Volvo Technology Direct metrics of driver performance Johan Engström Volvo Technology Corporation Driver Metrics Workshop Ottawa, October 2-3, 2006 Volvo Technology Humans System Integration Volvo Technology Outline • • • Background of research (HASTE and AIDE) Metrics • • • Lane keeping Steering Eye movements • Time sharing • Gaze concentration Conclusions, lessons learned and topics for further research Volvo Technology Humans System Integration Volvo Technology Outline • • • Background of research (HASTE and AIDE) Metrics • • • Lane keeping Steering Eye movements • Time sharing • Gaze concentration Conclusions, lessons learned and topics for further research Volvo Technology Humans System Integration Volvo Technology Research in HASTE and AIDE on performance metrics • • • • • Research conducted 2003-2006 in the HASTE and AIDE EU-funded projects General objective of the studies: • Investigate systematically the effects of visual and cognitive load on driving performance -> define metrics for IVIS safety evaluation Data collected in simulators (of varying grade) and field (HASTE) Further analysed in AIDE Work reported here performed in collaboration with VTI (Swedish National Transport Research Institute) Volvo Technology Humans System Integration Volvo Technology The HASTE WP2 data set • • • • • Collected in HASTE WP2 during 2003-2004 9 parallel studies at different sites in Europe and Canada Same general methodology and experimental design Varied mainly with respect to test set-up (desktop simulator, meduim-high-fidelity simulators and field trials) Secondary tasks: Auditory Continuous Memory Task (aCMT) Visual: Arrows task Auditory/cognitive: aCMT 3 difficulty levels each Volvo Technology Humans System Integration Volvo Technology The HASTE WP2 data set (cont’d) • • Three general driving scenarios: Motorway, Rural, and Urban Present analyses based on data from three sub-studies • • • VTEC fixed-base simulator (rural and motorway) VTI moving-base simulator (rural and motorway) Volvo-VTI field study (instrumented Volvo S80 on motorway) VTEC simulator Volvo Technology Humans System Integration VTI simulator Volvo S80 Volvo Technology General result Visual and cognitive load have qualitatively different effects on driving… Volvo Technology Humans System Integration Volvo Technology Summary of results: Visual task Parameter Effect Interpretation Lane keeping variation + Reduced lateral control due to visual time sharing Steering wheel activity (e.g. reversals) ++ (mainly large reversals, 2-5 deg.) Increased steering effort to correct lane keeping errors Speed - Compensation for reduced lateral control to maintain safety margins Headway and TTC + Compensation for reduced lateral control to maintain safety margins Glance frequency ++ (increased with Increased visual complexity -> more task difficulty) glances required Glance duration ++ (increased with Increased visual complexity -> longer task difficulty) glances required Volvo Technology Humans System Integration Volvo Technology Summary of results: Cognitive task Parameter Effect Interpretation Gaze concentration to road centre ++ Less cognitive resources available for visual monitoring to the periphery Lane keeping variation - Indirect effect of increased gaze concentration Steering wheel activity (e.g. reversals) ++ (mainly small reversals.) More active and precise steering as a result of more visual input Speed +- (inconsistent) Dependent on test scenario Mean Headway and TTC No effect No compensation Headway variation + Less consistent car following Volvo Technology Humans System Integration Volvo Technology General conclusion: Visual and cognitive load have different effects • • Visual Visual diversion • Steering hold • Lane keeping error • • Large corrective steering movements Slowing down & increasing headway to compensate Cognitive • Interference with attention selection mechanisms • • • • Volvo Technology Humans System Integration Gaze concentrates to road centre More visual control input than during normal driving More active and precise steering More accurate lane-keeping Reduced visual detection/ decision making Volvo Technology Outline • • • Background of research (HASTE and AIDE) Metrics • • • Lane keeping Steering Eye movements • Time sharing • Gaze concentration Conclusions, lessons learned and topics for further research Volvo Technology Humans System Integration Volvo Technology