Cognitive Load Measurement Research Articles

Usability Comparison of Conventional Direct Control Versus Pattern Recognition Control of Transradial Prostheses

The goal of this study was to compare the usability of two control schemes for a transradial myoelectric prosthesis, including conventional direct control (DC) and pattern recognition (PR) control, when used by able-bodied individuals. Three types of response measures were captured to assess the control schemes, including learnability, performance, and cognitive workload. Prior research has applied performance and cognitive workload metrics for evaluation of prosthetics; however, workload measures applied in these studies (e.g., heart rate, electroencephalography, and respiration rate) have many limitations. This study used eye tracking to compare cognitive load implications of the different control schemes for a two degrees-of-freedom myoelectric prosthesis. In total, 12 participants were assigned to either control condition (six persons each) or perform a clothespin relocation task. Results revealed the PR scheme to be more intuitive for users and superior to DC across all response measures. We observed a lower learning percentage (i.e., greater learning potential), lower cognitive load, and greater productivity in task performance. This preliminary study illustrates efficacy of using eye-tracking-based measures of cognitive load and standardize test paradigms for assessment of upper limb prosthetic usability and supports PR prosthetic device control as an intuitive alternative to DC.

Methodological Constraints for Detecting the Modality Effect

Introduction. The aim of our research was to investigate the modality effect in more detail by measuring it in a direct way. Two studies were conducted using the same subject and material.Method. Computer-based learning material was presented on several screens, each containing a short text and a picture. Modality was varied by presenting written versus oral text. Presentation time was system-paced and matched the time for oral text presentation. Based on the dual-task approach cognitive load was measured directly in terms of response time towards a stimulus. In study 1, we used a between-subjects design. Participants got the learning material (all screens) either presented visually (picture + written text, N = 24) or audiovisually in a combination of picture and oral text (N = 24). In study 2, we used the same learning material and cognitive load measurements as in study 1. However, instead of a between-subjects design, we varied text modality within participants (N = 34) in such as that each participant received screens with text presented alternately in written and oral form. Starting modality was counterbalanced.Results. In study 1, we found no modality effect with respect to response time and learning outcome whereas in study 2, we found a significant difference in response times between screens with visually and orally presented text in the direction of the modality effect.Conclusion. The results are discussed with regard to error variance in different experimental designs and effect size of the modality effect as well as with regard to capabilities of participants to strategically adjust to disadvantageous instructional conditions.

Development and Validation of Two Instruments Measuring Intrinsic, Extraneous, and Germane Cognitive Load

Cognitive Load Theory is one of the most powerful research frameworks in educational research. Beside theoretical discussions about the conceptual parts of cognitive load, the main challenge within this framework is that there is still no measurement instrument for the different aspects of cognitive load, namely intrinsic, extraneous, and germane cognitive load. Hence, the goal of this paper is to develop a differentiated measurement of cognitive load. In Study 1 (N = 97), we developed and analyzed two strategies to measure cognitive load in a differentiated way: (1) Informed rating: We trained learners in differentiating the concepts of cognitive load, so that they could rate them in an informed way. They were asked then to rate 24 different learning situations or learning materials related to either high or low intrinsic, extraneous, or germane load. (2) Naïve rating: For this type of rating of cognitive load we developed a questionnaire with two to three items for each type of load. With this questionnaire, the same learning situations had to be rated. In the second study (N = between 65 and 95 for each task), we improved the instrument for the naïve rating. For each study, we analyzed whether the instruments are reliable and valid, for Study 1, we also checked for comparability of the two measurement strategies. In Study 2, we conducted a simultaneous scenario based factor analysis. The informed rating seems to be a promising strategy to assess the different aspects of cognitive load, but it seems not economic and feasible for larger studies and a standardized training would be necessary. The improved version of the naïve rating turned out to be a useful, feasible, and reliable instrument. Ongoing studies analyze the conceptual validity of this measurement with up to now promising results.

Cognitive Resilience and Psychological Responses across a Collegiate Rowing Season.

