2024 24th International Conference on Control, Automation and Systems (ICCAS)

Jeju Shinhwa World, Jeju, Korea. Oct 29 - Nov 1, 2024

Simulation Study of Autonomous Vehicle with Restricted Sensor Network for Autonomous
Valet Parking System
Chanho Lee1, Junghwan Gil1, Chulsoo Kim1*
1
Korea Automotive Technology Institute,
Chungnam 31214, Korea ([email protected])

Abstract: In this study, we propose the simulation study of autonomous vehicle with restricted sensor network for
Autonomous Valet Parking System (AVPS). Various sensors are installed in autonomous driving vehicles. During the
AVPS scene, short range objects were hard to detect by using the installed autonomous driving vehicle’s sensors. For
driving and parking sensor network are required for different detection area. The restricted sensors for the AVPS is used
as achieving position data of vehicle such as heading angle and position data. The parking algorithm is implemented in
specific scene as vertical parking. The scenario and situation of this paper were demonstrated via CARLA simulation.
The CARLA simulation was chosen by flexibility of vehicle and sensors. The flexibility makes our scenarios accurately.
Simulation results were shown that vehicle parking operation is successful.

Keywords: Autonomous Valet Parking System (AVPS), Path Planning, CARLA, Autonomous Vehicle

1. INTRODUCTION chapter 2, we explain the untact platform for untact store

project vehicle. The sensor network is explained in
Recent research in Autonomous Driving Systems chapter 2. The three stage of AVPS scenario is shown in
(ADS) has been continuously advancing. ADS is capable chapter 3. In chapter 4, the parking algorithm of AVPS is
of operating in various environments, including explained via pseudo code. The simulation setup and
highways and city streets. The various researches have results are shown in chapter 5. Finally, the conclusion
been conducted on ADS in highway settings [1,2]. and future works of this paper is illustrated in chapter 6.
Similarly, numerous studies have focused on city driving
environments [3,4].
The Untact Store Project (USP) aims to sell goods
2. UNTACT PLATFORM
without physical contact between people. This project is
particularly necessary in situations when physical contact UADV is designed for USP, focusing on ADS. The
is impossible, such as during pandemics. Even without base vehicle is the KIA EV Bongo, characterized by its
pandemics, USP can be utilized in areas such as using long length compared to passenger cars. This commercial
conventional retail stores is difficult to access. vehicle (as shown in Table.1 and Fig.1) chosen to
USP researches ADS in city streets for selling goods accommodate a large service cabin necessary for the
without physical contact between people. USP aims to mobile retail store.
allow consumers to summon the vehicle to their location Table 1. Kia EV Bongo spec (mm)
and perform desired actions, such as selling specific
products or providing simple meals. Length Width Tread Wheelbase
The Untact Autonomous Driving Vehicle (UADV) F : 1490
5115 1740 2810
project had focused on enabling ADS on city streets. R : 1340
However, the previous UADV research aimed at
stopping on the roadside to sell items to a single
consumer, without the Autonomous Valet Parking
System (AVPS). As the number of users increases and
situations require more time for simple cooking, UADV
require parking, not stopping. For those purpose, AVPS
research is required for UADV.
In this study, we propose the simulation study of Fig. 1. Kia EV Bongo spec
autonomous vehicle with restricted sensor network for
AVPS. Various sensors are installed in UADV for
autonomous driving. However, short range objects were
hard to detect by using the installed UADV sensors. For
driving and parking sensor network are required for
different detection area. The installed two Differential
Global Positioning System (DGPS) sensors are
affordable for activating AVPS using heading angle and
position data. This paper is structured as follows. In
Fig. 2. Modified Appearance of UADV

