Microsoft Word - brain_vol8_issue3_v6_ok1.docx

Validation of Enhanced Emotion Enabled Cognitive Agent Using Virtual
Overlay Multi-Agent System Approach

Faisal Riaz

Department of Computer Sciences & IT,
Mirpur University of Science & Technology, AJ&K,

College Rd, New Mirpur City 10250, Pakistan
Phone: +92 582 7961037

Department of Computing and Technology,
Iqra University,

5, Khayaban-e-Johar, Islamabad, Pakistan
fazi_ajku@yahoo.com

Muaz A. Niazi

Department of Computer Sciences,
COMSATS Institute of Information Technology,

Park Road, Islamabad 45550, Pakistan
Phone: +92 51 9247000

muaz.niazi@gmail.com

Abstract
Making roads safer by avoiding road collisions is one of the main reasons for inventing

Autonomous vehicles (AVs). In this context, designing agent-based collision avoidance components
of AVs which truly represent human cognition and emotions look is a more feasible approach as
agents can replace human drivers. However, to the best of our knowledge, very few human emotion
and cognition-inspired agent-based studies have previously been conducted in this domain.
Furthermore, these agent-based solutions have not been validated using any key validation
technique. Keeping in view this lack of validation practices, we have selected state-of-the-art
Emotion Enabled Cognitive Agent (EEC_Agent), which was proposed to avoid lateral collisions
between semi-AVs. The architecture of EEC_Agent has been revised using Exploratory Agent
Based Modeling (EABM) level of the Cognitive Agent Based Computing (CABC) framework and
real-time fear emotion generation mechanism using the Ortony, Clore & Collins (OCC) model has
also been introduced. Then the proposed fear generation mechanism has been validated using the
Validated Agent Based Modeling level of CABC framework using a Virtual Overlay MultiAgent
System (VOMAS). Extensive simulation and practical experiments demonstrate that the Enhanced
EEC_Agent exhibits the capability to feel different levels of fear, according to different traffic
situations and also needs a smaller Stopping Sight Distance (SSD) and Overtaking Sight Distance
(OSD) as compared to human drivers.

Keywords: Autonomous Vehicles, Cognitive Agent, Emotions, SimConnector, VOMAS
Agent, Validation

1. Introduction
Making roads safer by avoiding road collisions is one of the main reasons for inventing

Autonomous vehicles (AVs) (S. Samiee, S. Azadi, R. Kazemi, A. Eichberger, B. Rogic, M.
Semmer, 2016). Vine et al. (S. Le Vine, …, 2016) have proposed collision free intersections using
autonomous cars by employing the clear distance ahead approach. Rodrıguez-Seda et al. (E. J.
Rodríguez-Seda, D. M. Stipanović, M. W. Spong, 2016) have developed collision–free maneuvers
in uncertain road environments for autonomous cars using Lyapunov-based analysis. Jimenez et al.
(F. Jiménez, J. E. Naranjo, Ó. Gómez, 2015) have made an autonomous car capable of avoiding
road collisions from pedestrians and surrounding vehicles using laser-scanner sensor and detailed
digital map. However, designing collision avoidance component of AVs inspired by agent based

BRAIN: Broad Research in Artificial Intelligence and Neuroscience
Volume 8, Issue 3, September 2017, ISSN 2067-3957 (online), ISSN 2068-0473 (print)

modeling, which truly represents human cognition and emotions, looks more feasible as they are
replacing human drivers.

Researchers have proposed many agent based collision avoidance systems. Reichardt (D. M.
Reichardt, 2008) has presented an emotional agent inspired driver’s assistant model which
simulates the emotional influence on the human driver’s behavior. The main purpose of this model
is to build a framework for learning algorithms, which help in building adaptive driver assistance
system. An Emotion Enabled Cognitive Agent (EEC_Agent) inspired lateral collision avoidance
scheme between AVs has been proposed by Riaz et al. (F. Riaz, …, 2013). A detailed analysis of
Cyberphysical systems for collision avoidance has also been proposed by Riaz and Niazi (F. Riaz,
M. A. Niazi, 2016). An in-vehicle virtual agent based driver assistance system has been proposed
by Joo and Lee-Won (Y. K. Joo, R. J. Lee-Won, 2016) , which help female drivers to improve their
performance during risky situations. A human behavior inspired agent has been proposed by
Waizman et al. (G. Waizman, S. Shoval, I. Benenson, 2015), which helps the AVs to avoid
collisions with pedestrians at black spots. An agent based Driver Assistance System (DAS) has
been proposed by Tiengo et al. (W. Tiengo, E. B. Costa, J. M. Fechine, 2016) for the rollover
prevention in the heavy duty vehicles. In another work, a rule based cognitive Agent inspired
intersection collision avoidance system has been proposed by Lu et al. (G. Lu, …, 2014). However,
the problem with these agent-based solutions is that, the claimed functionalities of these agent-
based systems have not been validated at any level.

In existing literature, different validation techniques for agent-based systems have been
proposed by the researchers. These include work on modeling the internet of things such as (K.
Batool, M. A. Niazi, 2017; S. Laghari, A. Niazi, 2016; A. B. Altamimi, R. A. Ramadan, 2016). For
complex-network based models, a validation methodology has also been proposed by Batool and
Niazi in (K. Batool, M. A. Niazi, 2014). Agents have also been demonstrated to be useful in the
domain of multi-agent foraging (O. Zedadra, N. Jouandeau, H. Seridi, G. Fortino, 2017). Fagiolo et
al. (G. Fagiolo, C. Birchenhall, P. Windrum, 2007) have proposed an empirical validation technique
to validate the agent-based systems. In another research work, a validation scheme based on
philosophical truth theories has been proposed by Schmid (A. Schmid, 2005). Barreteau et al. (O.
Barreteau,… , 2003) have proposed an iterative participatory approach, known as companion
modeling for the validation of agent based systems. Makowsky (M. Makowsky, 2006) has proposed
the agent based simulation itself as a validation technique to validate the functionality of proposed
agent system. Niazi et. al (M. A. Niazi, A. Hussain, M. Kolberg, 2017) have proposed a novel
concept of agent based validation using Virtual Overlay Multi-Agent System (VOMAS) under the
framework of Cognitive Agent Based Computing (CABC) framework (M. A. Niazi, A. Hussain,
2012). However, none of these above mentioned validation techniques have been applied to validate
such cognitive and emotional agents, which have been tailored especially for autonomous vehicles
to enhance their collision avoidance capabilities.

Contribution: The main contribution of this research work is to validate the existing state-of
the-art EEC_Agent proposed by Riaz et al. (F. Riaz, …, 2013). However, some secondary
contributions have been made as well, which were necessary to perform its validation. The details
of the contribution are given as under.

1) The enhanced version of EEC_Agent known as Enhanced Emotion Enabled Cognitive
Agent (EEEC_Agent) has been proposed by introducing proper emotion generation
mechanism using Ortony, Clore & Collins OCC model (A. Ortony, G. L. Clore, A. Collins,
1990), as it lacks in simple EEC_Agent (F. Riaz, …, 2013). For this purpose the Exploratory
Agent Based Modeling (EABM) level of CABC framework has been employed.

2) The validated Agent Based Modeling level of CABC framework has been utilized to
validate the EEEC_Agent functionalities. To validate the EEEC_Agent Virtual Overlay
Multi-Agent System (VOMAS) approach has been employed for comparing and validating
the performance of EEEC_Agent with that of a human driver during the rear end collision
situation. For comparison and validation of EEEC_Agent, Stopping Sight Distance (SSD)

F. Riaz, M. A. Niazi - Validation of Enhanced Emotion Enabled Cognitive Agent Using Virtual Overlay Multi-
Agent System Approach

defined by the American Association of State Highway and Transportation
Officials (AASHTO) (A. Aashto, 2001) and Overtaking Sight Distance (OSD) defined by
Indian Road Congress (IRC) (M. Sarhan, Y. Hassan, 2008) are used.
The extensive experiments prove that EEEC_Agent enabled AVs can avoid rear end and

lane changing collisions with smaller SSDs and OSDs respectively as compared to the human-
driven vehicles. Hence, then validated and truly emotional, cognitive agent-based collision
avoidance solution for the autonomous vehicles is revealed.

