• Is technology-independent;
• Has safety-related criteria;
• Is cost effective;
• Is appropriate for any system design; and
• Is validated through real-world testing.
The objective of the experiments in this Workpackage (WP2) was to investigate the impact of IVIS task load on driving performance and safety. To achieve this, two surrogate IVISs (SIVISs) were created, one representing cognitive load and the other visual load. Using these SIVISs, it was possible to vary secondary task load systematically. Separate assessments of the effects on driving of the different types of task load were carried out, with as clean a distinction as possible between visual and cognitive load.
The objective was also to identify the advantages and disadvantages of the different assessment methods (laboratory, simulator, field), and finally to identify which road types and scenarios are the most productive for testing IVIS. Different groups of drivers were used and scenarios varied in accordance with the protocol and procedure for safety assessment of IVIS as outlined in Deliverable 1 (Roskam et al., 2002).
A very large set of experiments was conducted. But in one sense this was one very large multi-national unified and integrated experiment with a common goal, a common experimental protocol and common indicators. The effect of IVIS use in three distinct road categories — urban, rural, and motorway — was investigated. To do this, a total of 14 separate driving simulator experiments were conducted, with each participant experiencing only one type of S-IVIS. All seven driving simulators were used to investigate driving with both S-IVISs on a common rural road. For the most part, each simulator road type had three levels of difficultly with the most difficult being driving when some critical event was triggered (the motorway had only two levels of difficulty: without and with events).
For the field (real road) studies, both types of S-IVIS were included in the drives for each participant, with the order of S-IVIS tasks counterbalanced. This was because, for these studies, there was no issue with the drivers learning what might happen in the critical events, since none were staged — all the driving was done in natural settings. The three field studies used different combinations of the road types and all roadway types were completed in a single session.
The overall number of experiments, both simulator and field, was 17. A total of 527 participants were used. A large number of indicators of driving performance, particularly related to longitudinal and lateral control, were collected. Also collected was information on secondary task performance (acting as the Surrogate IVIS), both static (not driving) and dynamic (whilst driving). The indicators can be classified into:
• Self-reported driving performance
• Lateral control
• Longitudinal control, i.e. control of speed and distance to a lead vehicle
• Workload, such as physiological measures and gaze behaviour
• Expert observations of driving performance
Following the anaysis of this data, a meta-analysis was carried out to compare the various studies and to identify the most effective indicators. This meta-analysis was intended to single out the most powerful scenarios and to assist in showing which indicators could be dispensed with in subsequent work testing the methods in evaluating some real IVISs.
The major dimensions of the study, to some extent in order of importance, were:
• IVIS type, visual versus cognitive
o Within IVIS type, IVIS level
o Within IVIS type, Static IVIS performance versus Dynamic IVIS performance
• Simulator and Laboratory studies versus Field Studies
o Simulator/Laboratory type
• Road category (urban, rural, motorway)
o Within Road category, Road level
• “Average” drivers versus Elderly drivers
• UK drivers versus Portuguese drivers
Findings across these dimensions were:
S-IVIS Type: The two types of S-IVIS had quite different effects on driving performance. The visual task had pronounced effects in terms of steering and lateral behaviour. On the other hand, the cognitive task caused reduced lateral deviation; in other words it “improved” steering behaviour, though there was also a tendency for drivers to compensate for the task load by shifting away from the road edge. This “improvement” in steering behaviour was accompanied by an increase in glances focussed on the road ahead, at the expense of the periphery. There were indications in some of the results that the predominant negative effect of the cognitive task on driving performance was on longitudinal control in car following.
S-IVIS level: Drivers were not always able to manage the trade-off between primary and secondary task, and there were many indications of driving performance being poorest when the secondary task demand was the highest. The elderly drivers were particularly poor at this task management.
Static S-IVIS Performance vs Dynamic Performance: Generally, the studies found that there was an interaction between S-IVIS performance across the baseline (static) and three levels of dynamic situation (i.e. the three levels of road difficulty). This advocates the HASTE approach of requiring the driving context to be considered in assessing an IVIS. Static performance did not reliably predict dynamic performance.
Simulator vs. Field: The field studies tended to pick up somewhat different effects of the systems than the simulator studies. Additionally, it has not proved possible to test elderly drivers with the visual task due to simulator sickness. This shows the value of the field tests, but also suggests that the incorporation of some additional scenarios or tests in the simulator roads should be considered. These could perhaps take the form of detecting objects in the periphery or detecting changes in the peripheral scene.
Simulator Type: The broad conclusion is that the type of simulator or laboratory used in the assessment did not have an effect.
Road Category: In the simulator studies, the rural road was the most diagnostic and the motorway the least diagnostic, i.e. the effect sizes from the rural road were generally larger. The urban road did not pick up any additional information that was not provided by the rural road. This means that, for simulator and laboratory assessments, the rural road can be used as the sole road category in the later work of HASTE assessing real IVIS systems as well as in the final HASTE test procedure. In the field studies with the cognitive task, the motorway produced the only indicator with a consistent effect.
Road Level: Road level is an important factor. It will be sensible in the later work of the project to consider dispensing with the easiest level of the road.
“Average” vs Elderly Drivers: The findings have confirmed the hypothesis proposed in Deliverable 1 (Roskam et al., 2002), that there would be severe problems for elderly drivers in using IVIS while driving, particularly at higher levels of task demand.
UK vs Portugal: The controlled comparison of the British and Portuguese showed the expected effect: the Portuguese drivers exhibited riskier driving behaviours. But, reassuringly, the analysis revealed there was no interaction effect of the “country” factor. In other words, results obtained with Portuguese drivers should be as reliable as those obtained with drivers from northern Europe.
As regards methodology, the results obtained form this very large set of studies confirm some of the initial decisions made in formulating the HASTE approach. There was clear value to the focus on dynamic evaluation, i.e. of looking at interaction with an IVIS while driving and of identifying the effects of that interaction on driving. Static testing cannot predict how an IVIS will affect steering behaviour or interaction with other road users. The different road levels proved their worth, particularly levels 2 and 3 of the rural road. There is also clear value to the inclusion of events (road level 3), but there is also some scope for improving the events so that the drivers are less able to adapt to their occurrence, by for example slowing down as the lead vehicle comes closer to them.
There may also be scope for the inclusion of peripheral detection tasks (PDTs) in the driving task, in order to gain a better understanding of drivers’ ability to assimilate information in the periphery, which is crucial to safety.
The results also confirm the value of using a very large number of indicators. Some of these indicators have turned out to be non-diagnostic and therefore can be abandoned in the next phase of the project. Others have turned out to be superfluous in that what they reveal overlaps with the diagnosis provided by other indicators. The meta-analysis has helped to sift through the indicators and test environments to identify the most powerful ones.
Important conclusions from the studies are:
The effect of the S-IVIS visual task on driving is very clear: increased distraction leads to problems in lateral control.
• The effect of the S-IVIS cognitive task is more complex, in that some driving parameters, particularly related to steering control and lateral position appear to improve. However, this improvement seems to be an artefact of greater concentration on the road straight ahead at the expense of information acquired from the periphery. Thought needs to be given to tasks or tests that might capture this loss of information acquisition from the periphery.
• Motorway driving in the various simulators and the laboratory was generally less diagnostic, than driving on other road types.
• Elderly drivers exhibited very risky driving while performing IVIS tasks
• The field studies provided some information that was not provided by the simulator assessments. The subsequent work in the project should consider simulator tasks that can provide analogous information.
|Publisher||European Commission EC|
|Number of pages||315|
|Publication status||Published - 2004|
|MoE publication type||D4 Published development or research report or study|