978-8-9932-1538-0 ©2024 ICROS 131

Authorized licensed use limited to: R V College of Engineering. Downloaded on April 25,2025 at 06:52:12 UTC from IEEE Xplore. Restrictions apply.
 The modified UADV (as shown in Fig. 2) includes a range objects. However, Ultrasonic sensors can only
service cabin. Service cabin equipped with a multi- provide distance information. Additionally, without
jointed robotic arm capable of selling specific products walls or other objects, it is challenging to determine the
and providing simple meals. vehicle's global position. AVM cameras provide all
The ADS sensors installed on UADV include: surrounding information as vision data. However, vision
information is easily exposed to significant distortion by
Table 2. Equipment sensor on UADV fisheye camera. Therefore, it is difficult to measure
accurate distances using vision information alone.
Number of mmWave radar is designed to explore closer areas
Type Name
sensor- compared to conventional radar [7,8]. mmWave’s
Lidar Hesai 6 detectable minimum range is approximately 10cm. That
Continental range required for operating AVPS. Additionally,
Radar
ARS-408-21 mmWave providing basic information about objects
Mobileye 1 (Dynamic object or static object). However, the installing
Camera
AVM 8 mmWave radar solely for AVPS is not affordable reason
GPS MGI- 2000 2 of cost and space.
ADS dSPACE SCALEXIO 1 Due to internal constraints within USP, installing
additional sensors are hard to mounted on UADV.
Therefore, this study aims to acquire data under limited
conditions. Among the sensors on UADV, the AVM
camera and two DGPS units must be used for AVPS.
The AVM vision data is not sufficient to obtain
accurate distance information. Thus, the primary sensors
is the two DGPS units. DGPS is an improved version of
GPS with reduced errors of less than 3cm(Horizontal
±10mm, Vertical ±10mm). The principle operates as
Fig. 3.a. Sensor for ADS follows [9]. However, DGPS can obtain accurate
distance only outside. UADV always operates outdoors.
UADV is not designed to operate indoors. So, UADV can
obtain accurate distance data. The two DGPS units on
UADV are installed at the vehicle's center and at the
center of the rear axle. Using the two positions obtained
from the two DGPS units, the vehicle's precise position
and heading angle could be determined.
Fig. 3.b. Equipment Sensor position at UADV In the USP scenario, ADS and AVPS operate within a
confined space, and it is assumed that UADV has an HD
The UADV is equipped with various sensors designed map. Thus, the UADV can park using only the heading
for safe ADS (as shown in Table 2 and Fig. 3.a,b). and position angle. This study implements AVPS using
However, UADV sensors are not entirely sufficient for the position and heading angle obtained through DGPS.
obtaining the data required for AVPS. ADS operates at
medium-to-high speeds, and as a result the most
significant information for ADS is data on mid-to-long 3. AUTONOMOUS VALET PARKING
range objects. [5,6]. Therefore, the sensors on UADV SCENARIO
are optimized for collecting data on objects in mid-to- The scenario for the AVPS consists of three stages.
long distances, considering the types, specifications, and The stages are as follows: Stage 1) Move from the drop-
installation position. For example, the average off zone to the parking lot entrance. Stage 2) Move form
installation height of UADV's LiDAR is 1 meter, while parking lot entrance to the near position of the parking
the highest position is 2.3 meters. The LiDAR's FOV slot. Stage 3) Park in the parking slot.
angle is 25 degrees downward. Therefore, the highest
LiDAR can detect low obstacles only if they are at least 3.1 Stage 1) Drop-off to Parking Lot Entrance
4.94 meters away. The average LiDAR requires a The drop-off zone is where the driver gets off the
distance of 2.15 meters to detect such obstacles. vehicle. This zone can be any space determined by the
Consequently, collecting precise information on small user or operator. The exact location of the drop-off zone
obstacles or other objects within 1 meter of the vehicle is is not fixed, But Stage 1 always operates on regular roads
challenging. with lane markings. Extensive research has been
To acquire the necessary data for AVPS, additional conducted on ADS.
sensors should be configured. Sensors that is easily
provide close-range information include ultrasonic 3.2 Stage 2) Parking Lot Entrance to Parking Slot
sensors, AVM cameras, and mmWave radars. Ultrasonic The interior of the parking lot is considered a free
sensors obtain accurate distance information for close- space rather than a road with typical lane markings. In

