You are on page 1of 6

11th IEEE International Symposium on Applied Computational Intelligence and Informatics • May 12-14, 2016 • Timişoara, Romania

Issues about Autonomous Cars

Claudiu Poznaa,b Csaba Antonya


a
Transilvania University of Brasov, Romania Transilvania University of Brasov Romania
b
Széchenyi István University, Győr, Hungary antonya@unitbv.ro
cp@unitbv.ro

Abstract— The Autonomous car is a complex topic, Figures 1-2) the real developments have been achieved by
different technical fields like: Automotive engineering, designing the car from scratch
Control engineering, Informatics, Artificial Intelligence etc.
are involved in solving the human driver replacement with an
artificial (agent) driver. The problem is even more complicated
because usually, nowadays, having and driving a car defines
our lifestyle. This means that the mentioned (major)
transformation is also a cultural issue. The paper will start
with the mentioned cultural aspects related to a self-driving
car and will continue with the big picture of the system.

Keywords— autonomous car, control architecture, mobile


robots

I. INTRODUCTION
At first glance the Autonomous Car (AC) concept can
be defined like a car which is able to drive itself. This Figure 2. The second Marcus car of 1888 at the Technical Museum in
Vienna [https://en.wikipedia.org/wiki/History_of_the_automobile]
means that we deal with a car which copies the human
driver’s performances.
Is this a mobile robot? The answer is yes, if we consider
And designing a car is only a small part of the changes.
only the autonomy, but no, if we have in mind that it is
Because this moving possibility (by cars) proves such a
necessary to add specific interfaces that are not needed in
success it lead, by a domino effect, to new infrastructure
the case of the mobile (we mention here the interface
(roads), transportation conventions (signs), etc. in fact to a
between the human passenger and the autonomous car,
new lifestyle. This new locomotion possibility creates also
etc.).
a powerful economic development caused by the cars
manufacturing factories and also by the related industries.
From these points of view the mentioned developments
have had a cultural impact.
Also, the AC is it an (artificial) intelligent agent which
has the ability to navigate like a car or is it a car which has
the ability to navigate in an intelligence way? We think
that it is not practical to turn this vague (not well defined
yet) subject in to another vague subject who refers to the
artificial intelligence. Unfortunately it is a trend in our
culture to name ordinary product like intelligent product
(starting with the famous screwdriver and ending by cars).
However it is clear that the AC is a mechatronic product,
which use information processing, this means also
learning technics, we imagine that the level of abstraction
Figure 1. M Cugnot's steam wagon, the second (1771) version of these programs far outstrips the level of nowadays
[https://en.wikipedia.org/wiki/History_of_the_automobile]
programs used by the cars ECU (Electronic Control
Units).
Is this a car? Once again the answer is yes, if we focus
on its navigation capability and no, if we observe that the If we extrapolate these developments we must be
difference between a car and an AC is more than a simple prepare for a revolution: a new transportation system -
aggregation of several systems. If our original system (the which we name nowadays the Autonomous Car (AC), in a
car) was not designed (form the beginning) to enable such similar way to our ancestors who named the innovation by
a development it is more suitable to start a new design. motor – cab new infrastructures, new driving conventions
From many points of view the difference between a cart etc.
and a car reflect the difference between the car and the Nowadays the car is much more than a transportation
AC. Even at the first trails the cars have been constructed possibility, like all important thinks from our life, it was
by adding an engine and a transmission to the cart (see transformed and cultural integrated in to a symbol. The

