See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/325475976

Design and Development of Intelligent Self-driving Car Using ROS and

Machine Vision Algorithm

Chapter · January 2019

DOI: 10.1007/978-3-319-78452-6_9

CITATIONS READS

4 3,009

5 authors, including:

Aswath Suresh Dhruv Gaba

New York University New York University
17 PUBLICATIONS 120 CITATIONS 8 PUBLICATIONS 49 CITATIONS

SEE PROFILE SEE PROFILE

Debrup Laha Siddhant Bhambri

Polytechnic Institute of New York University Bharati Vidyapeeth College of Engineering, Delhi
7 PUBLICATIONS 44 CITATIONS 7 PUBLICATIONS 44 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Aswath Suresh on 19 May 2019.

The user has requested enhancement of the downloaded file.

Cite this paper as:
Suresh A., Sridhar C.P., Gaba D., Laha D., Bhambri S. (2019) Design and Development of Intelligent Self-driving Car Using ROS
and Machine Vision Algorithm. In: Kim JH. et al. (eds) Robot Intelligence Technology and Applications 5. RiTA 2017. Advances in
Intelligent Systems and Computing, vol 751. Springer, Cham

Design and Development of Intelligent Self-Driving Car using ROS and

Machine Vision Algorithm

Aswath Suresh1, Dhruv Gaba1, Debrup Laha1, and Siddhant Bhambri2

1
Department of Mechanical and Aerospace Engineering, New York University, USA
2Department of Electronics and Communication Engineering, Bharati Vidyapeeth’s College of
Engineering, India

[email protected], [email protected], [email protected], [email protected]

Abstract. The technology of self-driving cars is quite acclaimed these days. In

this paper we are describing the design and development of an Intelligent Self
Driving Car system. The system is capable of autonomously driving in places
where there is a color difference between the road and the footpath/roadside,
especially in gardens/parks. On the basis of the digital image-processing
algorithm, which resulted into optimal operation of the self-driving car is based
on a unique filtering and noise removal techniques implemented on the video
feedback via the processing unit. We have made use of two control units, one
master and other is the slave control unit in the control system. The master control
unit does the video processing and filtering processes, whereas the slave control
unit controls the locomotion of the car. The slave control unit is commanded by
the master control unit based on the processing done on consecutive frames via
Serial Peripheral Communication (SPI). Thus, via distributing operations we can
achieve higher performance in comparison to having a single operational unit.
The software framework management of the whole system is controlled using
Robot Operating System (ROS). It is developed using ROS catkin workspace
with necessary packages and nodes. The ROS was loaded on to Raspberry Pi 3
with Ubuntu Mate. The self-driving car could distinguish between the grass and
the road and could maneuver on the road with high accuracy. It was able to detect
frames having false sectors like shadows, and could still traverse the roads easily.
Thus, self- driving cars have numerous advantages like autonomous surveillance,
car- parking, accidents avoidance, less traffic congestion, efficient fuel
consumption, and many more. For this purpose, our paper describes a cost-
effective way for implementing self- driving cars.
Keywords: Autonomous, Drive, Image Processing, ROS, SPI

1 Introduction

An autonomous car is a vehicle that is capable of sensing its environment and

navigating without human input. Many such vehicles are being developed, but as of
May 2017 automated cars permitted on public roads are not yet fully autonomous.
All self-driving cars still require a human driver at the wheel who is ready to take
control of the vehicle at any moment. Autonomous cars use a variety of techniques to
detect their surroundings, for path planning and efficient decision-making they uses
radar, laser light, GPS, odometer, and computer vision techniques and many more for
correct behavior of the car. Advanced control systems interpret sensory information to
identify appropriate navigation paths, as well as obstacles and relevant road signs. Path
planning and obstacle avoidance utilizing the information obtained by the above-
mentioned sensors and techniques to drive to the goal without intervention of a driver
[1].
Cite this paper as:
Suresh A., Sridhar C.P., Gaba D., Laha D., Bhambri S. (2019) Design and Development of Intelligent Self-driving Car Using ROS
and Machine Vision Algorithm. In: Kim JH. et al. (eds) Robot Intelligence Technology and Applications 5. RiTA 2017. Advances in
Intelligent Systems and Computing, vol 751. Springer, Cham