The rest of the paper has been organized as follows. Section 2 presents the background.
Section 3 discusses EABM based design of proposed EEEC_Agent. SimConnect design of the
proposed EEEC_Agent simulation has been presented in section 4. Section 5 presents the validation
model. Section 6 presents experimental setup. Section 7 gives results and discussion along with the
details of experiments. Practical validation of the proposed fear generation mechanism of
EEEC_Agent has been performed in section 8 using prototype AV. The paper concludes in Section
9.

2. Background
This background section gives the preliminary information about the key terms utilized in

this research work. First of all, the role of emotions in human life has been elucidated along with the
OCC model of emotions. Subsequently, the terms sight distances, Cognitive agent based computing,
SimConnector and VOMAS agent have been discussed.

2.1. Emotions
The term ‘emotion’ has been used to refer mental and physical course of action that includes

aspects of subjective experience, evaluation and appraisal, motivation and body responses such as
arousal and facial expression (E. E. Smith, 2009). According to Aristotle, emotion is defined as a
hat that leads one’s state to become so transformed that his judgment is affected, which is
accompanied by pleasure or pain (E. E. Smith, 2009) .There are several emotion models, which are
proposed by different researchers. However, we are interested in the OCC model (A. Ortony, G. L.
Clore, A. Collins, 1990).The reason for choosing OCC is the primary interest of its authors in the
role of cognition which is used to generate emotion. According to OCC model, the emotions may be
generated due to three major aspects of the world or changes in the world, namely events, agents, or
objects. When humans focus on events they are interested in their consequences, whereas they focus
on agents and actions, they are interested in their actions. In our work, we are interested in emotions
generated due to the reactions to the expected events. These types of emotions are also known as
prospect based emotions.

2.2. Cognitive Agent-Based Computing (CABC)
Agent-based modeling (ABM) and complex networks (CN) are two popular modeling tools

for understanding Complex Adaptive System (CAS). In 2011, a unified framework named
Cognitive Agent-based Computing (CABC), combining these two modeling paradigms, was
proposed by Muaz et al. (M. A. Niazi, A. Hussain, 2012) for the better understanding of CAS.
Agent based modeling (ABM) and complex networks (CN) are two popular modeling tools for
understanding Complex adaptive system (CAS). The CABC helps the cross-disciplinary researchers
to develop the understanding of their area related CAS using different types of models. It provides
guidelines to the multidisciplinary researchers regarding how they can develop computational
models of CAS even though they belong to social science, life science or computer science. The
unified framework provides four understanding and development levels of CAS along with related
case studies.

Complex Network Modeling. The first level of framework, which is useful in modeling
Complex systems, is complex network modeling. When the interaction data between network nodes

BRAIN: Broad Research in Artificial Intelligence and Neuroscience
Volume 8, Issue 3, September 2017, ISSN 2067-3957 (online), ISSN 2068-0473 (print)

is available, then using complex network modeling can be useful. This level helps the researchers in
building the complex network models along with the network classification. Further this level helps
in extracting the useful information from the network by determining the global and local
quantitative measures related to this network. Other statistical and more traditional mathematical
models have not such capability to provide details of emergent behavior and patterns of complex
networks, which can be achieved by complex network modeling.

Exploratory Agent Based Modeling. The second level of framework is Exploratory Agent
Based Modeling (EABM). When the researchers are interested in extending existing ideas related to
the agent based modeling which belong to the other fields, the EABM is a useful guideline
paradigm in this regard. Using EABM, researchers can build experimental or proof of concepts,
which help in defining the further scope and feasibility of the future research. Using EABM
researchers still failed to solve some important problems.

DREAM. Using DREAM level, quantitative comparison of different models can be
performed without execution of simulation experiments. The processes of reverse engineering and
replication of model can be easily done using DREAM. Hence the ABM can be examined visually
other than textual description. The visual examination of ABM is better in the sense that a model
can be analyzed abstractly without visiting its source code. Hence using visual approach the
comparison between domains across models can be done easily along with teaching cas models.
Another ultimate benefit of DREAM is translation of different sub-models from visual model to
pseudo code specification model to an agent-based model. The proposed methodology should allow
for a translation from these different sub-models such as from a visual (Complex network-based)
model to pseudo code specification model to an agent-based model.

Validated Agent Based Modeling. Verification and validation are the techniques used for
the evaluation of a product or system to determine that it meets its objectives, requirements and
specification. These approaches are used together but they are different from each other. The
validated agent-based modeling level of the proposed framework is concerned with developing
verified and validated agent-based models. This level allows performing in-simulation verification
and validation of the agent-based models using a Virtual Overlay Multi-agent System (VOMAS).
To solve this problem Muaz et.al proposed a novel concept of Virtual Overlay Multi-Agent System
(VOMAS) in (G. Nehate, M. Rys, 2006). One of core benefits of VOMAS is that all kinds of agent
based models can be validated by using it. The basic idea of VOMAS is performing validation by
making an overlay on the top of agent based simulation without taking an active part in the
simulation. Using VOMAS, agent based simulations can be validated both spatially and non-
spatially.

In this research paper we have utilized two levels of CABC framework. The EABM, which
helps us in exploring the role of the OCC model in enhancing the emotion generation capabilities of
EEEC_Agent and Validated agent based modeling level, which provides us guidelines to validate
the functionality of EEEC_Agent.

2.3. Sight Distances
Sight distances are the different types of safe distances between the vehicles which should

be maintained to avoid collisions. Sight distances are further subdivided in stopping sight distance
and overtaking sight distance.

Stopping Sight Distance (SSD). The minimum sight distance at any point which enables
the driver to stop the vehicle safely without the collision is known as stopping sight distance. It is
one of the very basic measures in traffic engineering defined by the (AASHTO) (C. Harlow, S.
Peng, 2001). SSD basically comprises two distances. The first one is the distance that a vehicle
travels during the reaction time of the driver and the second one is the distance that vehicle takes to
stop after applying the brakes.

F. Riaz, M. A. Niazi - Validation of Enhanced Emotion Enabled Cognitive Agent Using Virtual Overlay Multi-
Agent System Approach

Overtaking Sight Distance (OSD). The Overtaking Sight Distance (OSD) is the distance
in front of the driver of a vehicle while trying to overtake against opposite direction. IRC has given
a method to compute OSD in (G. Nehate, M. Rys, 2006).

2.4. VOMAS
Agent-based models are getting popular due to their easy application to the different fields

of life, such as modeling human muscle development, heat diffusion, Turing machine, flocking and
different social science mechanisms, etc. In spite of such popularity of agent-based model,
validation of these models is still a challenging task. During the modeling of agent-based models, it
is important to ensure that the model is working correctly, i.e. verification and the model is giving
required outputs i.e. validation. To solve this problem, Muaz et al. (M. A. Niazi, Q. Siddique, A.
Hussain, M. Kolberg, 2010) proposed a novel concept of VOMAS. Using VOMAS, agent-based
simulations can be validated both spatially and non-spatially. In our work for the validation of
proposed agent-based model of EEC_Agent, we are using multi-invariant based validation method,
which lies in non-spatial validation category.

2.5. SimConnector
As the natural disaster events occur rarely,testing the performance of disaster alert systems

is a challenging task. To overcome this problem, a novel approach was proposed by Muaz et.al in
(M. Niazi, …, 2011) by developing and testing real-time disaster early warning and alerting system
by combining two different software environments. In (M. Niazi, Q. Siddique, A. Hussain, G.
Fortino, 2011), the agent-based simulation was used to generate a rare forest fire event and a web-
based alert decision support system for generating warnings. For a detailed study of SimConnector,
the interested readers are referred to (M. Niazi, Q. Siddique, A. Hussain, G. Fortino, 2011).