Student-athletes face numerous challenges across their competitive season. Although mood states have been previously studied, little is known about adaptations in other psychological responses, specifically cognition. The purpose of this study was to characterize cognitive function, mood, sleep, and stress responses at select time points of a season in collegiate rowers. It was hypothesized that during baseline, typical training, and recovery, athletes would show positive mental health profiles, in contrast to decreases in cognition with increases in negative mood and measurements of stress during peak training. Male and female Division I rowers (N = 43) and healthy controls (N = 23) were enrolled and assessed at baseline, typical training, peak training, and recovery. At each time point, measures of cognitive performance (Stroop color-naming task), academic and exercise load, perceived cognitive deficits, mood states, sleep, and stress (via self-report and salivary cortisol) were recorded. Repeated-measures ANOVA revealed significant group-time interactions for perceived exercise load, cognitive deficits, mood states, and perceived stress (P < 0.05). For athletes during peak training, the perception of cognitive deficits was positively correlated with mood disturbance (r = 0.54, P < 0.05) and perceived stress (r = 0.55, P < 0.05) and negatively correlated with response accuracy during incongruent Stroop trials (r = -0.38, P < 0.05). Cognitive performance did not change over the course of the season for either group. Cortisol and sleepiness changed over the course of the season but no significant interactions were observed. These results demonstrate that various psychological responses change over the course of a season, but they also highlight adaptation indicative of cognitive resilience among student-athletes.

Detecting Users’ Cognitive Load by Galvanic Skin Response with Affective Interference

Experiencing high cognitive load during complex and demanding tasks results in performance reduction, stress, and errors. However, these could be prevented by a system capable of constantly monitoring users’ cognitive load fluctuations and adjusting its interactions accordingly. Physiological data and behaviors have been found to be suitable measures of cognitive load and are now available in many consumer devices. An advantage of these measures over subjective and performance-based methods is that they are captured in real time and implicitly while the user interacts with the system, which makes them suitable for real-world applications. On the other hand, emotion interference can change physiological responses and make accurate cognitive load measurement more challenging. In this work, we have studied six galvanic skin response (GSR) features in detection of four cognitive load levels with the interference of emotions. The data was derived from two arithmetic experiments and emotions were induced by displaying pleasant and unpleasant pictures in the background. Two types of classifiers were applied to detect cognitive load levels. Results from both studies indicate that the features explored can detect four and two cognitive load levels with high accuracy even under emotional changes. More specifically, rise duration and accumulative GSR are the common best features in all situations, having the highest accuracy especially in the presence of emotions.

Measuring achievement goal motivation, mindsets and cognitive load: validation of three instruments' scores.

Measurement of motivation and cognitive load has potential value in health professions education. Our objective was to evaluate the validity of scores from Dweck's Implicit Theories of Intelligence Scale (ITIS), Elliot's Achievement Goal Questionnaire-Revised (AGQ-R) and Leppink's cognitive load index (CLI). This was a validity study evaluating internal structure using reliability and factor analysis, and relationships with other variables using the multitrait-multimethod matrix. Two hundred and thirty-two secondary school students participated in a medical simulation-based training activity at an academic medical center. Pre-activity ITIS (implicit theory [mindset] domains: incremental, entity) andAGQ-R (achievement goal domains: mastery-approach, mastery-avoidance, performance-approach, performance-avoidance), post-activity CLI (cognitive load domains: intrinsic, extrinsic, germane) and task persistence (self-directed repetitions on a laparoscopic surgery task) were measured. Internal consistency reliability (Cronbach's alpha) was > 0.70 for all domain scores except AGQ-R performance-avoidance (alpha 0.68) and CLI extrinsic load (alpha 0.64). Confirmatory factor analysis of ITIS and CLI scores demonstrated acceptable model fit. Confirmatory factor analysis of AGQ-R scores demonstrated borderline fit, and exploratory factor analysis suggested a three-domain model for achievement goals (mastery-approach, performance and avoidance). Correlations among scores from conceptually-related domains generally aligned with expectations, as follows: ITIS incremental and entity, r=-0.52; AGQ-R mastery-avoidance and performance-avoidance, r=0.71; mastery-approach and performance-approach, r=0.55; performance-approach and performance-avoidance, r=0.43; mastery-approach and mastery-avoidance, r=0.36; CLI germane and extrinsic, r=-0.35; ITIS incremental and AGQ-R mastery-approach, r=0.34; ITIS incremental and CLI germane, r=0.44; AGQ-R mastery-approach and CLI germane, r=0.48 (all p<0.001). We found no correlation between the number of task repetitions (i.e. persistence) and mastery-approach scores, r=-0.01. ITIS and CLI scores had appropriate internal structures and relationships with other variables. AGQ-R scores fit a three-factor (not four-factor) model that collapsed avoidance into one domain, although relationships of other variables with the original four domain scores generally aligned with expectations. Mastery goals are positively correlated with germane cognitive load.

