Physiology-based Affect Recognition During Driving in Virtual

Environment for Autism Intervention

Dayi Bian

, Joshua Wade

, Amy Swanson

, Zachary Warren

4,5,6

and Nilanjan Sarkar

1,2

Department of Electrical Engineering, Vanderbilt University, Nashville, TN, U.S.A.

Mechanical Engineering and

Computer Science, Vanderbilt University, Nashville, TN, U.S.A.

Department of Pediatrics and Psychiatry, Vanderbilt University, Nashville, TN, U.S.A.

Department of Special Education, Vanderbilt University, Nashville, TN, U.S.A.

Vanderbilt Kennedy Center, Treatment and Research Institute of Autism Spectrum Disorders, Vanderbilt University,

Nashville, TN, U.S.A.

Keywords: Virtual Reality, Autism intervention, Affect recognition, Physiological sensing.

Abstract: Independent driving is believed to be an important factor of quality of life for individual with autism

spectrum disorder (ASD). In recent years, several computer technologies, particularly Virtual Reality (VR),

have been explored to improve driving skills in this population. In this work a VR-based driving

environment was developed for skill training for teenagers with ASD. Eight channels of physiological

signals were recorded in real time for affect recognition during driving. A large set of physiological features

were investigated to determine their correlation with four categories of affective states: engagement,

enjoyment, frustration and boredom, of teenagers with ASD. In order to have reliable reference points to

link the physiological data with the affective states, the subjective reports from a therapist were recorded

and analyzed. Six well-known classifiers were used to develop physiology-based affect recognition models,

which yielded reliable predictions. These models could potentially be used in future physiology-based

adaptive driving skill training system such that the system could adapt based on individual affective states.

1 INTRODUCTION

Autism spectrum disorder (ASD) has a prevalence

rate as high as 1 in 68 children in U.S. (CDC 2014).

While at present there is no single accepted

intervention, treatment, or known cure for ASD,

there is growing consensus that intensive behavior

and educational intervention programs can

significantly improve long-term outcomes for

individuals and their families (Rogers 1998; Cohen,

Amerine-Dickens et al. 2006). However, many

current intervention approaches show limited

improvements in functional adaptive skills because

traditional skill-based methodologies often failed to

systematically match intervention strategies to

specific underlying processing deficits associated

with targeted skills. Additionally, such intervention

approaches may have difficulties creating

opportunities for addressing such skills and deficits

within and across naturalistic settings in

appropriately intensive dosages (Goodwin 2008). In

this regard, technological intervention paradigms,

including Virtual Reality (VR) platforms, have been

suggested as potentially powerful tools for

addressing these limits of current intervention

paradigms. Moreover, given the limited availability

of professionals trained in autism intervention, it is

likely that emerging technology will play an

important role in providing more accessible and

individualized adaptive intervention in the future

(Standen and Brown 2005; Tartaro and Cassell 2007;

Lahiri, Bekele et al. 2013).

VR-based intervention could be utilized to help

children with ASD generalize learned skill to the

real world not only by providing more control over

how the basic skills are taught, but also the ability to

systematically employ and reinforce these skills

within many different, controllable, realistic

interaction environments. In addition, the virtual

world can be designed to break down, repeat, add

and subtract aspects of the environment in any

manner necessary to achieve a task goal. While VR-

based ASD intervention has become an active

research field in recent years, more in-depth studies

137

Bian D., Wade J., Swanson A., Warren Z. and Sarkar N..

Physiology-based Affect Recognition During Driving in Virtual Environment for Autism Intervention.

DOI: 10.5220/0005331301370145

In Proceedings of the 2nd International Conference on Physiological Computing Systems (PhyCS-2015), pages 137-145

ISBN: 978-989-758-085-7

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

are required to explore how skills learned in virtual

environment are translated into real-world situations.