3. Improvement in the Existing Architecture of EEC_Agent Using EABM
In the introduction section, it has already been discussed that in the current work we are

going to improve the existing architecture of EEC_Agent (F. Riaz, S. I. Shah, M. Raees, I. Shafi,A.
Iqbal, 2013). The main drawback of EEC_Agent is that it claims to utilize emotions in making
collision avoidance decisions, but no proper emotion generation mechanism has been proposed,
which helps the cognitive agent to feel emotion according to the changing in the dynamic
environment. Hence to overcome this issue, we have redesigned the architecture of EEC_Agent
using EABM and explored the role of OCC model in the fear generation of EEC_Agent. In existing
literature OCC model has not been utilized to model the collision avoidance problems of AVs using
an agent approach. To our best knowledge, it is the first research work, which is going to explore
the role of OCC model in this context. The architecture and functionality of the EEEC_Agent are
discussed as follows.

Architectural Features and Functionality
The architecture of the proposed EEEC agent has been shown in figure 1. For the sake of

less complexity only two main neurons are considered i.e. Hypothalamus neuron (NH) and
Amygdala neuron (NA) because of their direct relation regarding the efficient processing in
emergency situations. In addition to generate the notion of fear in an EEC agent, the OCC Model
Based Fear Generation Module has been introduced (K. Batool, M. A. Niazi, 2014).

It can be seen from the figure 1 that the proposed architecture consists of five main modules:
Sensory module, Artificial Thalamus module, OCC Model Based Fear Generation Module, and
Motor module. To begin with, step 1, the Sensory module keeps track of the distance between
neighboring AVs on a road segment. The Sensory module of the EEC_Agent converts the stimulus
information into electrical signals. The Artificial Thalamus module, similar to the hypothalamus
module in the human brain, receives these sensed signals in step 2 for further processing. The

BRAIN: Broad Research in Artificial Intelligence and Neuroscience
Volume 8, Issue 3, September 2017, ISSN 2067-3957 (online), ISSN 2068-0473 (print)

signals received by Thalamus module have different frequencies that reflect on the prevailing inter-
vehicular distance. In step 3, these signals are then checked against the maximum allowed threshold
after computing likelihood, desirability and Ig variables using OCC Model Based Fear Generation
Module. In step 4 and 5, the Artificial Amygdala module also computes the fear intensity level in
coordination with Artificial Thalamus and OCC Model Based Fear Generation Modules. These
different intensity levels are used as controlling interrupts, which are passed to the Motor Module,
in step 6, and in turn it execute the suitable collision avoidance maneuver.

Figure 1. Emotion Enabled Cognitive Agent Architecture

4. Proposed SimConnector Design For Implementation of EEEC_Agent
The SimConnector approach proposed in (M. Niazi, Q. Siddique, A. Hussain, G. Fortino,

2011) consists of an agent-based simulation, which generates artificial disaster events and then
provided this data to the web-based decision support system for generating warnings.

Figure 2. SimConnector design for the implementation

However, in our case, agent-based simulation is used for validating the performance of
prospect based emotion, i.e. fear inspired road collision avoidance system in real road accident
scenarios. We first developed a fuzzy logic based simulator for finding the numeric values of fear
related variables such as Desirability, Likelihood, and Ig. These numeric values are provided to the

Prospect Based
Emotions
Variables

Fuzzy Logic
Inference

Engine

Crisp Values of
Prospect Based
Emotion (Fear)

SimConnector

Agent
Based
Mode

Emotions
Enabled
Collision

F. Riaz, M. A. Niazi - Validation of Enhanced Emotion Enabled Cognitive Agent Using Virtual Overlay Multi-
Agent System Approach

NetLogo platform based model of EEC_Agent for generating fear and testing its performance in
rear-end collisions. Fig. 2 depicts our SimConnector design.

5. EEEC_Agent Validation Model
As mentioned earlier, non-validated simulation and results are one of the drawbacks of

previously proposed EEC_Agent (F. Riaz, S. I. Shah, M. Raees, I. Shafi, A. Iqbal, 2013). To
overcome this problem, we have proposed VOMAS agent-based simulation validation
methodology. First, we have elucidated the VOMAS agent design, and then we have raised the
validation question. In the end, three invariants have been defined which will act as filters to
validate the agent-based simulation of EEEC_Agent.

VOMAS Agent Design - In our simulation model, we have designated AV as a VOMAS
agent. The VOMAS agent computes the changes in AV’s fear, according to the OCC model defined
fear related variables and computes the required SSD and OSD distances required for efficient
collision avoidance during rear end and overtaking scenarios respectively.

5.1. Validation Question and Invariants
Our validation question can thus be defined as. “How can we validate that the emotion

generation mechanism of EEEC_Agent is working properly and EEEC_Agent installed AV avoids
the rear end and overtakes collisions more efficiently than human drivers.” To find out the answer
of these validation questions Invariants methodology proposed by Niazi et al. (M. A. Niazi, Q.
Siddique, A. Hussain, M. Kolberg, 2010) has been utilized and in this regard following four
invariants have been defined. The first two invariants help validating that the proposed fear
generation mechanism of EEEC_Agent is working properly. The third and fourth invariants help
validating that EEEC_Agent installed AV requires smaller SSD and OSD during rear end collision
avoidance and overtaking scenarios respectively as compared to - human drivers.

a) Invariant1_TypeA
If the pre-condition that “Distance between the rear end of the first AV and the front end of

the second AV is very small” is true, then the fear level exhibited by AV would result in a post-
condition of “Intensity of a fear of a bullet autonomous vehicle is high or very high”.

b) Invariant1_TypeB
If the pre-condition that “Distance between the rear end of the first AV and the front end of

the second AV is decreasing” is true, then the variation in the distance of AV would result in a post-
condition of “Intensity of a fear of a bullet autonomous vehicle is increasing accordingly”.

c) Invariant2
If the pre-condition that “Bullet_Agent autonomous vehicle is successfully reacting to the

rear end collision threat using EEEC_Agent short route reaction time” is true, then stopping sight
distance required would result in a post-condition of “EEEC_Agent requires smaller SSD as
compared to the SSD required by Human driver”.

d) Invariant3
If the pre-condition that “Bullet_Agent autonomous vehicle is successfully reacting to the

overtaking collision threat using EEEC_Agent short route reaction time” is true, then overtaking
sight distance required would result in a post-condition of “EEEC_Agent requires smaller OSD as
compared to the OSD of the human driver”.

BRAIN: Broad Research in Artificial Intelligence and Neuroscience
Volume 8, Issue 3, September 2017, ISSN 2067-3957 (online), ISSN 2068-0473 (print)

6. Experiments
This section describes the experiments related to the quantitative computation of prospect-

based emotion using fuzzy logic and the experiments related to the validation of EEC_Agent.

6.1. Experiment 1
To compute the fear, we have built a Mamdani fuzzy inference system, which uses the

traceability algorithm defined in (S. Laghari, A. Niazi, 2016).

If Prospect (v, e, t) and Undesirable (v, e, t) < 0
Then set Fear-Potential (v, e, t) = ff [|Desire (v, e, t) |, Likelihood (v, e, t), Ig (v, e, t)]
If Fear-Potential (v, e, t) > Fear-Threshold (v, t)
Then set Fear-Intensity (v, e, t) = Fear-Potential (v, e, t) - Fear-Threshold (v, t)
Else set Fear-Intensity (v, e, t) =0

Implementation Details Of Fuzzy Logic To Compute The Numeric Values of the Fear

Emotion

For the sake of brevity, we are just giving here short details of the Mamdani fuzzy inference

system for the computation of different intensities of fear. The details of Likelihood variable are
given in the coming section along with Linguistic tokens and fuzzy rules.

Sub-Experiment 1a
The likelihood variable helps in computing the chances of an accident. Suppose that the AV

is following a truck. If the distance between AV and the truck is higher or equal to the SSD than the
likelihood of accidents will be low and vice versa. After a detailed analysis, it has been observed
that distance and speed are two such factors which directly affect the likelihood of an accident. For
example, if the speed is high and distance is low, then the likelihood of an accident will be high, and
on the other hand, if the speed is low and distance is high then the likelihood of an accident will be
low. The experiment 1a has been carried out to compute Likelihood variable using fuzzy inference
editor as shown in Fig. 3. The two input variables Distance and Speed are defined on the left side
and the Likelihood variable on the right side. The mathematical function of TRIMF has been
utilized as a membership function for two input and one output variables. To compute the
Likelihood variable, five linguistic tokens VLLH, LLH, MLH, HLH and VHLH have been defined
which represent Very low likelihood, Low likelihood, Medium likelihood, High likelihood and
Very High likelihood respectively.