978-1-5090-2380-6/16/$31.00 ©2016 IEEE – 13 –


C. Pozna, Cs. Antonya • Issues about Autonomous Cars

mobility was related with the freedom, the engine power II. THE PROPOSED SYSTEM ARHITECTURE
with authority. As a result the car became a symbol of
masculinity, fact which is use abundantly by cars
designer. Intuitively we can infer that the AC will not For the currently development stage of the AC
have these characteristics. The AC will belong to a construction, characterized mostly by the aggregation
network; the driver will become a passenger which is not between a traditional car and a control system we sustain
more in charge; the shape of the AC will be different the idea of using an expert system. The goal of the
because the many subsystems of the car will be mentioned expert system is to manage the Driver
unnecessary (steering wills, gear box lever, etc.) and other Behaviors (DB). The proposed control system
will appear (working or entrainment habitat, architecture use the human driver behavior model
communication with other AC, etc.) or will be developed concept.
(communication etc.). The AC will have other symbols Preliminary discussions are necessary. The “Driver’s
and significance in our future culture. Behaviors” model is used in the simulation [15], [16] and
Because the AC is an exciting subject, not only for also in autonomous car design field [14]. It is known that
universities and cars manufacturers but also for the the first researches on the subject take place in 1950 [15]
general public, the danger of misconceptions appears. and began with the “Skill-based driving model”,
continued with the “Motivational model” which take into
From [1] we have selected the following ideas: account the drivers’ emotional state (we mention here the
• “Driver assistance systems will evolve gradually “Risk compensation”; “Risk avoidance” and “Risk
into fully autonomous cars”. In fact the AC is not threshold models”). Lately, the model became a
a result of driver assistant systems evolutions. “Hierarchical control structure” (Milchon). The
Adding capabilities like lateral dynamic “Hierarchical control structure” splits the driving process
assistance stop and go assistance, etc. will not into three levels of control: a strategic level, here the goal
generate (in the end) the AC. All these system of the driving – the task-is establish; a tactical level, used
are designed for the traditional cars; for solving parts of the tasks; and (in the end) an
• “The first models of fully autonomous cars will operational level, which implements the mentioned
be targeted to the consumer and will be available solutions using low level control. Using this
for purchase”. In fact the autonomous driving “Hierarchical control structure” numerous scientific
problem complexity will increase the time papers develop problems like: “Longitudinal behaviors
horizon until such a system (the AC) will models” [14]; “Lateral behaviors model” [17]; “Brake
becomes an available product; behavior” [14] etc. We proposed [18] a heuristic
approach related to the control programs architectures
• “Public demonstrations of self-driving cars
which model the human behavior [15], [16].
provide an indication of their capabilities”; In
fact even nowadays encouraging results have Our idea claims that it is more suitable to model and
been proved by the researcher, the problem implement the “human driver decisions act”, than the
complexity far exceeds these success. “human driver actions”. Using this concept we transfer
the modeling (approximation) of the human driver
behaviors from a mechanical problem to a computer
We briefly sum up our conclusions concerning the science problem.
developments of the autonomous cars:
From care to define the problem correctly, preliminary
• The AC subject performs nowadays its baby step. (phenomenological - here the words wisdom
• There are several projects made by powerful car methodology) analyses [20] need to answer to the
companies. The results of these projects are following questions: “What are driving behaviors?”;
summarized by the sentence: “It can be done, but “Can we obtain fundamental true about these behaviors
it is too expensive”; and use them in our construction?”
• The autonomous car construction implies a
mechanical and electrical design. There are two The phenomenological researches start with the
solutions: transform a real car into an semantic characteristics of “Driving behavior”. The
autonomous car [2] or design a new vehicle [3]. category tree of behavior word consist on {act → activity
→ (behavior, practice,...)}. The behavior is defined like
• The mobile robots’ navigation has been defined “an action or a set of actions performed by a person
from mathematical point of view [8], this means under specified circumstances that reveal some skill,
that we have the axiomatic background for the knowledge or attitude”.
solutions;
• Each mentioned work try to solve the navigation Researches on driving behavior [14-19] revealed that
problem or a part of this problem in a particular the driving behavior has a special feature. In order to
way. There are used classical robust control describe it, we concentrate on the word “custom” which
techniques [9], fuzzy logic or neural network belongs to the same category tree {act → activity →
strategies [4, 10-13] or solution based on practice → habit,.}. Custom is defined as: “accepted or
Bayesian theory of probability [ 3, 5, 6, 7, 8]; habitual practice”. These habits have a special property:
• One drawback of the mentioned solutions is the the automatism - any reaction that occurs automatically
time consuming computation, needed in solving without conscious thought or reflection. Concatenating
complex situations [8]. the previous definitions, “A Driving Behavior” can be
defined like an action or a set of actions performed by a
person under driving circumstances, actions which tend
to be transformed in habits and even in automatisms.

