to the car body and are allowed to roll or spin but disallowed to slip. This ensures that the rear wheels are always tangent to the automobile orientation. The corresponding parameters of the vehicle listed in Table I are defined as follows
The rear wheel is always tangent to the automobile orientation. Under normal conditions, the reserved velocity is about 5km per hour slow; thus, we can suppose that it appears a no-slipping condition and the velocity of the rear wheel in vertical director must be zero. This so-called nonholonomic constraint is described as

W x fr = xref + ( L1 + l ) cos + 2 y = y + ( L + l ) sin W 1 2 fr ref

(4)

xrr (t ) xob and yrr (t ) yob , the rear-right corner of the vehicle would hit the obstacle. Similarly, if x fr (t ) xob
If and

y fr (t ) yob , the front-right corner of the vehicle would

hit the obstacle. Through this test, the suitable region could be characterized as illustrated in Fig. 4.

xr = v cos cos yr = v sin cos

(1)

=v

sin l
Figure 2. Car parking with obstacle avoidance; coordinates of the left-upper corner of the obstacle are ( xob , yob )

This is used to generate the next backward state position of the vehicle when the present states and control input are given.

C. Reference Trajectories forParallel Parking It is important to provide a reasonable reference path so that the car can successfully accomplish the parallel-parking mission. In order to implement a feasible and smooth trajectory for the rear reference path, a method based on the derivation of a fifth-order polynomial [9] has been well adopted.

L1

W
( x fr , y fr )

P51

l
L2
W 2

( xref , yref )
T6

T6 P1 P2 T1 P6 T4 T3 P4 T6 P7 P8 T5 P9

T1

( xrr , yrr )

P3

T6

L1

TG T5 P5 PG

Figure 3. Definitions of the key variables.

T4 T3

15
feasible region collision with obstacle curvature > max k5 distance < W/2

(b) flow chart Figure 5. Petri net flow chart.

10 Y (meters)

A. Optimal Path Planning with GA In this section, we use the well known genetic algorithm (GA) to determine an optimal parking path.
0 5 X (meters) 10 15

Figure 4. Different initial positions for feasible and unfeasible vehicle parking regions.

While designing an optimal path for vehicle parking, there are various performance indices that are generally considered. In the proposed method, an additional factor is placed into consideration, that is, the number of gear changes. In general, a parking procedure that requires more than two gear shifts would be considered inefficient.
Distance: In the process of parking a vehicle, the relative position of the parking spot with respect to the car is a very important factor. Therefore, it is essential to consider the forward and backward movement when weighing the distance. Curvature: The curvature of a vehicle directly affects the rotary angle of the wheels. The maximum value usually doesnt exceed 45 degrees. An overlarge curvature for a vehicle in motion could cause the passengers a physical discomfort or even dizziness. Number of gear changes: Generally, when considering the parking process, the procedure requires more than simply reversing a vehicle into the desired parking spot. The number of gear shifts needs to be considered as another important factor. More gear shifts will result in more extended distances; therefore, the corresponding cost should also be minimized.

III.

PETRI NETS AND GAS

For human drivers, especially the beginners, it is usually difficult to determine a suitable parking space. Most often, the decisions are made instinctively. The parking process is a discontinuous action involving moving forward and reverse; therefore, different actions must be taken to find a suitable path for an optimal solution. Petri nets based on discrete event formalisms have been proved as a potential means of achieving collision avoidance, task-preserving human-intervention, and route controlling. We has utilized an effective method with the concept of Petri nets while solving this particular problem. Petri net flow chart with initial and target markings is developed as shown in Fig. 5. P P9 are defined as nine subareas in this parking field. 1

According to the performance indices previously specified, the fitness function for the optimization problem can be designed. (5)
fitness = 1 W1 D Ng + W2 + W3 D max Ng max

PG

i =1

i =1

i max

(a) Definition of P P9 1

where

Wi : weighing scalar

D : = trajectoryi total distance moved

i =1

and the horizontal angle

between the coordinates of

( x, y ) and ( xG , yG ) shown as in Fig. 7.

From which once can have

Ng : number of gear changes

i : orientation of the steering wheels with respect to the frame of

the vehicle of the i-th moving segment

The design of the optimal parking path for a vehicle illustrated in Fig. 6 is summarized as the following steps. The initial position ( x0 , y0 ) is randomly selected first.
Step1:

G = tan 1

yG y xG x

(6)

d e = ( xG x) 2 + ( yG y ) 2 u1 = G c
u2 = f G
where

(7) (8) (9)

Search for the point ( x p , y p ) randomly within each P5-position, p = 1, 2, , n . (Randomly generate n chromosomes on the initial population.)