Cognitive Heat

Current digital systems are largely blind to users’ cognitive states. Systems that adapt to users’ states show great potential for augmenting cognition and for creating novel user experiences. However, most approaches for sensing cognitive states, and cognitive load specifically, involve obtrusive technologies, such as physiological sensors attached to users’ bodies. This paper present an unobtrusive indicator of the users’ cognitive load based on thermal imaging that is applicable in real-world. We use a commercial thermal camera to monitor a person’s forehead and nose temperature changes to estimate their cognitive load. To assess the effect of different levels of cognitive load on facial temperature we conducted a user study with 12 participants. The study showed that different levels of the Stroop test and the complexity of reading texts affect facial temperature patterns, thereby giving a measure of cognitive load. To validate the feasibility for real-time assessments of cognitive load, we conducted a second study with 24 participants, we analyzed the temporal latency of temperature changes. Our system detected temperature changes with an average latency of 0.7 seconds after users were exposed to a stimulus, outperforming latency in related work that used other thermal imaging techniques. We provide empirical evidence showing how to unobtrusively detect changes in cognitive load in real-time. Our exploration of exposing users to different content types gives rise to thermal-based activity tracking, which facilitates new applications in the field of cognition-aware computing.

Effects of attention on the speech reception threshold and pupil response of people with impaired and normal hearing

For people with hearing difficulties, following a conversation in a noisy environment requires substantial cognitive processing, which is often perceived as effortful. Recent studies with normal hearing (NH) listeners showed that the pupil dilation response, a measure of cognitive processing load, is affected by ‘attention related’ processes. How these processes affect the pupil dilation response for hearing impaired (HI) listeners remains unknown. Therefore, the current study investigated the effect of auditory attention on various pupil response parameters for 15 NH adults (median age 51 yrs.) and 15 adults with mild to moderate sensorineural hearing loss (median age 52 yrs.). Both groups listened to two different sentences presented simultaneously, one to each ear and partially masked by stationary noise. Participants had to repeat either both sentences or only one, for which they had to divide or focus attention, respectively. When repeating one sentence, the target sentence location (left or right) was either randomized or blocked across trials, which in the latter case allowed for a better spatial focus of attention. The speech-to-noise ratio was adjusted to yield about 50% sentences correct for each task and condition. NH participants had lower (‘better’) speech reception thresholds (SRT) than HI participants. The pupil measures showed no between-group effects, with the exception of a shorter peak latency for HI participants, which indicated a shorter processing time. Both groups showed higher SRTs and a larger pupil dilation response when two sentences were processed instead of one. Additionally, SRTs were higher and dilation responses were larger for both groups when the target location was randomized instead of fixed. We conclude that although HI participants could cope with less noise than the NH group, their ability to focus attention on a single talker, thereby improving SRTs and lowering cognitive processing load, was preserved. Shorter peak latencies could indicate that HI listeners adapt their listening strategy by not processing some information, which reduces processing time and thereby listening effort.

Neural, physiological, and behavioral correlates of visuomotor cognitive load

Visuomotor ability is quite crucial for everyday functioning, particularly in driving and sports. While there is accumulating evidence regarding neural correlates of visuomotor transformation, less is known about the brain regions that accommodate visuomotor mapping under different cognitive demands. We concurrently measured cortical activity and pupillary response, using functional near infrared spectroscopy (fNIRS) and eye-tracking glasses, to examine the neural systems linked to pupil dilation under varying cognitive demands. Twenty-three healthy adults performed two sessions of a navigation task, in which the cognitive load was manipulated by either reversing the visuomotor mapping or increasing the speed of the moving object. We identified a region in the right superior parietal lobule that responded to both types of visuomotor load and its activity was associated with larger pupillary response and better performance in the task. Our multimodal analyses suggest that activity in this region arises from the need for increased attentional effort and alertness for visuomotor control and is an ideal candidate for objective measurement of visuomotor cognitive load. Our data extend previous findings connecting changes in pupil diameter to neural activity under varying cognitive demand and have important implications for examining brain-behavior associations in real-world tasks such as driving and sports.

Let me be your guide: physical guidance improves spatial learning for older adults with simulated low vision.