Focus of this presentation • • Metrics intended for task-based IVIS evaluation Types of metrics covered: • • • Lane keeping Steering Eye movements Volvo Technology Humans System Integration Volvo Technology Outline • • • Background of research (HASTE and AIDE) Metrics • • • Lane keeping Steering Eye movements • Time sharing • Gaze concentration Conclusions, lessons learned and topics for further research Volvo Technology Humans System Integration Lane position (m) ion (m) km/h) St. wheel angle (deg) St. Radial gaze (deg) wheel angle (deg) Radial gaze ( ion (m) Lane position (m) km/h) Radial gaze (deg) St. wheel angle (deg) St. wheel angle (deg) Radial gaze 50 50 30 40 4030 20 Volvo Technology 20 3020 3020 20 10 20 10 Example data: Straight 1010 1010driving, rural road, Driver straight rural road 7, 39, straight 80 130 100 ht ruralDriver road 120 140 rural road 940 760950 50 30 40 20 30 20 20 10 VTEC 10 10 simulator 760 780 Baseline Baseline Baseline 4 4 5 30 50 30 40 22 200 30 020 0 20 10 -2 10 -2 10 -5 80 130 100 120 140 80 100 120 130 780 140 760 4 5 30 50 30 40 2 2 200 30 020 0 2010 -2 10 -2 10 -5 760950 780 940 960 940760 950120780 960 14 1000 1010 120 1020 960 980 Cognitive task, level Visual task, level 3 23 Cognitive task, level 4 4 5 50 30 30 402 2 20 0 30 20 0 0 2010 -2 10 -2 10 -5 120 14 1000 1020 120 1000 1010 1020 140 140 960 1010 980 0.6 4 0.50.6 54 0.4 0.4 22 00.2 0.2 0 00 00 -2 -0.5-0.2 -2 -0.2 -5 120 140 80 130 100 80 100 120 130 780 140 760 0.6 4 0.5 0.6 54 0.4 0.4 22 0 0.2 0.2 0 00 00 -0.5-0.2 -2-2 -0.2 -5 940 960 760950 780 940760 950120780 960 0.6 4 0.5 0.6 4 5 0.4 0.4 22 0 0.2 0.2 00 0 0 0 -2 -0.5-2 -0.2 -0.2 -5 1000 1020 120 14 120 1000 1010 1020 140 140 960 1010 980 120 120 0.6 0.6 85 0.5 Technology 0.4 Volvo 110 Humans System Integration 0.4 110 80 780 960 120 140 Baseline Baseline 4Cognitive task, level 3 120 120 0.6 0.6 85 0.5 0.4 110 0.4 110 80 120 120 0.6 0.6 85 0.5 0.4 110 0.4 110 80 Volvo Technology Individual subjects Frequency analysis 0 10 -1 10 -2 10 -3 10 -4 1 Frequency content in st. wheel signal 10 -1 10 0 10 Frequency (Hz) 1 10 Driver 8, fixed-base simulator, straight rural road All subjects 10 Baseline Visual task Task length frequency 1 10 10 -2 10 0 Baseline Visual task Task length frequency 1 10 0 10 -1 10 -2 10 -3 10 -4 10 10 10 -2 10 -1 -2 10 -2 10 -1 10 Frequency (Hz) 0 10 1 Frequency content in st. wheel signal Frequency content in st. wheel signal Baseline, fixed base simulator, motorway, joint sequence, all drivers Baseline, fixed base simulator, rural straight, joint sequence, all drivers Baseline, fixed base simulator, rural curved, joint sequence, all drivers Baseline, moving base simulator, rural straight, joint sequence, all drivers Baseline, moving base simulator, rural curved, joint sequence, all drivers Baseline, field, motorway, joint sequence, all drivers Task length frequency Frequency content in st. wheel signal Driver 1, fixed-base simulator, straight rural road -1 10 0 1 10 10 Frequency (Hz) Driver 5, fixed-base simulator, straight rural road Baseline Visual task Task length frequency 1 10 0 10 -1 10 -2 10 -3 10 -4 Volvo Technology Humans System Integration 10 -2 10 -1 10 0 10 Frequency (Hz) 1 10 Volvo Technology Common types of lane keeping metrics Focus here Position-based TLC (Time-to-line-crosssing) -based e.g. standard deviation of lane position (SDLP) e.g. proportion of lane exceedences (LANEX) e.g. mean of TLC minima (MN_TLC) e.g. proportion of TLC minima < X s (PR_TLC) Continuous Volvo Technology Humans System Integration Event-based Non-normal distribution & too few instances -> difficult to use for task-based evaluation Less sensitive than position-based metrics and yield roughly similar results -> no obvious advantage for present purposes Volvo Technology (Modified) Standard deviation of lane position (SDLP) (1) • Operational definition (AIDE D2.2.5 – Östlund et al. 2006): • ”Standard deviation of lateral position data, high-pass filtered with a cutoff frequency of 0.1 Hz, where lateral position is defined as the average distance between the right side of the front or rear right wheel and the inner (closest) edge of the right hand lane marking.” Volvo Technology Humans System Integration Volvo Technology SDLP depenency on data duration High-pass filtering needed to overcome this problem (Östlund et al., 2006) Volvo Technology Humans System Integration Volvo Technology