Most of the driver assistance systems for the self-driving cars are developed to assist
the drivers and to improve the driver's’ safety such as in lane keeping [2], autonomous
parking [3], and collision avoidance [4]. Such systems are used heavily to reduce traffic
and accident rates. And currently rigorous Research is going on in this field to improve
the accuracy of the self-driving cars.
The major impetus which leads to the development of the self-driving car
technology was the Grand Challenge and Urban Challenge organized by DARPA
(Defense Advanced and Research Projects Agency) [5][6]. These were first organized
in 2004, then in 2005 (Grand Challenge), and then followed in 2007 (Urban
Challenge). This has led to numerous studies regarding accurate path planning, which
are categorized into three major areas namely, Potential-field approach [7], Roadmap
based approach [8] [9] [10] [11], and Cell decomposition based approach [12].
In paper [13], Voronoi cell based approach is used for path representation, collision
detection and the path modification rather than waypoint navigation. But the
inaccuracy in predicting the size of these cells in a dynamic environment makes the
approach ineffective. On the other hand, our color segregation based approach
(explained in Section 4) that does not require cells and path planning done in real time
environment, has a higher performance.
The complete system is subdivided into the drive mechanism and the control system
of the car. The drive mechanism is responsible for the locomotion of the car as it has
the motors, tires, mount for RaspiCam, chassis etc. for the operation of the car. But,
the control system is the most important part of the system because the entire
intelligence of the system is residing in this part. The control system is subdivided into
a master control unit and a slave control unit, which are explained in detail in the
Section 3. The reason for choosing distributed control systems for our self-driving car
is to improve the overall performance of the system.

In this paper we are explaining the algorithm that was created for the autonomous
car and its implementation on the self-driving car prototype to verify its capabilities.
The key aspect of our algorithm behind the autonomous behavior of the car is the
detection of color difference between the road and the green grass to determine the
direction in which the car must turn based on the run-time camera input and removal of
noise for error free decision making is also done and explained in the Section 4. The
programming of the prototype was done in “OpenCV (Python)”.

The upcoming sections explains the following information:

Section 2: Mechanical Structure and Design of the system

Section 3: Electrical connections and architecture of data flow

Section 4: Algorithm used for decision making

Cite this paper as:
Suresh A., Sridhar C.P., Gaba D., Laha D., Bhambri S. (2019) Design and Development of Intelligent Self-driving Car Using ROS
and Machine Vision Algorithm. In: Kim JH. et al. (eds) Robot Intelligence Technology and Applications 5. RiTA 2017. Advances in
Intelligent Systems and Computing, vol 751. Springer, Cham

Fig 1: Assembled Autonomous System

2 Mechanical Design

The mechanical design of the car is shown in the Fig. 1. The car in the image is a
prototype for representing our algorithm for effective determination of the direction of
the car’s movement.
For the prototype of the RC car, at first the circuit of the RC car is removed, and the
control of the system is given to the Arduino development board using the Adafruit
shield. In this mechanical structure, Ping Sensors (Ultrasonic Sensor) are attached for
the autonomous braking to avoid collision.
The RC car used for the prototype is based on differential drive mechanism.
Differential drive robots are based on differential drive mechanism in which the motion
of the system is divided into two sections which are right and left parts. The wheels of
both segments are attached to common axis and have independent motion and rotation
capabilities. If the system consists of two wheels, both the wheels can independently
be driven either forward or backwards. In differential drive, we can vary the velocity
of each wheel independently and change the direction of motion of the whole unit.
Further, the camera mount is designed for mounting the RaspiCam on the front end
of the car to get the required field of vision for the efficient image processing. The
Camera angle from the car must be 20° at the height of 20 cm. For this purpose, we
have designed and 3D printed the camera mount, as shown in the Fig. 2 and Fig 3.
Also, prototype of car driving autonomously in park is as shown in Fig. 4.
Cite this paper as:
Suresh A., Sridhar C.P., Gaba D., Laha D., Bhambri S. (2019) Design and Development of Intelligent Self-driving Car Using ROS
and Machine Vision Algorithm. In: Kim JH. et al. (eds) Robot Intelligence Technology and Applications 5. RiTA 2017. Advances in
Intelligent Systems and Computing, vol 751. Springer, Cham