– 14 –
11th IEEE International Symposium on Applied Computational Intelligence and Informatics • May 12-14, 2016 • Timişoara, Romania

Input Interface:
Establish the AC task Strategic Level

Behaviors: Program Manager:


Analyze the task versus driving
Generate behaviors; circumstances (DC);
Import the behaviors; Decide to run the appropriate behavior
Improves the behaviors;
Store behaviors;
Initiate behaviors;
Filter DC:
Construct the driving circumstance (DC);

Tactical Level

Output Interface: Actuator Sensors:


Output the state of the AC; Communications: Read the state of the AC and
Trigger the actuators the environment characteristics

Operational Level

Figure 3. The proposed architecture

The “Driving Behaviors” is a collection of behaviors The “Human Driving Behaviors” is a set of linked
(the driver’s behavior for car ignition, the driver’s programs (behaviors) which are stored in a memory. The
behavior for stop the car, etc.). manager program decides to run a certain program. This
The “Fundamental truths” have been inferred from the decision is possible knowing the driving goal and on the
“Driving Behavior” concept and have been used, like environment conditions which we have named driving
Meta Knowledge, in the control architecture design: circumstance (DC).
Each behavior is a state machine, a set of actions.
1. A priori, the driver establishes the current These actions compute parameters which are used in
driving goal; actuators triggering
2. A behavior is a set of actions; Some comments are necessary (see Figure 3):
3. The behaviors are linked together, creating a • The strategic level is an interface where the user
system which allows the driver to obtain define the AC task
solutions in the driving circumstance; • The AC analyzes the task, using the “Program
4. The change from one behavior to another is Manager”, compare it with the DC and decide to
triggered by an occurrence event; run a certain “Behavior”;
5. This system is developed by learning - • The DC have been defined by sensors fusion in
experience; the Filter DC subsystem;
6. Behaviors presume decisions with an incomplete • The “Behaviors” subsystem is used to:
set of information; o Generate behaviors i.e. learn new
7. In time, these sets of actions tend to be behaviors
transformed in habits and automatisms. o Import the behaviors from the AC
network
Transferring the previous propositions in to the o Improves the behaviors i.e. correct and
Milchon architecture we obtain (see Figure 4) the optimize actions which are already
strategic level, which models the phenomenon when the included or add new actions;
driver establishes his goal; the tactical level, which
o Store behaviors;
models the driver acts when he most find the solution in
order to accomplish the goal and the operational level, o Initiate the behaviors proposed by the
which models the implementations of these solutions. Program Manager