132

Authorized licensed use limited to: R V College of Engineering. Downloaded on April 25,2025 at 06:52:12 UTC from IEEE Xplore. Restrictions apply.
free space, it is not possible to use the same algorithms as Algorithm 1: Pseudo-code of Parking Algorithm
in Stage 1. Instead, the vehicle should identify drivable
areas by recognizing surrounding area features and then 1 If the vehicle arrives near the parking slot:
generate a path accordingly. This involves identifying 2 Goal point = center of the parking slot
objects in the environment, distinguishing between 3 Local_x, local_y = relative distance between
drivable and non-drivable areas, and then generating a current position and goal
drivable path for the vehicle. Since the vehicle is not a point
holonomic system, it cannot move sideways or rotate in 4 r-turn = turn radius at maximum steering angle
place. Various path generation algorithms considering
the vehicle's dynamic constraints have been researched 5 Step 1: FREE DRIVING
[10]. One commonly used method in free space is the 6 While Local_y > r-turn/2:
Hybrid A* algorithm, which is used to overcome this 7 Move straight
stage.
In the USP scenario, ADS and AVPS operate within a 8 Step 2: FIND FINAL PATH
confined space, and it is assumed that UADV has an HD 9 While searching for a suitable final path radius
map of the parking lot. On USP, Stage 1 and 2 have (Θ𝑓𝑖𝑛𝑎𝑙 ):
already implemented similar research in previous studies. 10 R = | (local_x^2 + local_y^2) / (2*local_y) |
11 If (local_y > r):
3.3 Stage 3) Park in the Parking Slot 12 Θ𝑓𝑖𝑛𝑎𝑙 = sin-1 * R
The section from the area around the parking slot 13 Else:
through arriving slot are suffered difference situation for 14 Θ𝑓𝑖𝑛𝑎𝑙 = sin-1 * R + π
ADS and AVPS. The AVPS involves more steering 15 If minimum turning radius < final_path_radius:
maneuvers, operates at lower speeds, and experiences 16 Θ𝑓𝑖𝑛𝑎𝑙 ≈ local heading
minimal slip compared to ADS. Additionally, while ADS 17 End
primarily move forward, AVPS requires both forward 18 Else:
and backward maneuvers. AVPS requires careful 19 If Θ𝑓𝑖𝑛𝑎𝑙 - local heading - π > 0:
consideration of heading errors. 20 Turn right at maximum steering angle
In ADS, the overcoming lateral path errors is more 21 Else:
critical than heading angle errors. On regular roads, 22 Turn left at maximum steering angle
overcoming lateral errors also resolves heading errors.
However, in AVPS, both lateral errors and heading errors 23 Step 3: Follow FINAL PATH
should be considered. The AVPS should achieve precise 24 While x-axis aligns with vehicle heading angle:
positions tracking error compared to ADS. This is 25 δ = arctan(wheelbase / Θ𝑓𝑖𝑛𝑎𝑙 )
because AVPS operates in much narrower spaces, such
as parking lots. 26 Step 4: STRAIGHTs
This study is focus on vertical parking into a parking 27 While local x ≈ 0:
slots as required by the USP scenario. 28 Reverse
Fig. 5. Parking Algorithm
4. AUTONOMUS VALET PARKING SYSTEM The prerequisite for this algorithm to operate is that
ALGORITHM Stage 2 mentioned in Section 3.2 has been completed.
This chapter focuses on the Stage 3 vertical parking And the vehicle has stopped near the parking slot. This
scenario, using the following Parking Algorithm: algorithm is based on [11].
Step 1: Free driving involves driving straight until the
vehicle reaches a distance equal to the radius of the
minimum turning circle from the parking slot. The radius
of the minimum turning circle represents the minimum
distance required to be tangent to the final path.
Step 2: Find final path involves turning at the
maximum steering angle until the vehicle's trajectory
intersects the final path circle, which has a turning radius
larger than the minimum turning radius and intersects
perpendicularly with the parking slot. Only the maximum
steering angle was considered because numerous
Fig. 4. Parking situation coordinates description possible scenarios would be created by considering other
steering angles. This step aims to improve computational
speed by quickly reaching the final path.
In Fig.4, we determine the parking coordinates.
Step 3: Follow final path entails following the final
path found in Step 2 by adjusting the steering angle
accordingly. The final path is defined as a circular path

133

Authorized licensed use limited to: R V College of Engineering. Downloaded on April 25,2025 at 06:52:12 UTC from IEEE Xplore. Restrictions apply.
with a larger turning radius than the minimum turning CarlaCola’s wheelbass is 2665mm and width is
radius, and it intersects the parking slot perpendicularly. 1785mm. CarlaCola’s maximum steering angle is 40
By inputting the steering angle derived from the final degree. r-turn’s diameter is 4254mm. Start point is (-
path, the vehicle can accurately follow the path into the 16.5, -45), End point is (-21.5, -42.0). Parking simulation
parking slot. result as shown in Fig 7.
Through this algorithm (is shown in Fig.5), AVPS is As shown in Fig 8, the Parking Algorithm works other
implemented. goal position. Fig. 8. a to d situation’s start point is (-1.5,
-35). Fig. 8. e situation’s start point is (-17, -35).