Monitoring one's safety during low vision navigation demands limited attentional resources which may impair spatial learning of the environment. In studies of younger adults, we have shown that these mobility monitoring demands can be alleviated, and spatial learning subsequently improved, via the presence of a physical guide during navigation. The present study extends work with younger adults to an older adult sample with simulated low vision. We test the effect of physical guidance on improving spatial learning as well as general age-related changes in navigation ability. Participants walked with and without a physical guide on novel real-world paths in an indoor environment and pointed to remembered target locations. They completed concurrent measures of cognitive load on the trials. Results demonstrate an improvement in learning under low vision conditions with a guide compared to walking without a guide. However, our measure of cognitive load did not vary between guidance conditions. We also conducted a cross-age comparison and found support for age-related declines in spatial learning generally and greater effects of physical guidance with increasing age.

Measuring Cognitive Load in Embodied Learning Settings.

In recent years, research on embodied cognition has inspired a number of studies on multimedia learning and instructional psychology. However, in contrast to traditional research on education and multimedia learning, studies on embodied learning (i.e., focusing on bodily action and perception in the context of education) in some cases pose new problems for the measurement of cognitive load. This review provides an overview over recent studies on embodied learning in which cognitive load was measured using surveys, behavioral data, or physiological measures. The different methods are assessed in terms of their success in finding differences of cognitive load in embodied learning scenarios. At the same time, we highlight the most important challenges for researchers aiming to include these measures into their study designs. The main issues we identified are: (1) Subjective measures must be appropriately phrased to be useful for embodied learning; (2) recent findings indicate potentials as well as problematic aspects of dual-task measures; (3) the use of physiological measures offers great potential, but may require mobile equipment in the context of embodied scenarios; (4) meta-cognitive measures can be useful extensions of cognitive load measurement for embodied learning.

Cognitive load and self-regulation: Attempts to build a bridge

The editors of the Special Issue called for a more integrative approach to the study of cognitive load and self-regulation. The goal formulated for the Special Issue is ambitious. In my role as a constructive critic, I first summarized the findings in the 6 papers, identifying important questions and concerns that emerged while reading the papers. I also identified some general issues that need further clarification and elaboration: I argued that there is a strong need to reach consensus on the conceptualization and measurement of cognitive load and that new methodologies should be developed to capture cognitive load in real time and link it to strategy use.

Cognitive load theory: Practical implications and an important challenge.

The field of medical education has adopted a wide variety of theories from other fields. A fairly recent example is cognitive load theory, which originated in educational psychology. Several empirical studies inspired by cognitive load theory and reviews of practical implications of cognitive load theory have contributed to guidelines for the design of medical education. Simultaneously, several research groups have developed instruments for the measurement of cognitive load in a medical education context. These developments notwithstanding, obtaining evidence for different types of cognitive load remains an important challenge. Therefore, the aim of this article is twofold: to provide medical educators with three key guidelines for the design of instruction and assessment and to discuss several fundamental issues in the remaining challenges presented by different types of cognitive load. The guidelines revolve around minimizing cognitive activity that does not contribute to learning, working with specific learning goals in mind, and appreciating the multifaceted relation between learning and assessment. Key issues around the types of cognitive load include the context in which learning occurs, the continued use of single-item mental effort ratings, and the timing of cognitive load and learning outcome measurements.

‘One Task Fits All’? The Roles of Task Complexity, Modality, and Working Memory Capacity in L2 Performance

The present study explores the independent and interactive effects of task complexity and task modality on linguistic dimensions of second language (L2) performance and investigates how these effects are modulated by individual differences in working memory capacity. Thirty‐two intermediate learners of L2 Spanish completed less and more complex versions of the same type of argumentative task in the speaking and writing modalities. Perceived complexity questionnaires were administered as measures of cognitive load to both L2 learners and native speakers to independently validate task complexity manipulations. Task performance was analyzed in terms of general (complexity and accuracy) as well as task‐relevant (conjunctions) linguistic measures. Quantitative analyses revealed that task modality played a larger role than task complexity in inducing improved linguistic performance during task‐based work: Speaking tasks brought about more syntactically complex output while writing tasks favored more lexically complex and more accurate language. In addition, relationships of working memory capacity with various linguistic measures were attested, but only when the cognitive complexity of tasks was enhanced.