– 15 –
C. Pozna, Cs. Antonya • Issues about Autonomous Cars

• The “Output Interface” allows the monitoring of the “Failure” events are signaled. The meaning of
the states ; “Crisis” is that a new Behavior is necessary. “Failure”
• The “Actuators Communications” outputs means that in the collection of the Behavior system no
commands to the actuators microcontrollers; proper Behaviors can be found other than to stop the AC
safely.
We have constructed and tray a part of the proposed
architecture in [18]. In that case the collection of III. CONCLUSIONS
behaviors has had a specific typology presented in figure
4.
Three different structures have been used: “The basic The proposed control architecture is composed from a
behaviors”, “The error behaviors” and “The simple strategical level a tactical level and an operational level.
behaviors”. The structural differences between these The first level task is to establish the AC task i.e. an
programs are the input output connections (P previous, N interface between the user and the AC; the second level
next, E error, QI quick in, QO quick out) and the mission is to solve situations which are probable but there
information flow direction. are not anticipated by the previous level (a pedestrian
which cross the road, an illegally maneuver of another
driver etc.); the last level use the task defined from the
Basic behavior
E tactical level and control the AC actuators, read the
sensors etc.
P
Action 1 Action n N The current issues of AC design consist on solving
unexpected driving problem, these issues refers to the
second level of the proposed architecture. From our point
Error behavior of view solving a tactical problem require the knowledge
of a certain driver behavior. A behavior is a well-defined
Action 1 Action n set of (here driver) actions. In turn an action is a basic
driver maneuver for example: pushes the brake pedal,
E
steers the wheels with a particular angel etc. The tactical
Simple behavior level will trigger a particular behavior in order to solve a
particular locomotion problem.
Action 1 Action nE We can model the behavior with a discrete system so
P N the state machine is one of the suitable mathematical
models. This means that the actions will be modeled like
QIn QO states. A specific network of states will define a behavior.
We imagine that the tactical level is an expert system
which manipulates behaviors according to its rules.
Figure 4. Different Behavior structures Knowledge consists on rules and behaviors (proposed
solutions).
In this way the challenges of tactical level design are
transferred to the expert system design. Here the keystone
AC Task problem is to avoid the expert system paradox: the
DC incompleteness of rules. For the AC, the incompleteness
refers to the behaviors set. There are always locomotion
problems which cannot be solved by the known set of
behaviors.
Our proposed solution to this issue is to acquire new
knowledges. We consider that the locomotion is a source
Continue to OK FAILURE Stop the of experience which can be transformed in knowledges
run the AC using learning technics. Once again the human driver
program example is follow. Starting with a basic collection of
CRISIS driving behaviors the human driving will acquire new
skills using its experience.
The challenges of tactical level design are now
Change the Behavior transferred to the learning system design. The learning
will improve continuously the capacity of the expert
system which will manage solutions for more and more
Figure 5. The Program Manager structure complex locomotion problem.
Even there are many well-known machines learning
techniques - supervise, unsupervised or reinforcement -
Comparing the AC task with the driving circumstance we consider that the experience itself is the hard problem
(DC) the “Program Manager” decides, about the to be solves. In [20] we have proposed a learning system
Behavior which will be run (see Figure 5). No changes used for an artificial intelligence agent. This system
are too made as long as the current behavior solves the combines perceptions, observations and experience.
task and DC demands. On the contrary, the “Crisis” or

– 16 –
11th IEEE International Symposium on Applied Computational Intelligence and Informatics • May 12-14, 2016 • Timişoara, Romania