Step2

apply these points and initial position to the fifth order equation and make sure that the curvature k does not exceed k max . If

de is the distance between vehicle and goal.

k ( x) kmax , the path allows the vehicle

to navigate to P5; trajectory1 is produced. If k ( x ) > k max , it is impossible for the vehicle to navigate directly to P5; therefore, the vehicle should be navigated to a different point where it is able to perform the maximum curvature k ( x ) = k max .
Step3: Step4:

We propose a two-stage fuzzy logic controller shown as in Fig. 8 to command the steering angle of the front wheels for parking task. Stage1 control input variables are u1 and u2 . The output variable is listed in Tables II-III. We introduce

e . The proposed fuzzy control rules are

de and e as the input linguistic variables for the Stage2 controller. d e is in the range of distance [ 0,100 ]
(m) and

Apply ( x p , y p ) and
trajectory0.

( xG , yG ) to obtain

is in the range of angle [ 180,180 ] (degrees). The

output linguistic variable is the steering angle actual condition of the vehicle, the [ 45, 45 ](degrees).
Y trajectory1 P5 trajectory2

. Consider the range of is

Find the number of gear changes. For example, the vehicle moved backward, forward, then backward, results in three gear changes. Find the change of . Calculate the fitness and select the better chromosomes with probabilities based on fitness. If the process will be terminated if the stopping condition is reached or the optimal solution is achieved, otherwise, repeat steps 2-7 for another run. IV. FUZZY CONTROL DESIGN

Step5: Step6:

PG
trajectory0 X

Step7:

Figure 6. Illustrations for parking trajectories.

f
( xG , yG )

In this section a human-friendly controller is implemented. We put our focus on designing the tracking controller for the vehicle to complete the task of parallel parking.
A. Problem Description & Analysis
( x, y )

Suppose the goal position defined by ( xG , yG , f ) , the vehicles current position ( x, y ) , the orientation of the vehicle

Figure 7. Target tracking control system with terminal orientation angle

f .

u1
Eqs. (6)-(9)
+

u2

Fuzzy controller 1 (Steering control)

e
Fuzzy controller 2 (motion control)

75

Plant

( x, y , )
y (meter)

70 65 60 55 50 45

de
Sensor

40 35

Figure 8. TABLE II.

Two-stage fuzzy logic controller. FUZZY RULES FOR STEERING CONTROL

30

25 15

20

25

30

35 X (meter)

40

45

50

55

u1

Figure 9. Experiments with (left ) and without (right) parameter optimization, initial position (20,30, 50) and terminal position (50, 70, 60) .

u2

(a) Membership function of TABLE III.

u1

FUZZY RULES FOR MOTION CONTROL

de

e
(b) Membership function of

u2

The objective of the controller is to make d e , u1 , and u2 converging to zero when t . As a result, the vehicle would just need to go straight to reach a specified goal. V. SIMULATION RESULTS

The simulation results presented here were based on a carlike platform (Toyota Camry) with the following dimensions: length 4.825 m, width 1.82 m, l = 2.755 m and max = 45 degrees. Each parking space is supposed to be of dimension 6.5 m 2.5 m . In case 1, the vehicle is with the starting position characterized by (20,30, 50 ) and the goal point, (50, 70, 60 ) . The experiments show the performance with and without parameter optimization. The resulting trajectories of the simulation results are illustrated in Fig. 9. The membership functions for the fuzzy controller, after optimization, are shown in Fig. 10

(c) Membership function of

Figure 10. Membership functions after optimization.

After optimizing the controller, the focus is switched to tracking control of the reference path. In Fig. 11, the working efficiency for the trajectory tracking of the reference path is shown. Next, we apply the proposed path optimization method and the fuzzy controller for finding the optimal parking path in a widespread parking lot. We set the generation=50 and the population=20; the corresponding results are shown in Fig. 12.

[3]

[4]

[5]

[6] Figure 11. Actual parallel parking trajectory with initial condition (10, 4, 10) and terminal condition (0, 0, 0)

[7]

[8]

[9]

[10]

Figure 12. Simulation result for random selection 2

[11]