(a) Isometric View of Lower part (b) Isometric View of Upper Part
Fig 2: Camera Mount Design

Fig 3: 3D Printed Camera Mount

The camera mount is fixed on the front end of the car as shown in the fig.3.

Fig 4: Car driving autonomously in park

3 Electrical Design
The control system of the autonomous car is subdivided into master and slave control
unit. The master control unit is the Raspberry Pi B3 microprocessor and the slave
control unit is based on Arduino development platform. We separated the two control
units and distributed the task allocated to them because by this manner we can achieve
higher responsiveness and better performance of the system.
Cite this paper as:
Suresh A., Sridhar C.P., Gaba D., Laha D., Bhambri S. (2019) Design and Development of Intelligent Self-driving Car Using ROS
and Machine Vision Algorithm. In: Kim JH. et al. (eds) Robot Intelligence Technology and Applications 5. RiTA 2017. Advances in
Intelligent Systems and Computing, vol 751. Springer, Cham

Architecture of Control system: Control and Data flow

The architecture and flow of control as well as data in our system is explained in
the following subsection as shown in Fig.5.:

Fig 5: Architecture of Data flow in the system.

3.1 RaspiCam Video Camera

The video is captured using the RaspiCam, which is a 5MP camera attached with the
Raspberry Pi B3 such that the video (i.e. image frames) are recorded and then sent to
the Raspberry Pi microprocessor for processing of data and decision making.

3.2 Master Control Unit: Raspberry Pi B3 Microprocessor

The master control unit is used for video processing in the system. The video
recorded by RaspiCam is converted into image frames and processed by Raspberry Pi
B3 microprocessor because of its higher processing power and additional capabilities
which makes it the most suitable choice for this application. The algorithm used for the
processing of the image frames is explained in detail in section 4. At the end of the
algorithm a message array of 8 bits is created which is transmitted wirelessly to the
Arduino platform for further executions.

3.3 Slave Control Unit: Arduino Development Platform

The slave micro-controller is used for controlling the drive system of the car. The
Arduino is connected to the Adafruit motor shield which drives power from the
Lithium Polymer battery. Arduino commands the motor shied and then it controls the
operation of the motors for the drive of the car.

3.4 Wireless Serial Peripheral Interface Communication: nRF24L01 Module

The results generated after the processing of data by implementing the algorithm in
the master control unit, are to be transmitted to the slave control unit (Arduino
Development Platform) for maneuvering the car. The communication between the
master and the slave control unit is achieved using the Serial Peripheral Interface (SPI)
communication. SPI wireless communication is achieved using the nRF24L01 (2.4
GHz) transceiver modules and the data is transmitted in the form of 8-bit signal. The
signal bits are interpreted in the following manner at the slave control unit:

1. Bits 1-3: Driving motor speed

2. Bits 4-6: Steering motor speed
3. Bit 7 : Driving motor direction
4. Bit 8 : Steering motor direction
Cite this paper as:
Suresh A., Sridhar C.P., Gaba D., Laha D., Bhambri S. (2019) Design and Development of Intelligent Self-driving Car Using ROS
and Machine Vision Algorithm. In: Kim JH. et al. (eds) Robot Intelligence Technology and Applications 5. RiTA 2017. Advances in
Intelligent Systems and Computing, vol 751. Springer, Cham

(a) Arduino and NRF module (b) Raspberry pi and NRF module

Fig 6: Circuit Diagram of NRF module

3.5 Ping Sensors (Ultrasonic Sensor)

For emergency breaking system of the autonomous self-driving car, we have used Ping
sensors. The slave control unit monitors the readings from the Ping Sensors, such that
if by any means something or someone comes in the way of the car the breaks are
applied immediately to avoid collision.