Historically, VR environments applied to

assistive intervention for children with ASD were

designed to develop skills based on performance

only (e.g., correct or incorrect and some other

performance metrics). However, current research

focuses the development of VR and other

technologies that respond not only to explicit

human-computer interactions (e.g., keyboard, mouse,

joystick, etc.), but also to implicit interactions like

eye gaze and physiological signals (Wilms,

Schilbach et al. 2010; Bekele, Lahiri et al. 2013;

Lahiri, Bekele et al. 2013). Such methods may offer

potential to individualize applications. Ultimately,

VR systems that not only assess performance in

specific task but also measure eye gaze or

physiological markers of engagement may lead to

optimization of learning (Welch, Lahiri et al. 2009;

Lahiri, Bekele et al. 2013).

The main objective of this paper is to explore the

reliability of using physiological signals to detect

affective states in a VR-based driving simulation

environment. The results show that physiological

signals can be used as a reliable way to detect

participants’ affective states in a driving task and

these affective states together with performance

could potentially be used to alter VR interactions.

While there exists a body of literature that

discusses interventions for individuals with ASD to

develop social skills, language development and

emotion recognition (Sundberg and Partington 1998;

Bauminger 2002; Golan, Ashwin et al. 2010), only a

few studies have addressed how to improve driving

skills of ASD population. Cox and his colleagues’

study (Cox, Reeve et al. 2012) reported parents’

experiences about driving of young adults with ASD

and provided suggestions to teach driving skills for

ASD teenagers. Huang et al. (Huang, Kao et al. 2012)

also addressed the factors associated with driving in

teenagers with ASD. Reimer and his colleagues

(Reimer, Fried et al. 2013) explored the differences

between an ASD group and a control group in terms

of physiology. However, only standard statistical

techniques were used in this study instead of

detecting affective states by using physiological

signals. Our previous study (Wade, Bian et al. 2014)

explored the differences between these two groups

in a more comprehensive way. These studies provide

us with useful information to design the driving

system and are the foundation of the proposed work.

As far as we know, there is no work on physiology-

based affect detection in driving skill training system

for the ASD population.

This paper is organized as follows. In Section II, we

provide a brief background on VR-based driving

task - the overall system description and how

physiology is used to measure the affective states of

the participants. This section is followed by a

description of the driving task. In Section IV, we

focus on the physiology-based affect detection

system description and results of physiological data

analysis. The implication of our results and future

work are discussed in the last section.

2 SYSTEM DESCRIPTION

The Virtual Reality (VR) based driving system

contained a VR module and three subsystems, which

were a peripheral physiological data acquisition

module, an EEG data acquisition module and an eye

tracker module (Fig. 1).

The virtual environment was developed via the

Unity game engine (www.unity3d.com). Within

Unity, we developed a graphical user interface,

created behavior for vehicles, pedestrians and traffic

lights, designed the driving scenario and embedded

traffic rules. Participants interacted with the driving

environment by operating a Logitech G27 driving

controller that was mounted on a playseat (Fig. 2).

The VR system was modeled as a video game with

three difficulty levels: easy, medium and hard. Each

level contained three assignments. Each assignment

had eight trials which were designed in order to

improve specific driving skill such as turning, speed-

maintenance, merging and following traffic laws.

Physiological data, EEG data and eye gaze data were

recorded continuously from the beginning of the

experiment to the end. A therapist rated the

participant’s affective states via a custom-designed

online rating program. More details of VR module

could be found in our previous papers (Bian, Wade

et al. 2013; Wade, Bian et al. 2014).

In this work, we only focused on the physiology-

based affect recognition during driving in VR. Four

categories of affective states, engagement,

enjoyment, frustration, boredom, were chosen

because of their importance in driving (Baker,

D'Mello et al. 2010) as well as their detectability

using peripheral physiological signals (Bradley and

Lang 2000; Sarkar 2002; Rani, Sarkar et al. 2003;

Liu, Rani et al. 2006; Welch, Lahiri et al. 2009). As

can be seen from the framework of our study (Fig. 1),

establishing an affect recognition model could lead

to the development of an adaptive closed-loop

driving skill training system.

PhyCS2015-2ndInternationalConferenceonPhysiologicalComputingSystems

138



Figure 1: Framework overview.

3 METHODS AND MATERIALS

3.1 Experimental Setup

The physiological signals were collected using the

Biopac MP150 physiological data acquisition system

(www.biopac.com) with a sampling rate of 1000 Hz.

