More than 500 people die and over 10.000 get injured annually in the Netherlands due to car accidents (SWOV 2014). One of the causes of accidents is distraction by for instance a cell phone or navigation system that are more and more used in the car nowadays. In this study the influence of distraction on the car following task is examined as well as the implications on traffic flow. For this study, data from a simulator experiment (Bellet and Bornard 2012) is used. In the experiment the participants drove an urban scenario, with a visual, cognitive and without secondary task (ST). With the visual ST participants had to indicate which of three pictograms match a fourth one appearing some seconds later on a display. The cognitive ST consists of indicating whether a letter did already appear in a sequence of letters orally presented to the participants. There were two following conditions, one free following and one constrained following conditions where they needed to keep 0.6s Inter-Vehicular-Time (IVT). Additional feedback was present about whether they were on this value. The different conditions of this experiment are statistically analysed with a paired t-test. The visual, cognitive and no secondary task scenarios in the free following and constrained condition are compared, which makes a total of 9 t-tests. The average distance to the lead car, average absolute speed difference and average inter vehicular time (IVT) are used as car following performance measures. As expected, the free following and constrained condition are significantly different on all measures. Main result is that with a visual ST in the free following condition, the subjects drive closer to the lead car, with less speed difference, compared to the no ST scenario. In the constrained following condition, this effect is not seen. Even a significant increase in speed difference is seen between the visual and no ST scenarios. Besides the car following performance, also the control effort of the participants is statistically analysed with a paired t-test. The measures used for control effort are: Root Mean Squared Error (RMSE) of derivative of steering wheel rotation; RMSE of derivative of gas pedal depression; RMSE of derivative of brake pedal depression and the frequency of brake actions. Expected is that the drivers show an intermitted controlling behaviour due to the secondary task. Only a decrease of control effort for the gas pedal, on the straight parts during a secondary task is seen but no discontinuous control is observed. Despite the secondary task, the participants keep on controlling the car continuously with almost the same effort. The data from the simulator experiment is used to estimate the parameters for the Helly and IDM car following model. These are both continuous models which are expected to describe the distracted behaviour adequate, because no evidence for intermittent control is found in the experimental data. An attempt is made to solve the known problem of the standard Helly model of underestimating the braking phase by introducing an extra gain during the braking phase. Another Helly variant (st-switch) is introduced which estimates a different gain for the velocity input during the secondary task. These models are compared to the standard Helly and IDM model. The brake model does perform slightly better regarding the variance accounted for (VAF), but it is still underestimating the braking phase. The same holds for the IDM model. The parameters estimated for the visual ST condition capture the distracted behaviour. The higher kv-value indicates the lower speed difference found in the data. The lower hv-value will lead to less distance to the lead car. From the st-switch model, it can be seen that the estimated kv-value during the visual secondary tasks doesn’t differ much from the kv-value during the period where no secondary task is presented. This leads to the conclusion that the participants didn’t change their behaviour during the secondary task. But the fact that a visual secondary task needed to be performed during the driving task made them more vigilant during the whole run. The multiple resource theory (Wickens 2002) partly describes the effect of an increase of mean speed difference in the visual ST scenario, compared to the no ST scenario in the constrained condition. The multiple resource theory says that a secondary task that draws from the same resources as the primary task will lead to compensatory behaviour in the primary task. This compensatory behaviour is found by several studies in literature (Jamson et al. 2004; Strayer, Drews, and Johnston 2003; Strayer and Drews 2004)., where drivers increased the distance to the lead car. In this study, this compensation on distance with a secondary task is not seen. In the free following condition, the mean distance to the lead car, mean speed difference and IVT even decreased in the visual ST scenario compared to the no ST scenario. This is unexpected and can’t be explained with the multiple resource theory. The inverse u theory (Hancock 1989), which describes the relation between workload and performance can explain this increased car following performance. This relation says that in an underload or overload situation, thus when the workload is too low or too high, the performance is not optimal. The free following task was probably too easy and the drivers were in an underload situation. The secondary task increased the workload and made the drivers more vigilant, as a consequence the performance increased with a secondary task compared to the no ST scenario. In the constrained condition, the workload was already high due to the additional task of keeping the vehicle at 0.6s IVT. During the secondary task, the participants keep continuously controlling the vehicle. This can be explained by the use of the internal mental model of the current driving situation to anticipate the future states. The deceleration of the lead car can also be incorporated in this mental model, because the speed difference of the lead car is due to the track layout. It is also possible that the participants perceive the changes of the lead car speed via their peripheral vision and are therefore able to continuously control the car. To see the influence of distraction on the road capacity a traffic flow simulation will be carried out. The Helly and IDM model describe the distracted behaviour as good as the non-distracted behaviour and can both be used to perform the traffic flow simulation of a distracted driver. This is done by simulating a fleet of vehicles on a single lane road stretch of 4000 m with an increasing inflow from 500 to 4000 veh/h. It is concluded that with a visual secondary task, the road capacity increases compared to the no ST condition, with 33%, from 2250 to 3000 veh/h. The Helly model is less suited for the traffic flow simulation because it only shows the free flow region and gets unstable in the congested region. A couple of limitations should be considered, first the study is conducted in a simple driving simulator. Therefore the results are not one on one transferable to driving in the real world. The speed variations of the lead car were mainly due to the track layout, so the participants might anticipate on this speed behaviour. This would make the car following task partly a track following task which is not captured in the car following models. Furthermore no additional hazards (crossing children, other cars on an intersection) were presented. Last limitations are the short scenarios of 90 seconds. The main conclusions of this report are that the effect of a secondary task depends on the primary driving task. In the free following condition, which is a low workload task, the secondary task has a positive effect on car following performance. In the constrained conditions the participants were asked to drive very close (0.6s IVT) to the lead vehicle the secondary task had no effect on the car following performance. Although the participants engaged in a visual and cognitive secondary task, the vehicle is continuously controlled. Therefore the continuous control models can be used to describe this distracted behaviour. This is confirmed by the kv gain of the model which didn’t change during the visual ST. The traffic flow simulation showed an increase of road capacity of 33% for the scenario of visual distraction compared to no distraction. The results of this study can be interesting for the automated vehicles. When the driver isn’t controlling the vehicle, in the automated mode, the ‘driving’ task becomes quite easy and dull. Therefore it is likely that when the driver is reclaiming control, the performance is not optimal. Future studies can focus on the effect of a secondary task in the automated vehicles in a real car experiment. It would be interesting to choose real secondary tasks like sending text messages, engaging in social media contacts or reading a paper. Furthermore it would be interesting if the results of this study also hold when the participants are driving for a longer period of time and engaging in a secondary task. (Author/publisher)
Abstract