3.6 Adafruit Motor Drive Shield

The Adafruit Motor Shied is attached to the Arduino development platform. Arduino
commands the motor shied which further controls the motors of the car. We have used
motor shield to segregate the high current battery supply for the motor from the low
current supply for the electronic systems.

3.7 Power Source: Lithium Polymer Battery

We are making use of Lithium Polymer Battery (11.1 V-3s) for providing power supply
to the entire system. Direct supply from the battery is given to the motors as they have
high current rating and regulated voltage (5V from 7805IC) for the electronics
controllers as they operate on 5 Volts. This is the common source of power for the
system.

3.8 ROS: Robot Operating System

All the software frame work of the system is well developed using ROS catkin
workspace with necessary packages and nodes. The ROS was loaded on to
Raspberry Pi 3 with Ubuntu Mate. Python scripts had been used for
programming in ROS.

3 Algorithm Explanation
We extract the video feedback from the RaspiCam, which is recorded by the Raspberry
Pi 3, and dynamically store in it for processing. Raspberry Pi processes the input frame
by frame, on which we sequentially implement the following filters for decision
making.

3.1 Resizing of Image Frame

The RaspiCam provides us with highly detailed images. It has 5MP camera for HD
recording. For rapid response of our system, we have chosen our image to have
600 Pixel of width because processing on large matrices (derived from the images)
will take larger amount of time to process it. So, to avoid lag in our system we are
Cite this paper as:
Suresh A., Sridhar C.P., Gaba D., Laha D., Bhambri S. (2019) Design and Development of Intelligent Self-driving Car Using ROS
and Machine Vision Algorithm. In: Kim JH. et al. (eds) Robot Intelligence Technology and Applications 5. RiTA 2017. Advances in
Intelligent Systems and Computing, vol 751. Springer, Cham

reducing the size of our image by using “imutils.resize()” command of “OpenCV”

library as shown in Fig.7. This way all our frames are resized to 600 pixels width,
which makes our filtering procedure more convenient and efficient.

Fig 7: Function for resizing the frame width.

3.2 Gaussian Blur Algorithm

Every image, which we get from the RaspiCam is having some or the other form
of high frequency noise. So, to remove this noise we are making use of low pass
filters. One of the most common forms of algorithm which will not only remove
the sharp noises, but also blur the image is the Gaussian Blur filters.

For operation of this filter we need to specify the width and height of the kernel
to the function, which should be positive and odd. Here we have used a kernel of
11 sizes. Further, we will have to specify the standard deviation in X and Y
direction, i.e. σX and σY; here we are using both as zeros, thus they are calculated
from the kernel size. Therefore, via Gaussian blur we are able to remove Gaussian
noise very effectively as shown in Fig.8.

Fig 8: Illustrating implementation of Gaussian blur and its function code.