Several physiological signals were investigated. The

acquired physiological signals were broadly

classified as cardiovascular activities including

electrocardiogram (ECG), photoplethysmogram

(PPG); electrodermal activities (EDA) including

tonic and phasic responses from galvanic skin

response (GSR); electromyogram (EMG) activities

from Corrugator Supercilii, Zygomaticus Major, and

Upper Trapezius muscles; respiration and skin

temperature.

These signals were measured by using light-

weight, non-invasive wireless sensors (Fig. 2). ECG

signal was collected from the chest using two

disposable electrodes to record electrical activity

generated by the heart muscle. PPG and GSR were

measured from toes instead of fingers in order to

reduce the motion artifact from driving. EMG was

measured by placing surface electrodes on

Corrugator Supercilii and Zygomaticus Major and

Upper Trapezius muscles. Respiration was used to

measure changes in thoracic circumference that

occur as a participant breathes. Skin temperature was

collected from the upper arm by using a temperature

sensor. In addition, an EEG cap and an eye tracker

were also used to detect EEG signal and eye gaze in

this setup.



Figure 2: Physiological sensors setup.

A socket-based communication module was

developed to transmit task-related (e.g., trial

start/stop) event triggers from the virtual driving

environment to the Biopac. Physiological signals

along with task-related event triggers were sent over

an Ethernet link to a physiological data logger

computer to enable acquiring and logging of the

signals in a time-synchronized manner with the VR-

based driving task (Fig. 3).

3.2 Procedure

Each participant completed six sessions in different

days. The first and last session were pre and post

sessions, which contained the exact same

assignments. Participants usually completed a single

session within approximately 60 minutes. At the

start of each session, physiological sensors and EEG

cap were placed on a participant’s body and the eye

tracker was calibrated. Participants watched a short

Physiology-basedAffectRecognitionDuringDrivinginVirtualEnvironmentforAutismIntervention

139



Figure 3: Experimental setup diagram.

instruction video which explained basic instructions

and game controls. After the tutorial, participants

were asked to remain calm and relax for three

minutes during which physiological, EEG, and eye

gaze baseline data were collected. The baseline data

were later used to offset environmental variability.

Participants also had three minutes free practice in

which there were no pedestrians and no other

vehicles in the VR environment. This practice mode

allowed participants to familiarize themselves with

the game controls and virtual environment.

After the three-minute practice, participants

began the first of three assignments. Through the

assignment, participants followed the navigation

system and tried to obey traffic rules. Disobeying

any traffic rules (i.e., running red light) caused a

performance failure. In addition, in gaze contingent

group, failing to look at a specific region of interest

in specific trials (i.e., did not look at speedometer in

school zone) also caused a gaze failure. Four failures

would cause the assignment end and the game would

go back to assignment selection menu. Time

duration for each assignment varied from 2 minutes

to 5 minutes depending on the participants’

performance.

Because of suspected unreliability of self-report of

teenagers with ASD, an experienced therapist was

involved in the experiment. The therapist was seated

next to the experiment room, watching the

experiment from the view of two cameras (Fig. 3).

The therapist rated the participants’ affective states

in four categories: engagement, enjoyment,

frustration and boredom by using a continuous rating

scale from 0 to 9 via an online rating program. For

each assignment, an overall rating was given after

the assignment. Also, the therapist provided ratings

when she felt the participants had obvious affective

state changes.

3.3 Participants

We have recruited 12 male teenagers with ASD for

this phase of the study. While it was not our

intention to recruit all male participants, they were

recruited randomly through the existing university

clinical research registry and happened to be all

males. This may partially be due to the fact that

ASD prevalence in male population is four times as

high as it is for female population (CDC 2014). All

participants had a clinical diagnosis of ASD from a

licensed clinical psychologist as well as cores at or

above clinical cutoff on the Autism Diagnostic

Observation Schedule (Lord, Risi et al. 2000). The

Institutional Review Board (IRB) approval was

sought and received for conducting the experiment.