Table 1. Linguistic tokens for Likelihood

Linguistic tokens Description
VHLH Very High Likelihood
HLH High Likelihood
MLH Medium Likelihood
LLH Low Likelihood

VLLH Very Low Likelihood

Twenty-five rules were defined to obtain the value of the variable likelihood, these rules are

presented in Table 2.

F. Riaz, M. A. Niazi - Validation of Enhanced Emotion Enabled Cognitive Agent Using Virtual Overlay Multi-
Agent System Approach

Table 2. Fuzzy Rules defined for Likelihood

If Distance is And Speed is Then Likelihood is
VHD VHS MLH
VHD HS LLH
VHD MS VLLH
VHD LS VLLH
VHD VLS VLLH
HD VHS HLH
HD HS MLH
HD MS VLLH
HD LS VLLH
HD VLS VLLH
MD VHS VHLH
MD HS VHLH
MD MS MLH
MD LS LLH
MD VLS VLLH
LD VHS VHLH
LD HS VHLH
LD MS HLH
LD LS MLH
LD VLS VLLH

V LD VHS VHLH
V LD HS VHLH
V LD MS VHLH
V LD LS HLH
V LD VLS MLH

Figure 3. Main simulation screen for Likelihood computation

The remaining fear variables, i.e. Undesirability and Ig are computed on the same pattern.

BRAIN: Broad Research in Artificial Intelligence and Neuroscience
Volume 8, Issue 3, September 2017, ISSN 2067-3957 (online), ISSN 2068-0473 (print)

6.2. Experiment 2
The purpose of the second type of experiment is to perform the validation of SimConnector

based design of the EEEC_Agent using VOMAS agent. For this purpose, Netlogo 5.3 has been
utilized which is a standard agent-based simulation environment. The NetLogo 5.1 environment
consists of patches and turtles.

6.2.2. Experimental Parameters
To perform the simulation experiments empirically, three types of simulation parameter sets

are defined. The first set consists of numeric values (Table 9, 10 and 11) of prospect based emotions
(i.e. Fear) variables like the likelihood of accident event, the undesirability of accident event, and
Ig. The second set of parameters consists of Stopping Sight Distance and overtaking sight distance
described by eq. (1) and (2).

V
VtSSD

2075.1
47.1  (1)

where SSD = Stopping Sight Distance in feet, V = design speed in mph, t = brake reaction
time in seconds and a = deceleration rate, 11.2 ft/s2.

s
VstVOSD bb

4
2  (2)

Here Vb = velocity of overtaking Vehicle, t = Reaction time, S=space before and after

overtaking and a = Maximum overtaking acceleration at different speeds.
The third miscellaneous set of parameters is:

- Number of Agents
- Number of VOMAS agents
- Testing Speed range (mph)
- EEEC_Agent Status
- Reaction Time

Here 0.4397 seconds and 3.8085 seconds are the reaction times for taking collision
avoidance maneuver by EEEC_Agent and human driver respectively. These reaction times are
taken in the guidelines of (Riaz et. al., 2013).

Figure 6 shows the experimental environment along with input and output parameters. The
two AVs are taking part in this validation simulation. The bullet AV is acting as a VOMAS agent
whereas the second one is leading AV which acts as a target agent. The left side of the simulation
world contains input sliders for providing fuzzy logic based numeric values of prospect based
emotion variables (Undesirability, Likelihood, Ig). It is important to recall here that these numeric
values of prospect based emotions were computed through experiments a), b) and c), presented in
section 5.1.1, using fuzzy logic and then provided to the agent based simulation using the proposed
SimConnector approach. The world size used in the simulation is (-25, -25) to (25, 25) and in this
way the total number of patches in the world is 25. To map the real-world distance in feet on 25
patches, each patch is representing a value equal to 100 feet.

F. Riaz, M. A. Niazi - Validation of Enhanced Emotion Enabled Cognitive Agent Using Virtual Overlay Multi-
Agent System Approach

Figure 6. Main Simulation Screen of EEEC_Agent based Collision Avoidance system in NetLogo
Environment

a) Experiments-Invariant1_TypeA
In this experiment, we have validated the invariant 1. For this purpose, six different types of

tests with different parameter values have been designed as presented in Table 7. These tests are set
up to test the pre-condition that if the distance between both AVs will be decreased, what will be its
effect on the intensity of the fear of EEEC_Agent. This set of tests has been performed using the
behavior space tool of NetLogo 5.3.1 environment and each test has been repeated 50 times.

BRAIN: Broad Research in Artificial Intelligence and Neuroscience
Volume 8, Issue 3, September 2017, ISSN 2067-3957 (online), ISSN 2068-0473 (print)

Table 7. Parameter values for Experiment-Invariant1_TypeA

Experime
nt No

Separati
on

Min
Veloci

Max
Velocity

Accelerati
on_

Bullet

Decelerati
on_Bullet

Accelerati
on_

Target

Decelerati
on_Target

EEE
C_Ag

ent
1 1 10 100 0.06 0.03 0.03 0.03 True
2 1 10 100 0.06 0.06 0.03 0.03 True
3 1 60 100 0.06 0.03 0.03 0.03 True
4 1 60 100 0.06 0.06 0.03 0.03 True
5 1 90 100 0.06 0.03 0.03 0.03 True
6 1 90 100 0.06 0.06 0.03 0.03 True

b) Experiments-Invariant1_TypeB
To validate the invariant 1_TypeB, a different set of experiments has been defined. This

experiment consists of five further tests. These five tests have been designed to check the behavior
of Bullet AV (EEEC_Agent) on different distances from the Target AV. The parametric values of 5
tests are presented in Table 8.

Table 8. Parameter values for Experiment-Invariant1_TypeB

Experime
nt No

Separati
on

Min
Velocity

Max
Velocit

Accelerati
on_

Bullet

Decelerati
on_Bullet

Accelerati
on_

Target

Decelerati
on_Target

EEEC_
Agent

1 5 10 100 0.06 0.03 0.03 0.03 True
2 9 10 100 0.06 0.06 0.03 0.03 True
3 13 10 100 0.06 0.03 0.03 0.03 True
4 13 60 100 0.06 0.06 0.03 0.03 True
5 17 10 100 0.06 0.06 0.03 0.03 True

c) Experiment-Invariant2
In this experiment, we have validated the invariant 2. The experiments are set up to test the

pre-condition that if the Bullet autonomous vehicle is successfully reacting to the collision threat
using EEEC_Agent short route reaction time, then EEEC_Agent requires smaller SSD as compared
to the SSD required by Human driver. For this purpose, a total of 12 experiments have been
performed.

d) Experiment-Invariant3
In this experiment, we have validated the invariant 3. The experiment is set up to test the

pre-condition that if the Bullet autonomous vehicle is successfully reacting to the collision threat
using EEEC_Agent short route reaction time, the EEEC_Agent then requires smaller OSD as
compared to the OSD required by Human driver. For this purpose, the experiment has been
repeated 10 times with different speeds and distances between both AVs.

7. Results and Discussion
This section describes the results of both experiments 1 and 2. The results are compared

with the state of the art EEC_Agent proposed in [6].

7.1. Experiment 1
Table 9 shows the quantitative values of undesirability from very low (VL) to very high

(VH). The terms VLD, LD, MD, HD, and VHD represent very low desirability, low desirability,
medium desirability, high desirability and respectively very high desirability. If the agent has a
value between 0-0.24 for its undesirability of an event, then it can be interpreted as the very low
undesirability. However, from an abstract analysis, it can be noted that due to the fuzzy nature of

F. Riaz, M. A. Niazi - Validation of Enhanced Emotion Enabled Cognitive Agent Using Virtual Overlay Multi-
Agent System Approach

the fear emotion, the boundary of one intensity level mixes in the boundary of another intensity
level. Hence, the intensity levels lying between 0.24 and 0.5 will be interpreted as low
undesirability and lower than these values as the very low undesirability. In the same way, the other
intensity levels of undesirability variable can be interpreted.

In the same way, Table 10 and Table 11 are showing the five quantitative values for finding
the different intensity levels of likelihood and Ig variables.