Cite this paper as:
Suresh A., Sridhar C.P., Gaba D., Laha D., Bhambri S. (2019) Design and Development of Intelligent Self-driving Car Using ROS
and Machine Vision Algorithm. In: Kim JH. et al. (eds) Robot Intelligence Technology and Applications 5. RiTA 2017. Advances in
Intelligent Systems and Computing, vol 751. Springer, Cham

ALGORITHM FLOWCHART:
Process Flow for OpenCV Program

Video extraction from RaspiCam

Frame by Frame feed

Applying Gaussian Blur

Conversion from BGR to HSV

Removing unwanted area Frame

Dynamic Threshold Parameter extraction from ROI

Masking based on Dynamic Threshold Parameters

Applying Erosion

Applying Opening

Applying Closing

Taking Decisions for Motor

Commands

Sending commands to Arduino

Through SPI
Fig 9: Flowchart for the decision making algorithm

The decision making algorithm of the self-driving car is as shown in flow chart in Fig.9.
3.3 Conversation from BGR (Blue, Green, Red) to HSV (Hue, Saturation,
Value)
We are changing the color space from BGR to HSV such that we can effectively
segregate the road from the green grass. In this step we are simply converting the entire
BGR image/frame (width = 600) to HSV color space so that just by simply making a
range mask, we are able to segregate the desired color range which is grey road from
the green grass in the case of a park as shown in Fig 10.
Cite this paper as:
Suresh A., Sridhar C.P., Gaba D., Laha D., Bhambri S. (2019) Design and Development of Intelligent Self-driving Car Using ROS
and Machine Vision Algorithm. In: Kim JH. et al. (eds) Robot Intelligence Technology and Applications 5. RiTA 2017. Advances in
Intelligent Systems and Computing, vol 751. Springer, Cham

Fig 10: Illustrating the conversion of BGR to HSV color space.

3.4 Removing unwanted area from Image Frame

After the image is converted from BGR color space to HSV color space, we find
the area having the most information present and remove the unwanted area from
the frame. For our system to perform with utmost accuracy in any unstructured
environment, we need to remove all unnecessary information from the frame. For
example, we will have to remove those regions which do not contain the road and
grass information. These unwanted areas like the sky, the trees, twigs, sticks, etc.
which do not have road and grass, are to be removed. This parameter of new
height of the frame to be used is dependent on the height of the camera and the
vertical inclination of the camera. Therefore, after this step we get the updated
frame for higher performance.
Here we are making use of np.size() function to get the height and width
dimension of the HSV color space. Using this, we adjusted the new frame for our
processing as shown in Fig.11.

Fig 11: Illustrating removal of unwanted areas.

Cite this paper as:
Suresh A., Sridhar C.P., Gaba D., Laha D., Bhambri S. (2019) Design and Development of Intelligent Self-driving Car Using ROS
and Machine Vision Algorithm. In: Kim JH. et al. (eds) Robot Intelligence Technology and Applications 5. RiTA 2017. Advances in
Intelligent Systems and Computing, vol 751. Springer, Cham

3.5 Dynamic Threshold Parameter

In this step we are adjusting our frame size such that we could determine the
required Region of Interest (ROI). As shown in the Fig. 12, we have used the
following command to get it.

Fig 12: Dynamic thresholding of the Parameter ROI

3.6 Dynamic Threshold Setting

In this code we could take values from the path [] matrix (created from hsv []
matrix) and assign its corresponding values to h [], s [], and v [] arrays. Then we
simply find the values which are the maximum and minimum of the h [], s [], and
v [] arrays and save them in hmax, hmin, smax, smin, vmax, and vmin. And then
correspondingly using them to create a two “numpi” arrays (“rangemin[] and
rangemax[]”), one having the maximum ranges and the other having the minimum
ranges as shown in Fig.13.

Fig 13: Code for finding the minimum and maximum ranges of hue, saturation, and
value in HSV color space.

3.7 Masking based on Dynamic Threshold Parameters

The following code creates the mask for transforming the image frame in
accordance to the min and max range provided from the previous steps as shown
in Fig.14. We have made us of cv2.inRange () function to create this mask.
Cite this paper as:
Suresh A., Sridhar C.P., Gaba D., Laha D., Bhambri S. (2019) Design and Development of Intelligent Self-driving Car Using ROS
and Machine Vision Algorithm. In: Kim JH. et al. (eds) Robot Intelligence Technology and Applications 5. RiTA 2017. Advances in
Intelligent Systems and Computing, vol 751. Springer, Cham

Fig 14: Illustrating mask creation from the max and min ranges of HSV.