Ten participants’ physiological data were used for

this paper because two of them were not able to

follow the instructions to get valid physiological

data.

Table 1: Participant data.

Participant NO. Age IQ

ADOS total

raw core

ADOS CSS

ASD01 13.6 -- -- --

ASD02 15.1 80 16 9

ASD03 14.3 86 14 8

ASD04 14.6 99 -- --

ASD05 17.1 118 8 5

ASD06 13.2 108 14 8

ASD08 17.5 125 13 8

ASD09 15.5 117 11 7

ASD10 16.6 88 22 10

ASD12 14.1 -- 11 7

Note: ADOS_CSS = Autism Diagnostic Observation

Schedule Calibrated Severity Score; IQ = composite score:

Differential Ability Scales (General Conceptual Ability) or

Wechsler Intelligence Scale for Children (Full Scale IQ).

4 PHYSIOLOGICAL DATA

ANALYSIS

In this study, a group model was developed to

classify affective states in four categories:

engagement, enjoyment, frustration and boredom. A

90-s window was chosen for sampling the

continuously-recorded physiological data. The 90-s

window was chosen for several reasons: it

approximates the time needed for autonomic signal

PhyCS2015-2ndInternationalConferenceonPhysiologicalComputingSystems

140

such as skin conductance to recover and it also

provides a level of smoothing when the features

were extracted. The samples were labeled by the

therapist’s overall rating for each assignment. The

therapist’s ratings were mapped into high and low

intensity for each category for binary classification.

The recorded physiological signals were

preprocessed for feature extraction. First, each signal

was filtered using different filters such as high/low

pass filter, notch filter etc. to reject outliers and

artifacts. The signals were then standardized to be

zero mean and unity standard deviation. In addition,

baseline wander was removed from the PPG signal

before peak detection as this signal is known to be

affected by baseline wander.

Several features were extracted for each channel

of physiological signal. A brief explanation for all

the features are listed in Table 2.

The Waikato Environment for Knowledge

Analysis (WEKA) (Hall, Frank et al. 2009), which is

recognized as a landmark system in machine

learning nowadays, was used to do feature selection

and classification in this study. For each category,

CorrelationAttributeEval (Hall 1999) algorithm was

used to select features. This algorithm evaluated the

value of a feature by measuring the correlation

between it and the class. It ranked feature subsets

according to a correlation based heuristic evaluation

function. The bias of the evaluation function was

toward subsets that contain features that were highly

correlated with the class and uncorrelated with each

other. Irrelevant features were ignored because they

would have low correlation with the class.

Redundant features were screened out as they would

be highly correlated with one or more of the

remaining features. Top ten features (Table 3) that

had the highest correlations with the classes were

chosen for further classification.

Six different well-known classifiers were used

for classification for each category. These classifiers

were:

BayesNet: SimpleEstimatior estimator and K2

search algorithm were chosen.

NaiveBayes: Numeric estimator precision values

were chosen based on analysis of the training data.

SVM: Radial basis function was chosen with a

degree of 3.

MultiLayerPerceptron: HiddenLayers were

chosen by using (attribs + classes) / 2, learningRate

was 0.3.

RandomForest: The number of trees to be

generated was 10, maxDepth was unlimited.

J48 DecisionTree: The minimum number of

instances per leaf was 2, 1 of 3 folds data was used

for reduced-error pruning.

10-fold cross validation was used. The

classification accuracies for each category from

different classifiers are shown in Figure 4.

The highest accuracy for engagement, enjoyment,

frustration and boredom were 77.78%, 79.63%,

79.63% and 81.48%, respectively. These results are

comparable to the accuracy of most up-to-date

affective computing systems (Tao and Tan 2005;

Jerritta, Murugappan et al. 2011).

As we can see from the selected 10 features of

each category, PPG, RSP, SCR, EMG_C and

EMG_Z are most common for the chosen affective

states. This indicates the possibility of using a

smaller set of features with a relatively low

computational cost for a potential closed-loop

system.

In this study, we focused on developing a group

affective state prediction model instead of model for

each individual. In the future, we want to use this

group model to provide affective state feedback in a

closed-loop system and potentially develop a more

efficient driving system to teach teenagers with ASD

basic driving skills.