These quantitative values of Desirability, Likelihood and Ig are presented in Table 9, 10 and
11 respectively. These values are then provided to the EEEC_Agent for computing different
intensities of fear in the next section by following the proposed SimConnector design.

Table 9 Quantitative Values of Five Intensity levels of Desirable Variable

VLD LD MD HD VHD
0-0.24 0.1-0.5 0.25-0.73 0.51-0.9 0.76-1

Table 10 Quantitative Values of Five Intensity levels of Likelihood Variable

VLL LL ML HL VHL
0-0.24 0.1-0.5 0.25-0.73 0.51-0.9 0.76-1

Table 11 Quantitative Values of Five Intensity levels of Global Variable (Ig)

VLIg LIg MIg HIg VIg
0-0.24 0.1-0.5 0.25-0.73 0.51-0.9 0.76-1

7.2. Experiment 2
In this section validation of fear generation mechanism of EEEC_Agent has been performed

using VOMAS agent methodology defined in the CABC framework.

7.2.1. Results and Discussion: Experiment-Invariant1_TypeA
For the detailed validation of proposed EEEC agent, we have designed two different sets of

experiments. Furthermore, these two sets of experiments consist of six and five tests respectively.
The details of these tests have been provided in the section 5.1.12. Before starting the discussion, it
is important to mention that the simulation for each test within experiments set 1 has been
performed for 100 ticks. However, due to the space limitations, the data of only 8-9 ticks have been
shown in the Tables 12 to Table 17. However, the graphs shown in figure 7 to figure 12 have been
generated over 100 ticks. Furthermore, to draw graphs more clearly, the intensities of fear have
been mapped from the range of [0-1] to the [0-100].

The first set of experiments has been designed to validate the performance of EEEC agent
on the very short distance between bullet and target AV. The results of the first test of experiments
have been presented in Table 12. In the first test, the bullet AV follows the target AV with low
speed, i.e. 10 mph with high acceleration rate, i.e. 0.06 mph and low deceleration rate i.e. 0.03 mph.

BRAIN: Broad Research in Artificial Intelligence and Neuroscience
Volume 8, Issue 3, September 2017, ISSN 2067-3957 (online), ISSN 2068-0473 (print)

Table 12. Computation of Fear at low speed
of Bullet AV

Speed=10 Acceleration=

0.06
Deceleration=
0.03

SSD Distance Fear Intensity
0.16 2.9 66
0.35 4.38 49
0.76 6.36 49
2.61 10.36 49
10.23 10.89 66
10.23 9.74 76
10.23 7.0 76
10.23 6.0 76
10.23 5.50 76

Figure 7. Computation of Fear at low speed of Bullet AV

With high acceleration and low deceleration

Table 13. Computation of Fear at low speed

of Bullet AV
Speed=10 Acceleration=

0.06
Deceleration=
0.06

SSD Distance Fear
Intensity

0.16 2.9 66
0.16 4.47 49
0.16 7.61 49
0.16 9.95 36
0.16 13.09 26
0.16 18.6 16
0.16 20.17 6
0.16 15.93 26
0.16 10.50 36

Figure 8. Computation of Fear at low speed of Bullet AV

With equally high acceleration and deceleration
Table 14. Computation of Fear at moderate

speed of Bullet AV

Speed=60 Acceleration=
0.06

Deceleration=
0.03

SSD Distance Fear
Intensity

3.83 2.87 76
5.87 3.71 76
9.45 3.66 76
10.23 3.39 76
10.23 2.96 76
10.23 2.83 76
10.23 2.68 76
10.23 2.55 76
10.23 2.40 76
10.23 2.27 76

Figure 9. Computation of Fear at moderate speed of

Bullet AV
With high acceleration and low deceleration

F. Riaz, M. A. Niazi - Validation of Enhanced Emotion Enabled Cognitive Agent Using Virtual Overlay Multi-
Agent System Approach

Table 15. Computation of Fear at moderate

speed of Bullet AV
Speed=60 Acceleration=

0.06
Deceleration=

0.06
SSD Distance Fear

Intensity
3.83 2.9 76
3.83 2.43 76
3.83 2.05 76
3.83 1.84 76
3.83 1.65 76
3.83 1.56 76
3.83 1.51 76
3.83 1.29 76
3.83 1.24 76

Figure 10. Computation of Fear at moderate speed of

Bullet AV with equally high acceleration and
deceleration

Table 16. Computation of Fear at high speed

of Bullet AV
Speed=90 Acceleration=

0.06
Deceleration=

0.03
SSD Distance Fear

Intensity
8.34 2.53 76
8.34 7.61 76
10.23 11.98 66
10.23 14.43 49
10.23 16.38 49
10.23 17.83 49
10.23 18.33 36
10.23 19.78 36
10.23 19.50 36

Figure 11. Computation of Fear at high speed of Bullet
AV with high acceleration and low deceleration

Table 17. Computation of Fear at high speed

of Bullet AV
Speed=9

0
Acceleration=

0.06
Deceleration=

0.06
SSD Distance Fear

Intensity
8.34 2.48 76
8.34 7.12 76
8.34 10.18 66
8.34 12.8 49
8.34 16.71 36
8.34 20.65 26
8.34 20.51 26
8.34 20.76 26
8.34 20.50 26

Figure 12. Computation of Fear at high speed of Bullet

AV with equally high acceleration and deceleration

BRAIN: Broad Research in Artificial Intelligence and Neuroscience
Volume 8, Issue 3, September 2017, ISSN 2067-3957 (online), ISSN 2068-0473 (print)

From the Table 12, it can be seen that at the beginning of travel the distance between bullet

and target AV is 2.9 feet and bullet AV requires 0.16 feet as SSD to avoid the collision. Because of
small differences between required SSD and current distance from target AV, bullet AV starts
feeling high fear i.e. 66. After feeling high fear, the bullet vehicle starts decelerating with the
deceleration rate of 0.03 mph and the distance between both AVs starts increasing. From the 2nd
entry of Table 12, it can be seen that due to increasing in distance from 2.9 feet to 4.38 feet the
bullet AV starts feeling medium fear i.e. 49. From the fifth entry of Table 12 it can be seen that
when the required SSD is very near to the distance between both AVs, bullet AV starts exhibiting a
high intensity of fear again and as the difference between required SSD and distance moves towards
high negative value then the bullet AV starts feeling very high fear. From these tests, it can be
validated that the proposed EEEC agent has the capability to feel abrupt fear as well due to the
sudden appearance of the leading vehicles on very short distance. The graphical representation of
the results of table1 has been shown in Figure 7.

Table 13 presents the simulation results of test 2. In test 2, the initial distance between the
bullet-target AVs and the speed of bullet AV has been considered same as test 1. However, the
acceleration and deceleration rates of bullet AV have been set to high rate i.e. 0.06 mph. From the
Table 13, it can be seen that in the beginning of the simulation, EEEC_Agent enabled bullet AV felt
high fear due to small differences between required SSD and current distance. In a result, bullet AV
starts decelerating with the rate of 0.06 mph. Due to deceleration maneuver, the distance between
both AVs starts increasing. In the result, the EEEC agent’s fear state switches from high fear to
medium fear i.e. 49. An interesting thing in this test can be noted that the SSD remains constant. It
is because of equal high acceleration and deceleration rates. From the 6th entry of Table 13, it can
be observed that when the distance between both AVs gets very high i.e. 18.6 feet (on simulation
scale) as compared to the required SSD i.e. 0.16 feet then the EEC agent fear state presents very
low fear i.e. 16. Again, the results of test 2 validate the performance of proposed EEEC agent over
very short, medium and high distances between bullet and target AVs. The graphical representation
of the results of table 13 has been shown in figure 8.