3.8 Applying Erosion, Opening, and Closing

In this step we are performing Erosion, Opening and Closing functions one by
one to remove various kinds of noises from our image frames, which is the basis
of our decision making.
- Erosion will remove the boundaries of the foreground object (displayed in
white) as the kernel (25x25) slides over the image (convolution), and the
pixel in the original image is considered 1 only if all the pixels in the kernel
are 1, otherwise it is eroded 0.
- Opening is another function to remove noise and it functions in a manner of
erosion followed by dilation, thus removing small particle noise from the
image frame.
- Closing is reverse of Opening, i.e. Dilation followed by Erosion; it is useful
for closing small holes, or small black points on the object.
Therefore, implementation of these filters one by one helps the robot to render the
clean and noise free image frame as shown in the Fig. 15.

Fig 15: Implementation of Erosion, Opening, and Closing filters in code and frames.

3.9 New Regions of Interest for Turning Decisions/ Taking Decisions for Motor
Commands
The following code represents the ‘if-else’ conditions which are used to determine
the direction of the drive motor and the steering motor. These ‘if-else’ loops check
for the white and black regions on the top left and right side of the image frame.
This way we can decide in which direction the car has to move, and then transmit
the data to the slave controller (Arduino).
Also, the second image Fig 16 shows the formatting of the 8-bit data encoding
which is transmitted to the slave controller. In Fig. 16 the first three bits represent
the Drive Wheel Velocity, the next three consecutive bits represent the Steer
Wheel Velocity, the second last bit represents the direction of the Drive Wheel,
and the last bit represents the direction of the Steer Wheel.

Fig 16: Determination of the direction of turning.

Cite this paper as:
Suresh A., Sridhar C.P., Gaba D., Laha D., Bhambri S. (2019) Design and Development of Intelligent Self-driving Car Using ROS
and Machine Vision Algorithm. In: Kim JH. et al. (eds) Robot Intelligence Technology and Applications 5. RiTA 2017. Advances in
Intelligent Systems and Computing, vol 751. Springer, Cham

3.10 Sending commands to Arduino through SPI

Once the algorithm has determined the direction in which the car has to move, the
master controller (Raspberry Pi) will transmit the data via wireless serial
communication to the slave controller (Arduino).

The dataTX takes the message generated from the previous step and it will save
it in the dataTX which is transmitted serially to Arduino. Also, in this step we
check if the length of the dataTX is less than 8. If it is less than 8, we will have to
append a 0 to make it 8-bit to avoid any errors which could happen during wireless
communication.

Now, serial communication begins and we record time using time.time() function
into the start variable. After that we write this data to the radio transmitter which
will transmit it wirelessly to Arduino for movement and simultaneously print it
onto the console as well.

The transmitted code is then received by the Arduino, which is further decoded
by it for implementation of the locomotion commands. For example, if the drive
or steering direction bit is 1 then move the respective motor in forward direction,
whereas if the drive or steering direction bit is 0 then move the respective motor
is moved in the backward direction as shown in Fig.17.

Fig 17: Illustration of message signal transmission from Raspberry Pi and decoding of the
message signal at Arduino.

Conclusion

Thus, for reducing the cost of self-driving cars we use normal cameras and sensors
instead of complicated sensors, which tend to be quite expensive. So, our system
resulted into a more cost-effective system as compared to the current ones.
Additionally, the digital image processing techniques used in the algorithm of the self-
driving car had satisfactory results. We implemented and tested the self-driving car in
Cadman Plaza Park.

Future Scope

As for future development, we could use machine learning and deep learning
techniques for even higher accuracy of the system and dynamic upgradation of the
systems parameters according to its surroundings could also be implemented onto the
system.