5 DISCUSSION

There is a growing consensus that development of

computer assistive therapeutic tools can make

application of intensive intervention for teenagers

with ASD more readily accessible. In recent years,

several applications of advanced intervention that

address deficit in driving for teenagers with ASD

were investigated. However, these application

lacked the ability of detect the affective cues of the

teenagers, which could be crucial given the affective

factors of teenagers with ASD have significant

impacts on the intervention practice.

In this work, we presented a physiology-based

affect recognition framework for teenagers with

ASD. 68 features were extracted from the recorded

physiological data. Subsequently 10 features were

selected by using CorrelationAttributeEval algorithm

to overcome the overfitting problem. Six most

commonly used machine leaning algorithms were

used to classify four category of affective states. The

developed model could reliably recognize affective

states of the teenagers with ASD and provide the

basis for physiology-based affect-sensitive driving

skill training system.

In the future, a real-time affect recognition

system which dynamically shape the driving task

will be developed. We will also incorporate EEG

Physiology-basedAffectRecognitionDuringDrivinginVirtualEnvironmentforAutismIntervention

141

Table 2: Physiological features.

Physiological signal Feature extracted Label used Unit of measurement

Electrocardiogram

(ECG/EKG)

Sympathetic power

Parasympathetic power

Very low-frequency power

Ratio of powers

Mean Interbeat Interval (IBI)

Std. of IBI

power_sym

power_para

power_vlf

para_vlf

para_sym

vlf_sym

mean_ibi_ekg

std_ibi_ekg

Unit/s

No unit

Standard deviation(no

unit)

Photoplethysmogram

(PPG)

Mean and std. of amplitude of the peak

values

Mean and std. of heart rate variability

ppg_peak_mean

ppg_peak_std

hrv_mean

hrv_std

µV

No unit

Electrodermal activity

(EDA)

Mean and std. of tonic activity level

Slope of tonic activity

Mean and std. of amplitude of skin

conductance response (phasic activity)

Rate of phasic activity

Mean and std. of rise time

Mean and std. of recovery time

SCL_mean

SCL_sd

SCL_slope

SCR_mean

SCR_sd

SCR_rate

tRise_mean

tRise_std

tHRecovery_mean

tHRevovery_sd

µS

µS/s

µS

Response peaks/s

Electromyographic

Activity

(EMG)

Mean of Corrugator, Zygomaticus and

Trapezius activities

Std. of Corrugator, Zygomaticus and

Trapezius activities

Slope of Corrugator, Zygomaticus and

Trapezius activities

Number of burst activities per minute of

Corrugator, Zygomaticus and Trapezius

Mean of Corrugator, Zygomaticus and

Trapezius burst activities

Std. of Corrugator, Zygomaticus and

Trapezius burst activities

Mean and Median frequency of

Corrugator, Zygomaticus and

Trapezius

Mean of the amplitude of Corrugator,

Zygomaticus and Trapezius burst

activities

Cemg_mean

Zemg_mean

Temg_mean

Cemg_std

Zemg_std

Temg_std

Cemg_slope

Zemg_slope

Temg_slope

Cburst_count

Zburst_count

Tburst_count

Cburst_mean

Zburst_mean

Tburst_mean

Cburst_std

Zburst_std

Tburst_std

Cfreq_mean

Cfreq_med

Zfreq_mean

Zfreq_med

Tfreq_mean

Tfreq_med

Cburst_amp_mean

Zburst_amp_mean

Tburst_amp_mean

µV

No unit

µV/s

/min

No unit

Hertz

µV



PhyCS2015-2ndInternationalConferenceonPhysiologicalComputingSystems

142

Table 2: Physiological features (cont.).