Table 14 presents the simulation results of test3, which is designed to test the behavior of the
EEEC_Agent over moderate speed, i.e. 60 mph with acceleration rate, i.e. 0.06 mph and low
deceleration rate i.e. 0.03 mph. From the first entry of table 14, it can be seen that over moderate
speed, bullet AV starts feeling very high fear at the very beginning of travel due to small differences
between required SSD and current distance. In the next move, bullet AV starts decelerating to avoid
the collisions and the distance between both vehicles starts increasing. But still, the EEEC_Agent
remains in the very high fear state because of the high danger of accident as compared to the test 1
and test 2 results, where over low-speed EEEC_Agent switches from high fear state to medium fear
state after performing deceleration maneuver. The results of test 3 and test 4 are almost the same.
However, the main difference is that in the test 4 the required SSD remains the same for each tick of
simulation due to equally high acceleration and deceleration rates. The graphical representations of
table 14 and table 15 are presented in figure 9 and figure 10 respectively.

Table 16 and 17 present the results of test 5 and 6 respectively. Test 5 and 6 are designed to
validate the performance of proposed EEEC_Agent over very high speed i.e. 90 mph with high
acceleration and low deceleration rates and with equal high acceleration and deceleration rates
respectively. From the table 16, it can be seen that the proposed EEEC_Agent switches between
different fear levels, according to the low or high differences between required SSD and current
distance. The graphical representations of table 16 and 17 are presented in figure 11 and figure 12
respectively. From all of these tests it can be concluded that the proposed EEEC_Agent has been
validated to have the capability to feel sudden fear over the different speeds with small initial SSDs.
Hence, these results validate the invariant1_Type A claim that if the pre-condition that
“Distance between the rear end of the first AV and the front end of the second AV is very small” is

F. Riaz, M. A. Niazi - Validation of Enhanced Emotion Enabled Cognitive Agent Using Virtual Overlay Multi-
Agent System Approach

true, then the fear level exhibited by AV would result in a post-condition of “Intensity of a fear of a
bullet autonomous vehicle is high or very high”.

7.2.2. Results and Discussion: Experiment-Invariant1_TypeB
Further validation of the proposed EEEC_Agent has been provided through an extensive set

of tests over different arrangements of experiments. Continuing the discussion section, the rest of
the validation scenarios have been presented through table 18-22. Five different sets of experiments
have been designed over different initial separation, covering a range from short to long distances
between bullet and target AVs. In Table 18, results have been given for validating the EEEC_Agent
by placing bullet and target AVs 5 separation apart. Bullet AV is moving at a low speed of 10 mph
and accelerating at a rate of 0.06 mph and decelerating with a low rate of 0.03 mph. At tick number
1, which is shown by the first record in table 18, bullet AV requires 0.16 feet SSD. Whereas, the
actual distance, i.e. 6.28 feet between vehicles is greater than the required SSD. That is why
medium level fear is felt by EEEC_Agent i.e. 49. As the bullet AV proceeds further by adding 0.06
mph to its current speed, a decrease in the distance has been recorded, which is shown by the
second entry of distance in table 18 i.e. 5.77 feet. This decrease in distance increased fear intensity
and shifted it to the high level. As bullet AV continues to accelerate the required SSD varies with
changes in speed on every tick. The fourth record in table 18 shows the status of bullet AV with an
increased SSD value which is 4.73 due to increasing in its speed. At this point, the autos’ separation
has crossed the safety sight distance limit, which is causing our EEEC_Agent to feel high positive
fear i.e. 76. After the violation of SSD, bullet AV tends to decelerate. The rest of the entries are
confirming the fact that deceleration causing an increment in distance and an ultimately decrement
in fear level. Graphical representation of all 100 ticks’ data has been provided in Figure 13.

Table 18. Computation of Fear at low speed
of Bullet AV with separation 5

Speed=10 Acceleration=
0.06

Deceleration=
0.03

SSD Distance Fear Intensity
0.16 6.28 49
2.61 5.77 66
3.95 4.60 66
4.73 3.83 76
10.22 7.61 76
10.22 11.98 66
10.22 14.15 49
10.22 18.15 36
10.22 17.50 36

Figure 13. Computation of Fear at low speed of

Bullet AV
With high acceleration and low deceleration

Table 19. Computation of Fear at low speed
of Bullet AV with separation 9

Speed=10 Acceleration=
0.06

Deceleration=
0.06

SSD Distance Fear
Intensity

0.16 10.24 36
0.16 12.21 26
0.16 13.27 26
0.16 16.44 16
0.16 19.49 16
0.16 19.60 16
0.16 20.14 160.15

Figure 14. Computation of Fear at low speed of

Bullet AV

BRAIN: Broad Research in Artificial Intelligence and Neuroscience
Volume 8, Issue 3, September 2017, ISSN 2067-3957 (online), ISSN 2068-0473 (print)

0.16 21.87 6 With equally high acceleration and deceleration

Table 20 Computation of Fear at moderate

speed of Bullet AV with separation 13
Speed=10 Acceleration=

0.06
Deceleration=

0.03
SSD Distance Fear

Intensity
0.15 14.84 26
0.46 16.84 16
5.57 23.28 16
7.47 23.39 26
10.22 21.68 36
10.22 18.17 49
10.22 14.15 66
10.22 11.28 66

Table 21 Computation of Fear at moderate

speed of Bullet AV with separation 13
Speed=60 Acceleration=

0.06
Deceleration=

0.06
SSD Distance Fear Intensity
3.83 14.9 36
3.83 16.04 26
3.83 20.03 16
3.83 24.02 6
3.83 22.99 16
3.83 19.30 26
3.83 15.31 36
3.89 11.59 49

Figure 15. Computation of Fear at moderate

speed of Bullet AV
With high acceleration and low deceleration

Figure 16. Computation of Fear at moderate

speed of Bullet AV with equally high acceleration
and deceleration

F. Riaz, M. A. Niazi - Validation of Enhanced Emotion Enabled Cognitive Agent Using Virtual Overlay Multi-
Agent System Approach

Table 22. Computation of Fear at high speed

of Bullet AV with separation 17
Speed=10 Acceleration=

0.06
Deceleration=
0.06

SSD Distance Fear
Intensity

0.15 18.37 16
0.15 20.21 6
0.15 22.44 6
0.15 24.12 6
0.15 25.35 6
0.15 20.32 6
0.15 19.88 16
0.15 18.92 16

Figure 17. Computation of Fear at high

speed of Bullet AV with equally high
acceleration and deceleration

Table 19 presents the figures regarding the validation test being performed with a low initial

speed of bullet AV, i.e. 10 mph with the initial separation of 9. Bullet AV has equally high
acceleration and deceleration rate of 0.06 mph. Initial separation has been increased by 4 points
than the previous setup. Staircase representation shown in figure 12 is substantiating the reality of
the increase in the distance causes a decrease in the fear intensity. Required SSD for bullet AV is
0.16 feet due to its low speed. Initial placement of the bullet and target vehicles is far apart. The
distance shown by the first record in table 19 is 10.24 feet, which has allowed bullet AV to move
continuously with mentioned rate by feeling positive low fear given by the value 36. An SSD
column of table 19 is baring a constant value due to an equal change in speed caused by both
acceleration and deceleration. Deceleration of bullet AV causing an increase in distance, for
instance, entry number 3 shows the distance value of 13.27 feet hence lowering the fear value to 26.
Fear continues to drop and that is because of the fact that bullet AV is decelerating without any
considerable fear intensity. Figure 14 presents a plot of 100 records to show the overall behavior of
EEEC_Agent.

In the continuation of the experiments, the next test has been performed with the initial
separation of 13. The results of this scenario have been shown in table 20 and table 21. Bullet AV is
moving with a speed of 10 mph. The first record in table 20 is giving 0.15 feet SSD, 14.85 feet as
initial distance and fear value equal to 26 i.e., low fear. Bullet vehicles accelerated, but being on
low speed as compared to target AV, an increase in the distance has been recorded given by the
second and third record in table 20. Gradual acceleration has put the bullet on a maximum speed of
100 mph and hence SSD required for such a high speed is given by 5th entry i.e. 10.22 feet. The rest
of the entries giving a constant figure of 10.22 feet for SSD, but gradual decrease in separation due
to the high speed of bullet AV. This continuing reduction in the distance value is causing uplift in
the fear intensity, exposing by 6th, 7th and 8th records of table 20. By maintaining the said initial
separation another test has been conducted by setting the initial speed of bullet AV to moderate
level i.e. 60 mph. Figure 16 depicts the truth of obtaining results that show a constant 3.83 feet SSD
value with different distances due to equal acceleration and deceleration rate. By examining the
table 21 records, it can be clearly validated that with the increase in distance between vehicles
caused (due to the deceleration of bullet vehicle or acceleration of target vehicle) a gradual decline
in fear values i.e. 0.36 to 0.26 and then finally to the 6. Record number 5 to 10 reveal the reverse of
earlier mentioned fact, as presented clearly in figure 16.

