Analysis of car simulator data

Abstract: Simulators are being used for a wide variety of purposes, not least for vehicle simulation as an aid in driver behaviour analysis. Common values to be measured are reaction times such as brake and steer reaction time. One problem with monitored simulator studies is that there are a limited number of test drivers. With only 50-100 test drivers, presented reaction times from the studies are often mean values. From a safety analytic perspective, it would be more interesting to show the entire distribution of values. The unmanned simulator at Universeum gives a large number of test drivers (~40k / year) which makes it possible to investigate entire distributions of values. Another benefit with the unmanned study at Universeum is that drivers tends to act more natural than in a monitored study, where drivers are aware of that they are being observed. On the other hand, the fact that users are not being observed, lead to a lot of questionable data. Drivers are exploring and do not behave as they would in real traffic. The main objective with our thesis project has been to find algorithms to extract trustworthy data from the simulator. It has been proven that there are large amounts of data that can be used for driver behaviour analysis. Methods to calculate common measures used in traffic safety analysis have been developed. An updated simulator software, better adjusted for an unmonitored study, has been developed and installed in the Universeum simulator. Summaries from the different scenarios in the simulator K2 Summary There are quite few drivers in the K2 scenario where a brake signal has not been registered at all. This is the main issue with the scenario and the explanation to this is that the scenario is very sensitive to the speed kept by the user. Speeding drivers will not experience the situation at all and since so many drivers are speeding, a lot of data is lost. By adapting the speed of the mover to the speed of the driver this problem would be reduced. The sensitivity to speed is also noticeable in the plots as the BRT seems to have very little influence on the results. The BRTs span from ~0.4 seconds to 2 seconds which is in range with other studies. [P1] There are some drivers steering but they are not many enough to draw any valid conclusions. K3 Summary The K3 scenario works much better than the other crossing scenarios, producing measurable data at a much higher rate than the other scenarios. This is probably thanks to the MeetAtPoint function which takes the drivers speed into consideration. The wider acceptance to speeds in the scenario gives better possibilities to analyse the impact of BRT and deceleration on the result. There are some drivers steering but they are too few to present any valid distributions. K5 Summary This scenario gives the least percentage of valid files among the K-scenarios. This is due to the very high urgency in the scenario. This together with the fact that the speed of the mover is not adapted to the speed of the driver allows very few drivers to actually experience the scenario at all. By adapting the speed of the mover to the speed of the driver, drivers will be able to experience the situation at a much higher rate than before. There are very few drivers steering instead of braking. Interesting is that, remembering that there are very few entries to investigate, the steering option has been quite successful in this scenario compared to the other K scenarios. U1 Summary The percentage of valid files from this scenario is at the same level as the other scenarios, i.e. quite low. The reason to the low percentage in this scenario is however not the speed kept. One big loss is the drivers which chooses to overtake the braking lead car and therefore fails to enter the situation as intended. The other problem is the distance between the lead car and the driver. The distance between them as the lead car brakes varies a lot and the BRT´s will therefore also vary. There are almost no drivers that try to steer away from the car instead of braking. U4 Summary The percentage of files where a brake signal has been registered is in level with the other scenarios. One explanation to the few registered brake signals is, as in the U1 scenario, that a lot of drivers choose to overtake the lead car and will therefore not experience the scenario at all. Overall the U4 data and results are questionable. Even when normalising the BRT by TTC no satisfying distributions can be found. The amount of files holding the U4 scenario (1761 or 23 % of all scenario files) are also remarkable high. And apart from the files holding a full scenario, there are a lot of files that holds only U4 start messages without any stop messages. This indicates that the U4 trigger points are trigged even when they are not supposed to. This will be investigated and discussed in chapter 3.1.5. There are a few drivers steering, an option that has proven quite successful in this scenario. M1 Summary The urgency in the scenario is low, which makes it possible for drivers to avoid a collision without steering. The lane position does not play any significant role on the type of reaction, but many drivers have a lane position to the right. A high amount of drivers have an offset which gives them a position more than 0.5 meters out on the road verge. The road verge is 3.2meters wide, which is much wider than in reality. The explanation to the high number of drivers that drives on the road verge can be that the width of the lane (3.2 meters) corresponds to the width of a road with speed limit 70km/h. Drivers may experience the road as narrow at a speed around 90km/h and therefore sometimes choose to drive on the road verge, which looks perfectly fine to drive on. [D] If the verge were less wide and looked less appealing to drive on this behaviour would probably be reduced. In reality the road verge on a road with a speed limit of 90km/h is 50cm. [I5] It was not possible to use the SWRR measure to analyse the driver behaviour. The SWRR is higher on parts of the road where it is more difficult for the driver to follow the road as in long turns. The SWRR increases linear as the speed increases but are too spread to make any conclusions from. M2 Summary The urgency in the scenario is high. It is hard to avoid a collision without steering. The lane position does not play any significant role on the type of reaction, but many drivers have a lane position to the right. It is not a normal behaviour to drive on the road verge and the meeting situation occurs in a left curve and a lane position to the right is not expected as many drivers would like to cut the curve, giving a lane position to the left. Despite the initial lane position, the most common type of reaction is steering right. 3,6% of all drivers are out on the grass (out of road). This can be considered as a high amount and probably depends on the wide road verge and the flat grass area, which looks perfectly fine to drive on. As in the M1 scenario, the wide, fully driveable, road verge is probably the reason to why 78% of the drivers choose to drive a least 50cm out on the road verge as they try to avoid the situation. Another explanation to the high number of drivers that drives on the road verge can be that the width of the lane (3.2m) corresponds to the width of a road with speed limit 70km/h. Drivers may experience the road as narrow at a speed around 90km/h and therefore sometimes choose to drive on the road verge, which looks perfectly fine to drive on. [D] It was not possible to use the SWRR measure to analyse the driver behaviour. The SWRR is higher on parts of the road where it is more difficult for the driver to follow the road as in long turns. The SWRR increases linear as the speed increases but are too spread to make any conclusions from.