Representative results from HASTE on SDLP VTEC simulator, rural road 0,6 BL 0,4 SLv1 0,3 SLv2 0,2 SLv3 0,1 0 Straight Visual task Volvo Technology Humans System Integration Curve st_lp (m) st_lp (m ) 0,5 0,4 0,35 0,3 0,25 0,2 0,15 0,1 0,05 0 BL SLv1 SLv2 SLv3 Straight Curve Cognitive task Volvo Technology (M)SDLP pros and cons • • • Advantages • • Easy to measure, at least in the simulator (feasible also in the field using off-the-shelf lane-tracking systems) Straightforward general interpretation as performance metric Disadvantages • • • Only moderately sensitivite to secondary task task load Strongly sensitive to environment factors (e.g. curvature, lane width) Sensitive to discontinuities due to lane changes and exceedences Relation to crash data • Open issue – no strong direct evidence of causal relation between increased SDLP and crash risk (however, indirect evidence via visual distraction) Volvo Technology Humans System Integration Volvo Technology Outline • • • Background of research (HASTE and AIDE) Metrics • • • Lane keeping Steering Eye movements • Time sharing • Gaze concentration Conclusions, lessons learned and topics for further research Volvo Technology Humans System Integration Volvo Technology Example data: Straight driving, rural road, VTEC simulator Radialgaze gaze(deg) (deg) St. wheel wheel angle angle (deg) (deg) Radial St. wheel angle (deg) St. wheel angle (deg) Radial gaze (deg) ht rural road Driver 7, straight rural road Driver 39, straight straight rural road Driver 39, rural road Driver 7, straight rural road ht rural road Baseline Baseline Baseline Baseline Baseline Baseline 50 3030 50 30 40 30 40 20 30 20 20 30 20 20 1010 20 10 10 10 10 80130 100 120 140 760 80 100 760 780 120 130 780 140 44 5 5 44 222 0 0 000 4 4 5 5 44 2222 00 0000 -2 -2-2 -2 -2 -2-2 n (m) -5-5 80130 100 80 100 120 140 760 780 760 120 130 780 140 (m) Cognitive task, level 3 Baseline Baseline Baseline Baseline Cognitive task, level 3 50 3030 50 30 40 30 40 20 30 20 20 3020 20 1010 20 10 10 10 10 760 780 140 940 950 960 780 120 940 760 950 120 960 140 0.6 Technology 0.6 0.5 0.50.6 Volvo 0.4 Humans System Integration -5-5 760 780 780 940 950 960 120 940 760 950 120 960 0.6 0.6 0.6 0.5 0.50.6 0.4 0.4 Cognitive task, level 2 3 Visual task, level 3level Cognitive task, Cognitive task, level Visual task, level 323 Cognitive task, level 5030 30 50 30 4030 40 20 3020 302020 2010 10 20 101010 10 120 14 1000 1020 960 980 9601010 980 1000 1010120 1020 140 4 4 55 4 4 22 2 2 0000 0 0 -2 -2-2-2 -5 -5 120 1000 1020 140 9601010 980 140 960 980 1000 1010120 1020 0.6 0.6 0.6 0.6 0.5 0.5 0.4 0.4 14 140 0 Volvo Technology Metrics investigated in AIDE • • • • Standard deviation of steering wheel angle High frequency steering – 3 versions Steering entropy – 2 versions (Boer, 2000;Boer, 2005) Steering wheel reversal rate – 2 versions (HASTE version; Modified version developed in AIDE, Markkula and Engström, 2006) Volvo Technology Humans System Integration Volvo Technology Results in sensitivity (effect size) – visual load Visual All metrics fairly sensitive in all conditions except Standard Deviation 1.6 Standardised effect size 1.4 1.2 Fixed, mw Fixed, rural, straight 1 Fixed, rural, curve 0.8 Moving, rural, straight 0.6 Moving, rural, curve 0.4 Field 0.2 St an da r d de via Re tio n ve rs al ra Re te 1 ve rs al ra te HF 2 st ee rin g1 HF st ee rin g2 HF st ee St ee rin rin g3 g en St tr o ee py rin 1 g en tr o py 2 0 Markkula and Engström (2006) Volvo Technology Humans System Integration Volvo Technology Results in sensitivity (effect size) – cognitive load Cognitive Reversal Rate2 and Steering entropy most sensitive 1.4 Standardised effect size 1.2 1 Fixed, mw Fixed, rural, straight 0.8 Fixed, rural, curve Moving, rural, straight 0.6 Moving, rural, curve Field 0.4 0.2 St an da r d de via Re tio n ve rs al ra Re te 1 ve rs al ra te HF 2 st ee rin g1 HF st ee rin g2 HF st ee St ee rin rin g3 g en St tr o ee py rin 1 g en tr o py 2 0 Markkula and Engström (2006) Volvo Technology Humans System Integration Volvo Technology Steering entropy (1) • • Operational definition • Entropy of the prediction errors made by a linear predictive filter applied on the steering wheel angle signal (see Boer 2005 for detailed mathematical definition) Interpretation • ”…increase in high frequency steering corrections that result after periods of diverted or reduced attention (i.e., in response