The last subject of discussion regarding experiment tests is describing the results by putting
vehicles distant apart, which is shown by the separation of 17 in the simulation. Bullet AV is
traveling at a speed of 10 mph with equally high acceleration and deceleration rates i.e. 0.06 mph.
The initial low speed of bullet AV is causing an increase in distance of 18.37 feet to 20.21 feet,

BRAIN: Broad Research in Artificial Intelligence and Neuroscience
Volume 8, Issue 3, September 2017, ISSN 2067-3957 (online), ISSN 2068-0473 (print)

from 20.21 feet to 25.35 feet and hence lowering the fear from 16 to 6 (shown by record number 1
to 5 in table 22). A gradual decrease in the distance and then a corresponding increase in the fear
intensity have been portrayed through the rest of the records of table 22. Figure 17 shows the actual
behavior of all 100 records that were gathered during 100 runs of the simulation. The downward
movement of the orange line (fear) with the upward movement of the gray line (distance) is
validating the invariant1_TypeB.

7.2.3. Results and Discussion: Experiment-Invariant 2
Figure 18 shows the validation results of invariant 2. A total of 12 experiments are

conducted to validate the claim that If the pre-condition that “Bullet Agent autonomous vehicle is
successfully reacting to the rear end collision threat using EEEC_Agent short route reaction time”
is true, then stopping sight distance required would result in a post-condition of “EEEC_Agent
requires smaller SSD as compared to the SSD required by Human driver”. The vertical axis of
figure 16 presents different SSD values, whereas the corresponding velocities are given on the
horizontal axis. From the figure 18, it can be seen that when the bullet AV successfully avoids the
collision by traveling at the velocity of 15 mph then it takes SSD= 31.733 feet as a safe distance.
Whereas in the same case the human driver needs SSD = 106.015 feet. In the same way, when the
bullet AV successfully avoids the collision by traveling at a high velocity of 50 mph then it takes
SSD= 277.184 feet as a safety distance. Whereas for the same case the human driver needs SSD =
524.790 feet. Hence, the results prove that the claim of invariant 2 is true and the emotions inspired
collision avoidance module is working properly in case of rear-end collisions.

The EEC_Agent proposed in (F. Riaz, S. I. Shah, M. Raees, I. Shafi, A. Iqbal, 2013) is
tested for lateral collisions at different speeds. However, the (F. Riaz, S. I. Shah, M. Raees, I. Shafi,
A. Iqbal, 2013) failed to validate the results of the simulation against any standard collision
avoidance parameters. We have validated our claim of rear end collisions against the standard
Stopping Sight Distance by adapting VOMAS agent approach defined in CABC framework.

Figure 18. EEEC_Agent SSD vs. Human Driver SSD Graph

7.2.4. Results and Discussion: Experiment-Invariant 3
Figure 19 shows the validation results of invariant 3. Total 7 experiments are conducted to

validate the claim that if the pre-condition that “Bullet vehicle is successfully reacting to the
overtaking collision threat using EEEC_Agent short route reaction time” is true, then overtaking
sight distance required would result in a post-condition of “EEEC_Agent requires smaller OSD as
compared to the OSD of the human driver”. From the figure 19 it can be seen that when the bullet
AV successfully avoids the collision by traveling at a velocity of 25 mph then it takes OSD= 63.408

F. Riaz, M. A. Niazi - Validation of Enhanced Emotion Enabled Cognitive Agent Using Virtual Overlay Multi-
Agent System Approach

feet as a safe distance. However, for the same case, the human driver needs OSD = 85 feet. In the
same way, when the bullet AV successfully avoids the collision by traveling at a higher velocity of
50 mph then it takes OSD= 145.264 feet as a safe distance. Whereas for the same case the human
driver needs OSD = 185 feet. Hence, the results prove that the claim of invariant 3 is true and the
emotions inspired collision avoidance module is working properly in case of overtaking.

Figure 19. EEEC_Agent OSD vs. Human Driver OSD Graph

The EEC_Agent proposed in (F. Riaz, S. I. Shah, M. Raees, I. Shafi, A. Iqbal, 2013) is tested for

lateral collisions at different speeds. However, the (F. Riaz, S. I. Shah, M. Raees, I. Shafi, A. Iqbal,
2013) failed to validate the results of the simulation against any standard collision avoidance
parameters. We have validated our claim of overtaking collisions against the standard Overtaking Sight
Distance parameters by adapting VOMAS agent approach defined in CABC framework.

8. Practical validation of the Validation of Fear Generation Mechanism of
EEEC_Agent
To further validate the results of VOMAS agent based simulation regarding fear generation,

EEEC_Agent agent has been deployed in a prototype autonomous vehicle. OCC model based
application has been developed in the Visual C# environment and it has been deployed in AV using
Windows 10 based tablet. Different AV-Obstacle tests have been performed to validate the results
of VOMAS agent based simulation regarding the fear generation mechanism of EEEC_Agent. The
overall architecture of rigorous validation using VOMAS agent has been presented in figure 20. The
prototype AV is equipped with high range ultrasonic sonars, which help to measure the distance of
AV from incoming obstacle. To control these sonars, Arduino ATmega2560 processor has been
utilized, which further pass the results of these sonars to the C# application, which computes the
intensity of fear. To validate that the fear generation processes of EEEC_Agent is practically
generating the same results as it generated in NetLogo simulation, VOMAS agent has been
considered, which further utilized the invariants defined in section 5.1.

BRAIN: Broad Research in Artificial Intelligence and Neuroscience
Volume 8, Issue 3, September 2017, ISSN 2067-3957 (online), ISSN 2068-0473 (print)

Figure 20. Proposed Practical Validation Architecture Using VOMAS Agent Approach

Figure 21 a and b present the results of different practical validation tests, which have been
performed using AV-Obstacle topology. From the figure 21a, it can be seen that when the AV moves at
low speed towards an obstacle, then in the beginning, it feels low fear due to high distance i.e. 210 feet.
Then gradually fear increases as the distance decreases. This confirms the simulation results that the fear
generation mechanism of EEEC_Agent is working properly.

(a)

EEEC_Agent Installed In AV

High Range Ultra
sonic Sonars

Arudino
Microcontroller

OCC Model
Based Fear
Generation

Coded in C#

Validation
Invariants

Testing Results

VOMAS Agent

100

150

200

250

Low Low Medium Medium High

100 100 100 100 100

Distance Vs. Fear Intensity at 100 RPM

Distance

F. Riaz, M. A. Niazi - Validation of Enhanced Emotion Enabled Cognitive Agent Using Virtual Overlay Multi-
Agent System Approach

(b)

Figure 21. Graphic representations of practical validation results a. Different
Fear Intensity levels perceived by EEEC_Agent installed AV at 100 RPM speed,
b. Different Fear Intensity levels perceived by EEEC_Agent installed AV at 200

RPM speed

Further, if we analyzed the results of figure 21b then it can be seen that at high speed

EEEC_Agent first feel low fear and then suddenly jump to the medium fear because the distance
suddenly decreased from 190 feet to 155 feet. In the same way, other results confirm that the fear
increases as the distance between EEEC_Agent and obstacles decreases. An interesting phenomenon
can be noted from the figure 21a and 21 b that at low speed, the EEEC_Agent feel medium fear after
some time, whereas with high speed the EEEC_Agent suddenly perceive medium fear after feeling low
fear. It shows the validation of the proposed fear generation mechanism that fear feeling capability is not
fixed but dynamic. Hence, it can be seen that the validation of validation has proved that the fear
generation mechanism of proposed EEEC_Agent works accurately according to the different situations.
These results can be further utilized by the researchers to build the EEEC_Agent based systems with
more confidence.