EEG activity as an objective measure of cognitive load during effortful listening: A study on pediatric subjects with bilateral, asymmetric sensorineural hearing loss

ObjectivesDeaf subjects with hearing aids or cochlear implants generally find it challenging to understand speech in noisy environments where a great deal of listening effort and cognitive load are invested. In prelingually deaf children, such difficulties may have detrimental consequences on the learning process and, later in life, on academic performance. Despite the importance of such a topic, currently, there is no validated test for the assessment of cognitive load during audiological tasks. Recently, alpha and theta EEG rhythm variations in the parietal and frontal areas, respectively, have been used as indicators of cognitive load in adult subjects.The aim of the present study was to investigate, by means of EEG, the cognitive load of pediatric subjects affected by asymmetric sensorineural hearing loss as they were engaged in a speech-in-noise identification task. MethodsSeven children (4F and 3M, age range = 8–16 years) affected by asymmetric sensorineural hearing loss (i.e. profound degree on one side, mild-to-severe degree on the other side) and using a hearing aid only in their better ear, were included in the study. All of them underwent EEG recording during a speech-in-noise identification task: the experimental conditions were quiet, binaural noise, noise to the better hearing ear and noise to the poorer hearing ear. The subjects' Speech Recognition Thresholds (SRT) were also measured in each test condition. The primary outcome measures were: frontal EEG Power Spectral Density (PSD) in the theta band and parietal EEG PSD in the alpha band, as assessed before stimulus (word) onset. ResultsNo statistically significant differences were noted among frontal theta power levels in the four test conditions. However, parietal alpha power levels were significantly higher in the “binaural noise” and in the “noise to worse hearing ear” conditions than in the “quiet” and “noise to better hearing ear” conditions (p < 0.001). SRT scores were consistent with task difficulty, but did not correlate with alpha and theta power level variations. ConclusionThis is the first time that EEG has been applied to children with sensorineural hearing loss with the purpose of studying the cognitive load during effortful listening. Significantly higher parietal alpha power levels in two of three noisy conditions, compared to the quiet condition, are consistent with increased cognitive load. Specifically, considering the time window of the analysis (pre-stimulus), parietal alpha power levels may be a measure of cognitive functions such as sustained attention and selective inhibition. In this respect, the significantly lower parietal alpha power levels in the most challenging listening condition (i.e. noise to the better ear) may be attributed to loss of attention and to the subsequent fatigue and “withdrawal” from the task at hand.

Classifying Drivers' Cognitive Load Using EEG Signals.

A growing traffic safety issue is the effect of cognitive loading activities on traffic safety and driving performance. To monitor drivers' mental state, understanding cognitive load is important since while driving, performing cognitively loading secondary tasks, for example talking on the phone, can affect the performance in the primary task, i.e. driving. Electroencephalography (EEG) is one of the reliable measures of cognitive load that can detect the changes in instantaneous load and effect of cognitively loading secondary task. In this driving simulator study, 1-back task is carried out while the driver performs three different simulated driving scenarios. This paper presents an EEG based approach to classify a drivers' level of cognitive load using Case-Based Reasoning (CBR). The results show that for each individual scenario as well as using data combined from the different scenarios, CBR based system achieved approximately over 70% of classification accuracy.

Contributions of mixed reality in a calligraphy learning task: Effects of supplementary visual feedback and expertise on cognitive load, user experience and gestural performance

The learning of handwriting and calligraphy can be improved by supplementary sensory feedback. Emerging technologies such as mixed reality devices can implement appropriate visual, auditory, or proprioceptive feedback. Yet there have been few studies on supplementary visual feedback. Our study was designed to fill this gap, by comparing the effects of two visual feedback types implemented on a graphic tablet with touchscreen (PenWidthFB versus ColoredVelocityFB feedback), in two groups with different levels of expertise in calligraphy (novice versus expert). We collected measures of cognitive load, user experience and gestural performance. Results showed that 1) there was no significant difference in cognitive load between experts and novices, but PenWidthFB feedback created a higher cognitive load than ColoredVelocityFB feedback, 2) for the user experience, there were no obvious differences between experts and novices, and between the two feedback types, 3) concerning the gestural performance, the experts were faster than the novices, and the applied pressure with ColoredVelocityFB feedback was higher than with PenWidthFB feedback.