to a perceived vehicle drift outside the acceptable tolerance margins that mounted during these periods of degraded information)” (Boer, 2005) Volvo Technology Humans System Integration Volvo Technology Steering Entropy pros and cons • • Advantages • • • Disadvantages • • • • Strongly sensitive to visual and cognitive load in a range of conditions SW data easy to measure, also in the field Relatively robust to differences in driving environment (road type, curvature, test set-up etc.) • • Fairly complex to compute (though straightforward) Somewhat difficult to interpret, even in terms of performance (increased SE may indicate both increased and reduced lateral control) Interpretation of free parameters (alpha and re-sampling rate) not entirely straightforward Requires baseline data for computation of task condition data ”Normalisation” to baseline data makes BL and Task data somewhat dependent Relation to crash data • No established relation to crash data (only indirectly via visual distraction) Volvo Technology Humans System Integration Volvo Technology Steering Wheel Reversal Rate (SRR) • Operational definition • The number, per minute, of steering wheel reversals larger than a certain angular value referred to as the gap size (see Markkula & Engström, 2006, for detailed mathematical definition) Volvo Technology Humans System Integration Volvo Technology Representative results from HASTE: SRR1, 1 degree gap size VTEC simulator, rural road 16 14 14 12 10 SLv1 8 SLv2 6 SLv3 4 BL 10 BL rr_st1 rr_st1 12 8 6 SLv1 4 SLv3 SLv2 2 2 0 0 Straight Curve Visual task Volvo Technology Humans System Integration Event Straight Curve Event Cognitive task Volvo Technology SRR Sensitivity (effect size) as a function of gap size 1.6 Visual, fixed, mw 1.4 Cognitive fixed mw Standardised effect size (σ) 1.2 Visual, fixed, rural, straight 1 Cognitive, fixed, rural, straight Visual, fixed, rural, curve 0.8 Cognitive, fixed, rural, curve 0.6 Visual, field 0.4 Cognitive, field Visual, moving, rural, straight Cognitive, moving, rural, straight Visual, moving, rural, curve 0.2 0 0.1 0.5 1 2 3 -0.2 Gap size (degrees) -0.4 Volvo Technology Humans System Integration 4 5 10 Cognitive, moving, rural, curve Volvo Technology SRR pros and cons • • • Advantages • • • • Strongly sensitive to visual and cognitive load in a range of conditions SW data easy to measure, also in the field Easier to interpret than steering entropy Does not involve normalisation of task data to baseline data (like Steering Entropy) Disadvantages • • Sensitive to environment factors Somewhat difficult to interpret in terms of performance - increased SRR may indicate both reduced and increased lane keeping performance (however, can be tuned by changing gap-size) Relation to crash data • Like other steering wheel metrics, no established relation to crash data (only indirectly via visual distraction) Volvo Technology Humans System Integration Volvo Technology Outline • • • Background of research (HASTE and AIDE) Metrics • • • Lane keeping Steering Eye movements • Time sharing • Gaze concentration Conclusions, lessons learned and topics for further research Volvo Technology Humans System Integration Volvo Technology Example data ition (m) St. wheel angle (deg) Radial gaze (deg) ition (m) St. wheel angle (deg) Radial gaze (deg) Driver 7, straight ruralrural roadroad Driver 39, straight ht rural road Baseline Baseline Baseline 50 30 40 30 30 20 20 20 10 10 10 50 30 4030 3020 20 20 10 1010 120 140 80 130 780 100 760 5 44 22 0 00 -5 Baseline Baseline Cognitive task, level 3 940760 950120780 960 50 3030 40 30 2020 20 1010 10 140 5 44 22 0 00 -2-2 120 140 80 130 780 100 760 0.6 0.50.6 0.4 Volvo Technology 0.4 Humans System Integration -2-2 -5 940760 950120780 960 0.6 0.6 0.5 0.4 0.4 Visual task, levellevel 3 3 Cognitive task, Cognitive task, level 2 1000 1020 140 120980 960 1010 54 4 22 00 0 140 -2-2 -5 1000 1020 140 120980 960 1010 0.6 0.6 0.5 0.4 0.4 Volvo Technology Outline • • • Background of research (HASTE and AIDE) Metrics • • • Lane keeping Steering Eye movements • Time sharing • Gaze concentration Conclusions, lessons learned and topics for further research Volvo Technology Humans System Integration Volvo Technology Factors to account for • • • • Total time spent looking away from the road Intensity (”how much looking away per time untit”) Distribution of single glance durations Eccentricity On-road A