9. Conclusion
A validated Enhanced Emotion Enabled Cognitive Agent (EEEC_Agent) has been proposed

to avoid the collision between autonomous vehicles. For this purpose, prospect based emotions
defined by OCC model are used to generate fear in EEEC_Agent. SimConnector approach is used
to join fuzzy logic environment results with NetLogo agent-based simulation. VOMAS Agent is
used to validate the performance of EEEC_Agent in the rear end and overtaking collisions. The
extensive experiments proved that validated EEEC_Agent can perform collision avoidance with
smaller SSD and OSD as compared to the human driver.

The authors declare that there is no conflict of interest in this manuscript.

References

Aashto, A. (2001). Policy on geometric design of highways and streets, American Association of
State Highway and Transportation Officials, Washington, DC, vol. 1, no. 990, pp. 158.

Altamimi, A. B. & Ramadan, R. A. (2016). Towards internet of things modeling: a gateway
approach, Complex Adaptive Systems Modeling, vol. 4, no. 1, pp. 25.

0
20
40
60
80

100
120
140
160
180
200

Low Medium Medium High High

200 200 200 200 200

Distance Vs. Fear Intensity at 200 RPM

Distance

BRAIN: Broad Research in Artificial Intelligence and Neuroscience
Volume 8, Issue 3, September 2017, ISSN 2067-3957 (online), ISSN 2068-0473 (print)

Barreteau, O., Antona, M., D'Aquino, P., Aubert, S., Boissau, S., Bousquet, F., Daré, W., Etienne,
M., Le Page, C., & Mathevet, R. (2003). Our companion modelling approach, Journal of
Artificial Societies and Social Simulation, vol. 6, no. 1.

Batool, K. & Niazi, M. A. (2014). Towards a methodology for validation of centrality measures in
complex networks, PloS one, vol. 9, no. 4, pp. e90283.

Batool, K. & Niazi, M. A. (2017). Modeling the internet of things: a hybrid modeling approach
using complex networks and agent-based models, Complex Adaptive Systems Modeling, vol.
5, no. 1, pp. 4.

Fagiolo, G., Birchenhall, C., & Windrum, P. (2007). Empirical validation in agent-based models:
Introduction to the special issue, Computational Economics, vol. 30, no. 3, pp. 189-194.

Harlow, C. & Peng, S. (2001). Automatic vehicle classification system with range sensors,
Transportation Research Part C: Emerging Technologies, vol. 9, no. 4, pp. 231-247.

Jiménez, F., Naranjo, J. E., & Gómez, Ó. (2015). Autonomous collision avoidance system based on
accurate knowledge of the vehicle surroundings, IET intelligent transport systems, vol. 9,
no. 1, pp. 105-117.

Joo, Y. K. & Lee-Won, R. J. (2016). An agent-based intervention to assist drivers under stereotype
threat: effects of in-vehicle agents' attributional error feedback, Cyberpsychology, Behavior,
and Social Networking, vol. 19, no. 10, pp. 615-620.

Laghari, S. & Niazi, A. (2016). Modeling the internet of things, self-organizing and other complex
adaptive communication networks: a cognitive agent-based computing approach, PLoS One,
vol. 11.

Lu, G., Li, L., Wang, Y., Zhang, R., Bao, Z., & Chen, H. (2014). A rule based control algorithm of
connected vehicles in uncontrolled intersection, pp. 115-120.

Makowsky, M. (2006). An agent-based model of mortality shocks, intergenerational effects, and
urban crime, Journal of Artificial Societies and Social Simulation, vol. 9, no. 2.

Nehate, G. & Rys, M. (2006). 3D calculation of stopping-sight distance from GPS data, Journal of
transportation engineering, vol. 132, no. 9, pp. 691-698.

Niazi, M. A., & Hussain, A. (2012). Cognitive Agent-based Computing-I: A Unified Framework for
Modeling Complex Adaptive Systems Using Agent-based & Complex Network-based
Methods. Springer Science & Business Media.

Niazi, M. A., Hussain, A., & Kolberg, M. (2009). Verification and validation of agent based
simulations using the VOMAS (virtual overlay multi-agent system) approach.

Niazi, M. A., Siddique, Q., Hussain, A., & Kolberg, M. (2010). Verification & validation of an
agent-based forest fire simulation model, p. 1.

Niazi, M., Siddique, Q., Hussain, A., & Fortino, G. (2011). SimConnector: An approach to testing
disaster-alerting systems using agent based simulation models, pp. 659-665.

Ortony, A., Clore, G. L., & Collins, A. (1990). The cognitive structure of emotions: Cambridge
university press.

Reichardt, D. M. (2008). Approaching driver models which integrate models of emotion and risk.
pp. 234-239.

Riaz, F. & Niazi, M. A. (2016). Road collisions avoidance using vehicular cyber-physical systems:
a taxonomy and review, Complex Adaptive Systems Modeling, vol. 4, no. 1, pp. 15.

Riaz, F., Shah, S. I., Raees, M., Shafi, I., & Iqbal, A. (2013). Lateral Pre-crash sensing and
avoidance in emotion enabled cognitive agent based vehicle-2-vehicle communication
system, International Journal of Communication Networks and Information Security, vol. 5,
no. 2, pp. 127.

Rodríguez-Seda, E. J., Stipanović, D. M., & Spong, M. W. (2016). Guaranteed Collision Avoidance
for Autonomous Systems with Acceleration Constraints and Sensing Uncertainties, Journal
of Optimization Theory and Applications, vol. 168, no. 3, pp. 1014-1038.

F. Riaz, M. A. Niazi - Validation of Enhanced Emotion Enabled Cognitive Agent Using Virtual Overlay Multi-
Agent System Approach

Samiee, S., Azadi, S., Kazemi, R., Eichberger, A., Rogic, B., & Semmer, M. (2016). Performance
Evaluation of a Novel Vehicle Collision Avoidance Lane Change Algorithm, Advanced
Microsystems for Automotive Applications 2015, pp. 103-116: Springer.

Sarhan, M. & Hassan, Y. (2008). Three-dimensional, probabilistic highway design: sight distance
application, Transportation Research Record: Journal of the Transportation Research
Board, no. 2060, pp. 10-18.

Schmid, A. (2005). What is the Truth of Simulation?, Journal of Artificial Societies and Social
Simulation, vol. 8, no. 4.

Smith, E. E. & Stepehen, M. K. (2009). Emotion and Cognition. In Asoke K. Gosh (Ed.), Cognitive
Psychology, 1st edition, New Delhi: PHI Learning.

Tiengo, W., Costa, E. B., & Fechine, J. M. (2016). Reducing Risk of Rollover in Curve for Heavy-
Duty Vehicles with an Agent-Based Advanced Driver Assistance System, pp. 65-72.

Vine, S. L., Liu, X., Zheng, F., & Polak, J. (2016). Automated cars: Queue discharge at signalized
intersections with ‘Assured-Clear-Distance-Ahead’driving strategies, Transportation
Research Part C: Emerging Technologies, vol. 62, pp. 35-54.

Waizman, G., Shoval, S., & Benenson, I. (2015). Micro-Simulation Model for Assessing the Risk
of Vehicle–Pedestrian Road Accidents, Journal of Intelligent Transportation Systems, vol.
19, no. 1, pp. 63-77.

Zedadra, O., Jouandeau, N., Seridi, H., & Fortino, G. (2017). Multi-Agent Foraging: state-of-the-art
and research challenges, Complex Adaptive Systems Modeling, vol. 5, no. 1, pp. 3.

Faisal Riaz is a Doctoral student at Iqra University, Pakistan and currently serving
as an Assistant Prof. in the dept. of computer sciences, Mirpur University of
Science and Technology. His research interests are Vehicular Cyber Physical
Systems, Cognitive Radio, Affective Computing, and Agent Based Modelling. He
has more than 24 research publications in various national/international research
journals and conferences. He is the author of two books as well. Furthermore, he is
head of Intelligent Transport Systems Lab at Mirpur University of Science and

Technology (MUST), AJ&K Pakistan. Under the banner of ITS lab, he has developed the first
Autonomous Vehicle of Pakistan. He is the reviewer of many well reputed journals. He has served as a
technical program committee member in many international IEEE conferences. His reviewing profile can
be viewed over https://publons.com/author/742771.