Measurement of cognitive load in multimedia learning: a comparison of different objective measures

Different indicators are interesting for analyzing human learning processes. Recent studies analyze learning performance in combination with cognitive load, as an indicator for learners’ invested mental effort. In order to compare different measures of cognitive load research, the present study uses three different objective methods and one subjective method, reviewing the seductive details effect in a computer-based multimedia learning instruction. An experimental two-group design (N = 50) was used, with exposure to seductive details during learning as the between group factor. Eye movements were analyzed concerning the indicated cognitive activity and cognitive load was measured by the rhythm method (Park 2010; Park and Brunken 2015), the index of cognitive activity (ICA) (Marshall 2007), as well as by subjective ratings of mental effort and task difficulty (Paas 1992). Results confirm the seductive details effect for learning success with a decrease in retention and comprehension performance, an increase in total cognitive activity indicated by eye movements and significant higher cognitive load, indicated by rhythm method. The ICA values and the subjective ratings on mental effort and task difficulty show no difference in cognitive load between the groups. The results provide evidence of the suitability of different objective measures for a direct and continuous cognitive load assessment in multimedia learning. Further, the results show the benefit of combining different methods to gain detailed insight into information processing while learning with multimedia as well as a differentiated access to the single cognitive load factors.

The Anatomy of Traditional and E‐Learning Education: How the spatial ability of learners can impact learning outcomes

Technological innovation is changing the landscape of higher education, and the competing interests and responsibilities of today's learners have propelled the movement of post‐secondary courses into the online environment. In the anatomical sciences, computer‐aided instruction and online learning tools have become a critical component of teaching the intricacies of the human body when physical classroom space and cadaveric resources are limited. Our previous research (n=70) compared a simple 2‐dimensional e‐learning tool (A.D.A.M. Interactive Anatomy) to a more complex tool that allows for a more 3‐dimensional perspective (Netter's 3D Interactive Anatomy). Despite the differences in how these e‐learning tools present information, student ability to learn anatomical material, and their mental effort while doing so, known as cognitive load, were identical between e‐learning tools. However, when students with low spatial ability studied anatomical content with the more complex tool (Netter's 3D Interactive Anatomy), their performance scores were significantly lower than those students with high spatial ability (p=0.007). These results indicate that e‐learning tool software design can differentially influence students based on their spatial ability, but it remains to be determined if traditional kinesthetic‐tactile learning approaches, such as manipulating a skeleton, are also impacted by a student's spatial ability. Using a novel dual‐task methodology with a cross over design, undergraduate anatomy students from The University of Western Ontario, Canada (n=71) were evaluated as they studied a bony joint using a physical skeleton as well as a simple commercial software program (A.D.A.M. Interactive Anatomy). We hypothesized that the acquisition of anatomical knowledge by students, regardless of their spatial ability, would be superior when learning is associated with a real model, rather than currently available e‐learning tools. Students were assessed using a baseline knowledge test, Stroop observation task response times (a measure of cognitive load), MRT scores (a measure of spatial ability) and an anatomy post‐test (a measure of learning). Results suggested that while students may experience more cognitive load while studying using a physical skeleton (p&lt;0.001), it does not detrimentally impact their performance; in fact student performance was significantly higher when they studied using the skeleton (p&lt;0.001, R=0.46). Furthermore our results also demonstrated that students with low spatial ability are at a significant disadvantage when they learn the bony anatomy of a joint and are tested on images of the contralateral joint (p=0.023, R=0.326). This study highlights a major weakness in the strategy to move traditional anatomical education online, and suggests that students should be taught the anatomy of both sides of the human body, regardless of the reality that human limbs are mirror images of each other. These results can be further applied to the training of future surgeons and medical specialists, where surgical and medical procedures should be taught and practiced on both sides of the human body, to ensure that all students, regardless of spatial ability, can take their anatomical knowledge into the clinic and perform successfully.Support or Funding InformationSocial Science and Humanities Research Council, Government of Canada

Cognitive Load Measurement in a Virtual Reality-based Driving System for Autism Intervention.

Autism Spectrum Disorder (ASD) is a highly prevalent neurodevelopmental disorder with enormous individual and social cost. In this paper, a novel virtual reality (VR)-based driving system was introduced to teach driving skills to adolescents with ASD. This driving system is capable of gathering eye gaze, electroencephalography, and peripheral physiology data in addition to driving performance data. The objective of this paper is to fuse multimodal information to measure cognitive load during driving such that driving tasks can be individualized for optimal skill learning. Individualization of ASD intervention is an important criterion due to the spectrum nature of the disorder. Twenty adolescents with ASD participated in our study and the data collected were used for systematic feature extraction and classification of cognitive loads based on five well-known machine learning methods. Subsequently, three information fusion schemes-feature level fusion, decision level fusion and hybrid level fusion-were explored. Results indicate that multimodal information fusion can be used to measure cognitive load with high accuracy. Such a mechanism is essential since it will allow individualization of driving skill training based on cognitive load, which will facilitate acceptance of this driving system for clinical use and eventual commercialization.