Off-road B On-road Off-road C On-road Off-road Volvo Technology Humans System Integration Volvo Technology Traditional (ISO 15007) glance-based metrics Measure Definition Glance frequency The number of glances to a target within a pre-defined time period, or during a pre-defined task, where each glance is separated by at least one glance to a different target (ISO 15007). Single glance duration Time from the moment at which the direction of gaze moves towards a target to the moment it moves away from it (ISO 15007). Mean single glance duration The average duration of the glances towards a target. Number of glances > 2 seconds The number of glances towards the system with a duration longer than 2 seconds. Total glance time (towards a target) Total glance time (or percentage of time) associated with a target (e.g. in-vehicle device). Volvo Technology Humans System Integration Volvo Technology Automating the ISO 15007 metrics: The VDM Tool (Larsson, 2002; Johansson et al., 2006) Volvo Technology Humans System Integration Volvo Technology Difficulties with automating the ISO 15007 metrics • • • • Standard originally intended for manual transcription Glance-based metrics are very sensitive to noise Requires careful calibration and signal pre-processing Much data still needs to be discarded (~30% in HASTE) Volvo Technology Humans System Integration Volvo Technology An alternative: Road centre-based metrics Road Centre On-road glances Off-road glances Volvo Technology Humans System Integration Volvo Technology Percent Road Centre (Victor, 2005) • Operational definition: • • PRC-Task: The percent of fixations directed towards the road centre (RC) during a task. Represents intensity only. PRC-Window: The percent of fixations directed towards the RC during a moving time window of 1 minute. If the task is shorter than 1 minute, the remaining time is completed with a constant PRC of 80%. The windowing adds a weighting for task duration. Volvo Technology Humans System Integration Volvo Technology Example data from HASTE (Victor, Harbluk and Engström, 2005) Volvo Technology Humans System Integration Volvo Technology Pros and cons of PRC • • • Advantages • Very sensitive to visual task difficulty • Allows for baseline data (which glance-based metrics to do not) • Should be more robust to measurement noise (focus measurement where eye tracking accuracy is normally best, data order does not matter) Disadvantages • PRC-Task measures only intensity • PRC-Window accounts for task duration, but somewhat arbitrarily • Does not account for eccentricity Relation to crash data • Strong empirical evidence on the relation between visual diversion from the forward road scene and accident risk (e.g. Wierwille and Tijerina, 1995; Klauer et al., 2006) Volvo Technology Humans System Integration Volvo Technology Further ideas • • • • • RC-based versions of the ISO metrics (Kronberg et al., 2006) Other ways to account for both intensity and duration Weighting function for single glance duration Account for eccentricity For example: VD g i Volvo Technology Humans System Integration 3/ 2 i E ( , ) gi=single off-road glance duration E=eccentricity weighting function Volvo Technology Outline • • • Background of research (HASTE and AIDE) Metrics • • • Lane keeping Steering Eye movements • Time sharing • Gaze concentration Conclusions, lessons learned and topics for further research Volvo Technology Humans System Integration Volvo Technology Measuring gaze concentration: Standard deviation of gaze angle • Operational definition • The standard deviation of the combined horizontal and vertical angles. The combined angle is the square root of the sum of squared vertical and squared horizontal angles (Pythagoras theorem) and thus is a one-dimensional angle between the origin and a gaze point Volvo Technology Humans System Integration Volvo Technology Effects of cognitive task on gaze concentration Gaze angles (pich and yaw) Baseline Cognitive task (levels 1-3 aggregated) VTEC simulator, rural road Volvo Technology Humans System Integration Volvo Technology Example data from HASTE (Victor, Harbluk and Engström, 2005) Volvo Technology Humans System Integration Volvo Technology Pros and cons of SD gaze angle • • • Advantages • Sensitive to cognitive load (more than PRC) – good metric of gaze concentration • Robust to noise since data order does not matter Disadvantages • Only applicable to