5. SIMULATION RESULT
6. CONCLUSIONS AND FUTURE WORK
The simulation environment was implemented using
This paper presented the simulation study of
CARLA simulator, an open-source program developed
autonomous vehicle with restricted sensor network for
by the Computer Vision Center (CVC) and other
AVPS. We successfully used the CARLA simulator to
institutions. CARLA offers flexibility in vehicle and
accurately demonstrated our vehicle, sensors, and
sensor configurations, allowing UADV's exact state to be
specific scene. The parking algorithm successfully
replicated. The simulation was run on a Windows 11
completed the parking process in the simulation
laptop using CARLA version 0.9.14, python3.7, with
environment with two DGPS. Although this study
TOWN 5 map (is shown in Fig.6). SV vehicle is
experimented with a vertical parking algorithm, the
CarlaCola commercial truck. Simulation confirmed the
established simulation setup allows for testing various
feasibility of parking even without surrounding objects.
AVPS algorithms.
In future research, we will apply various AVPS
algorithms to develop a more efficient parking situation.
Additionally, we plan to use AVM information with
DGPS positional information. By utilizing AVM vision,
the vehicle is able to recognize parking lines and close
objects. If vehicle recognize parking lines and objects by
itself, an HD map is not necessary. Using vision
information will also make obstacle avoidance. The data
accumulated from simulations will be used to conduct
parking tests with the actual UADV.

Fig. 6. Simulator map (TOWN 5) and Parking lot

ACKNOWLEDGMENT
This work was supported by the Technology
Innovation Program (RS-2024-00406173, Development
of functional reconstruction and self-recovery element
technology to respond to defects and errors in
autonomous vehicle electrical components) funded By
the Ministry of Trade Industry & Energy (MOTIE,Korea)

REFERENCES
Fig. 7. Simulator’s GPS Result [1] J.Y. Do, K. Han, S.B. Choi, “Inference of Vehicle
Lane Change Intention Using Multiple Model
Estimatorin Automated Highway Driving”, in Proc,
International Conference on Control, Automation
and Systems (ICCAS), Busan, Korea, pp. 366-372,
Nov, 2022
[2] N. Albarella, D.G. Lui, A. Petrillo and S. Santini,
“A Hybrid Deep Reinforcement Learning and
a) (-6, -36.8) b) (-6, -20) c) (-6, -22.5) Optimal Control Architecture for Autonomous
Highway Driving” MDPI Energis, 2023, 16, 3490,
doi : https://doi.org/10.3390/en16083490
[3] S.H Bae, S.H. Joo, J.W. Pyo, J.S. Yoon, K. Lee and
T.Y. Kuc, “Finite State Machine based Vehicle
System for Autonomous Driving in Urban
d) (-6, -28) e) (-21.5, -28) Environments”, in Proc. International Conference
Fig. 8. Parking other position on Control, Automation and Systems (ICCAS),
Busan, Korea, pp. 1181-1186, Oct, 2020

134

Authorized licensed use limited to: R V College of Engineering. Downloaded on April 25,2025 at 06:52:12 UTC from IEEE Xplore. Restrictions apply.
[4] T. Vidano and F. Assadian “Control Performance
Requirements for Automated Driving Systems”,
MDPI Electronics, 2024, 13, 902, doi:
https://doi.org/10.3390/electronics13050902
[5] H. Cho, Y.W. Seo, B.V.K.V. Kumar and R.R.
Rajkumar, “A Multi-Sensor Fusion System for
Moving Object Detection and Tracking in Urban
Driving Environments” Proc. IEEE International
Conference on Robotics & Automation, Hong Kong,
China, pp. 1836-1843, June 2014
[6] S. Royo and M. Ballesta-Garcia, “An Overview of
Lidar Imaging Systems for Autonomous Vehicles”,
MDPI Applied Sciences, 2019, 9(19), 4093.
https://doi.org/10.3390/app9194093
[7] M. Ankush, “mmCaliper: High resolution sensing
of distances using mmWave Radars” M.S. thesis,
Dept. Computer Science and Engineering, Univ. of
Pennsylvania, Philadelphia, PA, USA, 2023
[8] Z. Renyuan, “mmWave Radar: Enhancing
Resolution, Target Recognition, and Fusion with
Other Sensors” Ph. D. thesis, Dept. Electrical &
Computer Engineering, Univ. of Arizona, Arizona,
USA, 2019
[9] Synerex, "MGI-2000," [Online]. Available:
https://www.synerex.kr/mgi-2000.
[10] D. Dolgov, S. Thrun, M. Montemerlo, and J. Diebel,
"Practical Search Techniques in Path Planning for
Autonomous Driving," in Proc. International
Symposium on Experimental Robotics (ISER),
Athens, Greece, pp. 1-9, 2008.
[11] M.F. Hsieh and Ü. Özguner, "A Parking Algorithm
for an Autonomous Vehicle," Proc. IEEE
Intelligent Vehicles Symposium, Eindhoven
University of Technology, Eindhoven, The
Netherlands, pp. 1155-1160, June 2008.

135

Authorized licensed use limited to: R V College of Engineering. Downloaded on April 25,2025 at 06:52:12 UTC from IEEE Xplore. Restrictions apply.