|Abstract||Roundabouts are considered a safer intersection treatment compared to traditional right-angle intersections, due to a reduced number of conflicting points, shallower impact angles and, potentially, lower impact speeds. However, previous research on the safety benefits of roundabout designs is mainly based on assumptions on the vehicle speed, and little is known in terms of the actual speed at which vehicles travel through roundabouts. This study aims to identify whether common design features of roundabouts may affect the actual speed at which drivers in Australia approach, travel through, and leave roundabouts.
Empirical data collected during the Australian Naturalistic Driving Study (ANDS)  is used to create a profile of the speed at which drivers typically travel through roundabouts in Australia. Vehicle front radar data is used to screen those trips where the vehicle was deemed to be travelling in a free-flow condition (i.e., the vehicle speed was not directly constrained by a closely preceding vehicle, a crossing pedestrian or obstructions along the roadway and in proximity to the vehicle).
Initially, several roundabouts in various areas of Sydney are identified and classified based on their geometrical characteristics. For each trip that satisfied the free-flow criterion, the maximum vehicle speed through the intersection as well as the speed at the roundaboutís approach, centre (i.e., the position where the vehicle is closest to the centre of the roundabout, as shown in Figure 1) and exit were considered.
The speed data were then grouped according to the identified categories of roundabouts, speed limit and traffic volumes. The distributions of the speed profiles collected were then analysed to identify any differences related to specific geometric features of the roundabout, such as outer radius, height and size of the centre island, number of lanes, presence of obstacles in the centre island, lane width, vehicle entry angle, etc.
Preliminary results have been drawn from the analysis of an initial sample of 33 roundabouts. This initial assessment is limited to the analysis of the average speed measured at the centre of the intersection for vehicles travelling along a virtual straight direction (i.e., leaving the roundabout on the opposite leg from where they entered the intersection) and have been assessed against roundabout features such as size, number of lanes, and presence of a raised centre island. As an example, Figure 2 shows the average speed measured at the centre of each the 33 analysed intersections grouped by the corresponding roundabout outer radius. Based on a linear regression of these data points, the average speed tends to increase slightly for larger values of the roundabout radius. This behaviour is likely a consequence of the fact that the centrifugal acceleration experienced by a turning vehicle is inversely proportional to the radius of curvature and, thus, the same level of cornering acceleration and comfort obtained in smaller roundabouts can still be achieved at a higher speed for larger roundabouts.