assessment of purely cognitive load Relation to crash data • No empirical data on the relation between gaze concentration and crash risk Volvo Technology Humans System Integration Volvo Technology Outline • • • Background of research (HASTE and AIDE) Metrics • • • Lane keeping Steering Eye movements • Time sharing • Gaze concentration Conclusions, lessons learned and topics for further research Volvo Technology Humans System Integration Volvo Technology Conclusions and lessons learned • The metrics addressed here mainly relevant for evaluating visually demanding tasks • • • • • Lateral control performance metrics somewhat problematic as surrogate safety metrics – no clear link to crash data Direct eye movement metrics seem to be the most promising (though still practical difficulties with data collection and analysis) For cognitive tasks, other metrics are needed to capture the main safetyrelevant effects (e.g. detection task metrics such as PDT) Lack of agreed driver model – very little consensus on how to interpret even the most common driving performance metrics Little discussion and emprical work on the link between performance metrics and safety (especially in Europe) Volvo Technology Humans System Integration Volvo Technology Topics for further research • • • • • Development of metrics representing sensory-motor coordination (e.g. correlation between steering and eye movements) More comprehensive visual demand metrics, taking into account both duration, intensity and eccentricity Establish relation between different performance metrics and crashes (using data from naturalistic field studies) -> valid criteria for IVIS safety evaluation and ADAS safety benefits analyses Investigate how to incorporate exposure data (frequency of use) into the IVIS evaluation methods (e.g. in a general formula for visual demand exposure) Establish stronger theoretical foundation for driving performance assessment • • Multiple resource theory (Wickens, 2002) does not explain all variance in the data (e.g. driver adaptation and effects of cognitive load in terms of gaze concentration, and ”improved” lane keeping) Incorporate modern perception and attention (”active vision”) theories into driving research (see Victor, 2005) Volvo Technology Humans System Integration Volvo Technology References • • • • • • • • • • • Engström, J., Johansson, E and Östlund, J. (2005). Effects of visual and cognitive load in real and simulated motorway driving. Transportation Research Part F, pp. 97-120. HASTE special issue (contains all papers from the WP2 studies): Transportation Research Part F: Traffic Psychology and Behaviour, Volume 8, Issue 2 Johansson, E., Engström, J., Cherri, C., Nodari, E., Toffetti, A., Schindhelm, R., Gelau, C. (2004) Review of existing techniques and metrics for IVIS and ADAS assessment. EU project AIDE, project IST-1-507674-IP, Deliverable 2.2.1 Johansson, E., Kronberg, P., Victor, T., Martens, M., Chin, E. and Nathan, F. (2006). Visual demand measurement tool development. AIDE Deliverable 2.2.2. EC contract No. IST-1-507674-IP. Kronberg, P., Victor, T. and Engström, J. (2006). Road-centre-based measures of visual demand. Vision in Vehicles, Dublin. Larsson, P. (2002). Automatic Visual Behavior Analysis. Dissertation for a Master of Science Degree Applied Physics and Electrical Engineering Control and Communication Department of electrical engineering Linköping University, Sweden. LiTH-ISY-EX-3259. Markkula, G. and Engström, J. In press. A Steering Wheel Reversal Rate Metric for Assessing Effects of Visual and Cognitive Secondary Task Load. ITS World Congress, London 2006. Östlund, J., Peters, B., Thorslund, B., Engström, J., Markkula, G., Keinath, A., Horst, D., Mattes, S. Foehl, U. 2005. Driving performance assessment: Methods and metrics. AIDE Deliverable 2.2.5. European Commission, IST-1-507674-IP. Östlund, J. Carsten, O., Merat, N., Jamson, S., Janssen, W., & Brouwer, R., et al. (2004). Deliverable 2— HMI and safety-related driver performance. Human Machine Interface And the Safety of Traffic in Europe (HASTE) Project. Report No. GRD1/2000/25361 S12.319626. Victor TW (2005) Keeping eye and mind on the road. PhD Thesis, Uppsala University, Sweden. Victor, T. W., Harbluk, J. L. & Engström, J. (2005). Sensitivity of eye-movement measures to in-vehicle task difficulty. Transportation Research Part F 8:167-190 Volvo Technology Humans System Integration