From the Driver Behaviors collections in [22] we have [7] Chagas, C., M., Koike, C., Bayesian Approach to Action Selection
developed the trajectory following and the obstacles and Attention Focusing An Application in Autonomous Robot
Programming. Doctoral Thesis l’École Doctorale Mathématiques,
avoiding Behaviors. Sciences et Technologies de l’Information, Informatique , 2005
[8] Pradalier, C., Navigation intentionnelle d’un robot mobile.
Doctoral thesis, l’École Doctorale Mathématiques, Sciences et
ACKNOWLEDGMENT Technologies de l’Information, Informatique, 2004
The presented research studies are sponsored by the [9] Ackerman, J., “Robust control for automatic steering” Proc. Amer.
Conf., San Diego, CA, 1990, pp. 795-800
TAMOP-4.2.2.C-11/1/KONV-2012-0012: Smarter
Transport - IT for co-operative transport system. The [10] Antsalkis, P., An introduction to Intelligent and Autonomous
Control, Kluver Academic Publisher Norwell 1993
Project is supported by the Hungarian Government and
[11] Layne, J. K.,”Fuzzy Learning Control for Antiskid Braking
co-financed by the European Social Fund. This work was System”, IEEE Trans. on Control Systems Techn. Vol. 1, No. 2,
supported in part by a by the Romanian Ministry of June 1993, pp. 122-129
Education, Research and Innovation through the PNII Idei [12] Jang, J., “Neuro-Fuzzy and Soft Computing” Prentice Hall 1997
project 842/2008. [13] Kosko, B., “Neural Network and Fuzzy Systems” Prentice Hall
1992
[14] Bengtsson, J. Adaptive Cruise Control and Driver Modeling.
REFERENCES Department of Automatic Control Lund Institute of Technology
Lund, November 2001
[1] http://www.driverless-future.com/
[15] Al-Shihabi, T., Mourant, R., A Framework for Modeling Human-
[2] http://robots,stanford.edu/papers/thurn. S. Thrun, M. Montemerlo,
like Driving Behaviors for Autonomous Vehicle in Driving
H. Dahlkamp, D. Stavens, A. Aron, J. Diebel, P. Fong, J. Gale,
Simulators. Proceedings 5th International Conference on
M. Halpenny, G. Hoffmann, K. Lau, C. Oakley, M. Palatucci,
Autonomous Agents, June 2001, 286-291
V. Pratt, P. Stang, S. Strohband, C. Dupont, L.-E. Jendrossek,
C. Koelen, C. Markey, C. Rummel, J. van Niekerk, E. Jensen, [16] Quispel, L., Automan, a psychologically based model of human
P. Alessandrini, G. Bradski, B. Davies, S. Ettinger, A. Kaehler, driver. Experimental and Work Pscychology. Department of
A. Nefian, and P. Mahoney.Stanley the robot that won the Psychology, University of Groningen Grote Kruisstraat 2/1, 9751
DARPA Grand Challenge. Journal of Field Robotics MN Groningen
[3] Coué, C., Pradalier, C., Laugier, T., Bessiere, P., Bayesian [17] Ungoren, A., An Adaptive Lateral Preview Driver Model. Vehicle
Occupancy Filtering for Multitarget Tracking: An Automotive System Dynamic
Application. The International Journal of Robotics Research Vol. [18] Pozna, C., Troester, F.: Research on the ACC autonomous car,
25, No. 1, January 2006, pp. 19-30 Journal of Automation, Mobile Robotics & Intelligent Systems,
[4] Pomerleau, D. (1990): Neural Network Based Autonomous Vol. 1, No. 1. pp. 32-40, 2007
Navigation. Vision and Navigation: The CMU Navlab, C. Thorpe [19] Salvucci, D., Modeling Driver Behavior in a Cognitive
(Ed.), Kluwer Academic Publishers, Boston Architecture. In Press, Human Factors February 16, 2005
[5] C. Pradalier, C., Hermosillo, J., C. Koike, C., Braillon, C., [20] Pozna, C., A Phenomenological Analyze Trial of AI Syntagma, In
Bessihre, P., Laugier, C., An Autonomous Car-Like Robot Proc. of the 7th International Symposium of Hungarian
Navigating Safely Among Pedestrians Preceding of the 2004 IEEE Researches, 2006, pp. 159-165
1ntermational Conference on Robotics& Automation New Orleans [21] Pozna, C.,Troester, F., Collision Avoidance Trajectory Design,
- April 2004 15th International Workshop on Robotics in Alpe-Adria-Danube
[6] Braillon., C., Kane Usher, K., Pradalier, C., Crowley, J., Laugier, Region RAAD, 2006
C., Fusion of stereo and optical flow data using occupancy grids [22] http://autocarsim.com/
Proc. of the IEEE Int. Conf. on Intelligent Transportation Systems
- Toronto (CA) - September 2006

– 17 –
C. Pozna, Cs. Antonya • Issues about Autonomous Cars

– 18 –

You might also like