Respiration

(RSP)

Mean amplitude

Std. of amplitude

Subband spectral entropy

Minimum and maximum difference

Change rate

Power spectrum density

Std. of Poincare plot geometry

Mean and std. of peak valley magnitude

Mean and std. of breath per minute

RSP_mean

RSP_std

RSP_subbandSpectr

alEntropy({1,2,3})

RSP_minmax_diff

RSP_rate

RSP_low_power

RSP_high_power

RSP_firstOrder_std

RSP_poincare_SD1

RSP_poincare_SD2

PVM_mean

PVM_std

RRI_mean

RRI_std

No unit

Peripheral temperature

(SKT)

Mean temperature

Slope of temperature

Std. of temperature

temp_mean

temp_slope

temp_std

F/s

No unit

Table 3: Selected features for each category of affective states.

Category Features selected

Engagement

RSP_subbandSpectralEntropy(1), hrv_mean, SCR_rate, Zemg_mean, RSP_mean, PVM_std, SCL_sd,

SCL_slope, Cburst_amp_mean, ppg_peak_mean

Enjoyment

hrv_mean, RSP_mean, ppg_peak_mean, Cburst_count, Cemg_slope, Zburst_count, Temg_slope,

PVM_std,Zburst_mean,Cburst_amp_mean

Frustration

Cemg_std, RSP_subbandSpectralEntropy(2), RSP_subbandSpectralEntropy(3), PVM_mean,

RSP_firstOrder_std, temp_slope, RSP_std, RRI_std, Zfreq_med, RSP_low_power

Boredom

tRise_sd, hrv_mean, temp_mean, tRise_mean, SCR_sd, SCR_rate, RSP_subbandSpectralEntropy(3),

RSP_subbandSpectralEntropy(2), Zfreq_mean, Cburst_count

Figure 4: Classification accuracies for each category of

affective states.



Figure 4: Classification accuracies for each category of

affective states (cont.).

Classification accuracy of engagement

100

Classification accuracy of enjoyment

100

Classification accuracy of frustration

100

Classification accuracy of boredom

Physiology-basedAffectRecognitionDuringDrivinginVirtualEnvironmentforAutismIntervention

143

signal and eye gaze in order to give more

individualized feedback.

ACKNOWLEDGEMENTS

The authors would like to thank the participants and

their families for making this study possible. We

also gratefully acknowledge National Science

Foundation Grant 0967170 and National Institute of

Health Grant 1R01MH091102-01A1 that partially

supported the research presented here.

REFERENCE

Baker, R. S., S. K. D'Mello, et al. (2010). "Better to be

frustrated than bored: The incidence, persistence, and

impact of learners’ cognitive–affective states during

interactions with three different computer-based

learning environments." International Journal of

Human-Computer Studies 68(4): 223-241.

Bauminger, N. (2002). "The facilitation of social-

emotional understanding and social interaction in

high-functioning children with autism: Intervention

outcomes." Journal of autism and developmental

disorders 32(4): 283-298.

Bekele, E. T., U. Lahiri, et al. (2013). "A step towards

developing adaptive robot-mediated intervention

architecture (ARIA) for children with autism." Neural

Systems and Rehabilitation Engineering, IEEE

Transactions on 21(2): 289-299.

Bian, D., J. W. Wade, et al. (2013). A novel virtual reality

driving environment for autism intervention.

Universal Access in Human-Computer Interaction.

User and Context Diversity, Springer: 474-483.

Bradley, M. M. and P. J. Lang (2000). "Measuring

emotion: Behavior, feeling, and physiology."

Cognitive neuroscience of emotion 25: 49-59.

CDC (2014). "Prevalence of autism spectrum disorder

among children aged 8 years-autism and

developmental disabilities monitoring network, 11

sites, United States, 2010." Morbidity and mortality

weekly report. Surveillance summaries (Washington,

DC: 2002) 63: 1.

Cohen, H., M. Amerine-Dickens, et al. (2006). "Early

intensive behavioral treatment: Replication of the

UCLA model in a community setting." Journal of

Developmental & Behavioral Pediatrics 27(2): S145-

S155.

Cox, N. B., R. E. Reeve, et al. (2012). "Brief Report:

Driving and young adults with ASD: Parents’

experiences." Journal of autism and developmental

disorders 42(10): 2257-2262.