Also, for those roads which do not have significant color difference between road
 Cite this paper as:
Suresh A., Sridhar C.P., Gaba D., Laha D., Bhambri S. (2019) Design and Development of Intelligent Self-driving Car Using ROS
and Machine Vision Algorithm. In: Kim JH. et al. (eds) Robot Intelligence Technology and Applications 5. RiTA 2017. Advances in
Intelligent Systems and Computing, vol 751. Springer, Cham

and the side roads, we could use the road dividers as markers to stay on the road as the
object recognition algorithms will track these road dividers for driving the car on the
road.

Acknowledgement

The authors would like to thank Makerspace, New York University to provide support
and resources to carry out our research and experiments.

References

[1] J. Wit, C. D. Crane, and D. Armstrong, "Autonomous Ground Vehicle Path Track-
ing," Journal of Robotic Systems, vol. 21, no. 8, pp. 439-449, 2004.
[2] J. Wang, S. Schroedl, K. Mezger, R. Ortloff, A. Joos, and T. Passegger, "Lane
Keeping Based on Location Technology," IEEE Transactions on Intelligent Transpor-
tation Systems, vol. 6, no. 3, pp. 351-356, September 2005.
[3] Li, TS and Yeh, Ying-Chieh and Wu, Jyun-Da and Hsiao, Ming-Ying and Chen,
Chih-Yang, "Multifunctional intelligent autonomous parking controllers for carlike
mobile robots", IEEE Transactions on Industrial Electronics, Vol. 57, No. 5, pp. 1687-
1700, 2010.
[4] Kwanghyun Cho, Seibum B. Choi, "Design of an Airbag Deployment Algorithm
Based on Pre-Crash Information", IEEE Transaction on Vehicular Technology, Vol.
60, No. 4, pp. 1438-1452, 2011.
[5] Ü. Özgüner, C. Stiller, and K. Redmill, "Systems for safety and autonomous be-
havior in cars: The DARPA Grand Challenge experience," Proc. IEEE, vol. 95, no. 2,
pp. 397-412, Feb. 2007.
[6] Thrun, S., et al., "Stanley: The robot that won the DARPA Grand Challenge",
Journal of Field Robotics, 23(9), 661-692, 2006.
[7] O. Khatib, "Real-Time Obstacle Avoidance for Manipulators and Mobile Robots,"
The International Journal of Robotics Research, vol. 5, no. 1, pp. 90-98, March 1986.
[8] S. Koenig and M. Likhachev, "Fast Replanning for Navigation in Unknown Ter-
rain," IEEE Transactions on Robotics, vol. 21, no. 3, pp. 354-363, June 2005.
[9] N. S. V. Rao, N. Stoltzfus, and S. S. Iyengar, "A Retraction Method forLearned
Navigation in Unknown Terrains for a Circular Robot," IEEE Transactions on Robot-
ics and Automation, Vol. 7, No. 5, pp. 699-707, October 1991.
[10] L. E. Kavraki, P. Švestka, J.-C. Latombe, and M. H. Overmars, "Probabilistic
Roadmaps for Path Planning in High-Dimensional Configuration Spaces," IEEE
Transactions on Robotics and Automation, Vol. 12, No. 4, pp. 566-580, August 1996.
[11] H. M. Choset, Principles of Robot Motion: Theory, Algorithms, and Implementa-
tion. Cambridge, MA: MIT Press, 2005.
[12] L. Zhang, Y. J. Kim, and D. Manocha, "A Hybrid Approach for Complete Mo-
tion Planning," in IROS, October 2007, pp. 7-14.
[13] Unghui Lee, Sangyol Yoon, HyunChul Shim, Pascal Vasseur, Cedric Demon-
ceaux, “Local path planning in a complex environment for self-driving car,” The 4th
Annual IEEE International Conference on Cyber Technology in Automation, Control
and Intelligent Systems June 4-7,2014, Hong Kong, China.

View publication stats