VI. CONSLUSION In this paper, a complete parking mechanism for autonomous car-like vehicles has been introduced and applied to solve the parallel parking problem. The architecture includes a Petri-net based, off-line global trajectory planner, a decisive kernel based on GAs, and a trajectory tracking controller. The advantage of the proposed method is that it yields an effective parking path and further extends the controlled workspace to a larger region. The simulation results have illustrated the effectiveness of the preliminarily developed algorithms. A realworld application of the proposed design is currently undergoing. An integrated test for the autonomous parking planner with the trajectory tracking controller to verify effectiveness and applicability of our proposed approach will be appeared in the near future. ACKNOWLEDGMENT This research was sponsored by National Science Council, Taiwan, R.O.C. under the Grant NSC-96-2221-E-005-038. REFERENCES
[1] L.E. Dubins, On curves of minimal length with a constraint on average curvature, and with prescribed initial and terminal positions and tangents, American Journal of Mathematics, vol. 79, pp. 497-516, 1957 J.A. Reeds and L.A. Sheep, Optimal paths for a car that goes both forwards and backwards, Pacific Journal of Mathematics, vol. 145, pp. 367-393, 1990

[12]

[13]

[14]

[15]

[16]

[17]

[18]

[19] [20]

[21]

[2]

G. Lamfferriere and H. Sussmann, Motion planning for controllable systems without drift, Proceeding of the IEEE International Conference on Robotics and Automation, vol. 2, pp. 1148-1153, April 1991 R. Murray and S. Sastry, Steering nonholonomic system using sinusoid, Proceeding of the IEEE International Conference on Decision and Control, pp. 2097-2101, December 1990 I.E. Paromtchik and C. Laugier, Autonomous parallel parking of a nonholonomic vehicle, Proceeding of the IEEE International Conference on Vehicle Symposium, pp. 13-18, 1996 S. Yasunobu and Y. Murai, Parking control based on predictive fuzzy control, Proceeding of IEEE International Conference on Fuzzy system, pp. 1338-1341, Junurary 1994 R.E. Jenkins and B.P. Yuhas, A simplified neural network solution through problem decomposition: the case of the truck backer-upper, IEEE Transaction on Neural Network, vol. 4, pp. 718-720, July 1993 K. Nishimori, S. Hirakawa, and H. Tokutaka, Fuzzification of control timing in driving control of model car, Proceeding 2nd IEEE Conference on Fuzzy System, vol. 1, pp. 297-302, April 1993 D. Lyon, Parallel parking with curvature and nonholonomic constraint, Proceeding of IEEE Intelligent Vehicles Symposium, pp. 341-346, July 1992 K. Watanable, J. Tang, M. Nakamura, S. Koga, and T. Fukuda, A fuzzy-Gaussian neural network and its application to mobile robot control, IEEE Transaction on Control Systems Technology, vol. 4, pp. 193-199, March 1996 K.S. Tang, K.F. Man, Z.F. Liu, and S. Kwong, Minimal fuzzy memberships and rules using hierarchical genetic algorithms, IEEE Transaction on Industrial Electronics, vol. 45, pp. 162-169, February 1998 Y.S. Zhou and L.Y. Lai, Optimal Design for fuzzy controller by genetic algorithm, IEEE Transaction on Industrial Applications, vol. 36, pp. 93-97, January 2000 S.H. Kim and C. Park, A self-organized fuzzy controller for wheeled mobile robot using an evolutionary algorithm, IEEE Transaction on Industrial Electronics, vol. 48, pp. 467-474, April 2001 S.J. Chang, Autonomous fuzzy parking control of a car-like mobile robot, IEEE Transaction on Systems, Man and Cybernetics, Part A, vol. 33, pp. 451-465, July 2003 C.S. Chiu and K.Y. Lian, Fuzzy gain scheduling for parallel parking a car-like robot, IEEE Transaction on Control Systems Technology, vol. 13, pp. 1084-1092, November 2005 M.M. Suruz and G. Wail, Intelligent parallel parking of a car-like mobile robot using RFID technology, ROSE 2007 International Workshop on Robotic and Sensors, pp. 1-6, October 2007 F. Hwang, J. King, R. Hale, and M. Rokonuzzaman, Discrete event control of mobile robotic systems with potential applications in healthcare, Proceeding of IEEE 22nd Annual EMBS International Conference, pp. 431-435, July 2000 C.G. Looney, Fuzzy petri nets for rule-based decisionmaking, IEEE Transaction on Systems, Man and Cybernetics, vol. 18, pp. 178-183, February 1988 T. Murata, Petri nets: properties, analysis and applications, IEEE Proceedings, vol. 77, pp. 541-580, April 1989 S.I. Ahson, Petri nets models of fuzzy neural networks, IEEE Transaction on Systems, Man and Cybernetics, vol. 25, pp. 926-932, June 1995 J.C. Canales, L. Xiaoou, and Y. Wen, Fuzzy knowledge learning via adaptive fuzzy petri net with triangular function model, The Sixth World Congress on Intelligent Control and Automation, vol. 2, pp. 4249-4253, 2006