Golan, O., E. Ashwin, et al. (2010). "Enhancing emotion

recognition in children with autism spectrum

conditions: An intervention using animated vehicles

with real emotional faces." Journal of autism and

developmental disorders 40(3): 269-279.

Goodwin, M. S. (2008). "Enhancing and accelerating the

pace of autism research and treatment." Focus on

Autism and Other Developmental Disabilities 23(2):

125-128.

Hall, M., E. Frank, et al. (2009). "The WEKA data mining

software: an update." ACM SIGKDD explorations

newsletter 11(1): 10-18.

Hall, M. A. (1999). Correlation-based feature selection

for machine learning, The University of Waikato.

Huang, P., T. Kao, et al. (2012). "Factors associated with

driving in teens with autism spectrum disorders."

Journal of Developmental & Behavioral Pediatrics

33(1): 70-74.

Jerritta, S., M. Murugappan, et al. (2011). Physiological

signals based human emotion recognition: a review.

Signal Processing and its Applications (CSPA), 2011

IEEE 7th International Colloquium on, IEEE.

Lahiri, U., E. Bekele, et al. (2013). "Design of a virtual

reality based adaptive response technology for

children with autism." Neural Systems and

Rehabilitation Engineering, IEEE Transactions on

21(1): 55-64.

Liu, C., P. Rani, et al. (2006). Affective state recognition

and adaptation in human-robot interaction: A design

approach. Intelligent Robots and Systems, 2006

IEEE/RSJ International Conference on, IEEE.

Lord, C., S. Risi, et al. (2000). "The Autism Diagnostic

Observation Schedule—Generic: A standard measure

of social and communication deficits associated with

the spectrum of autism." Journal of autism and

developmental disorders 30(3): 205-223.

Rani, P., N. Sarkar, et al. (2003). Affective

communication for implicit human-machine

interaction. Systems, Man and Cybernetics, 2003.

IEEE International Conference on, IEEE.

Reimer, B., R. Fried, et al. (2013). "Brief report:

examining driving behavior in young adults with high

functioning autism spectrum disorders: a pilot study

using a driving simulation paradigm." Journal of

autism and developmental disorders 43(9): 2211-2217.

Rogers, S. J. (1998). "Empirically supported

comprehensive treatments for young children with

autism." Journal of clinical child psychology 27(2):

168-179.

Sarkar, N. (2002). Psychophysiological control

architecture for human-robot coordination-concepts

and initial experiments. Robotics and Automation,

2002. Proceedings. ICRA'02. IEEE International

Conference on, IEEE.

Standen, P. J. and D. J. Brown (2005). "Virtual reality in

the rehabilitation of people with intellectual

disabilities: review." Cyberpsychology & behavior

8(3): 272-282.

Sundberg, M. L. and J. W. Partington (1998). "Teaching

language to children with autism and other

developmental disabilities." Pleasant Hill, CA:

Behavior Analysts.

PhyCS2015-2ndInternationalConferenceonPhysiologicalComputingSystems

144

Tao, J. and T. Tan (2005). Affective computing: A review.

Affective computing and intelligent interaction,

Springer: 981-995.

Tartaro, A. and J. Cassell (2007). "Using virtual peer

technology as an intervention for children with

autism." Towards universal usability: designing

computer interfaces for diverse user populations.

Chichester: John Wiley 231: 62.

Wade, J., D. Bian, et al. (2014). Design of a Virtual

Reality Driving Environment to Assess Performance

of Teenagers with ASD. Universal Access in Human-

Computer Interaction. Universal Access to

Information and Knowledge, Springer: 466-474.

Welch, K. C., U. Lahiri, et al. (2009). An affect-sensitive

social interaction paradigm utilizing virtual reality

environments for autism intervention. Human-

Computer Interaction. Ambient, Ubiquitous and

Intelligent Interaction, Springer: 703-712.

Wilms, M., L. Schilbach, et al. (2010). "It’s in your eyes—

using gaze-contingent stimuli to create truly

interactive paradigms for social cognitive and affective

neuroscience." Social cognitive and affective

neuroscience: nsq024.

Physiology-basedAffectRecognitionDuringDrivinginVirtualEnvironmentforAutismIntervention

145