This section describes the single steps of the unified scenario generation for advanced driver assistance systems. The scenario generation bases on a 4-level model. A scenario can be flexibly assembled on a selection of one or more of the four levels. E.g. static scenarios without dynamic elements can be defined on the basis of the first two levels. On every level of the scenario generation, the principle of the systematic test case generation is applied, which is described in detail in the Fig. 7.2 Efficient test cases
are the intersection of test cases, which are non-redundant,
representative, unified, and reproducible
second part of this sub-chapter. Figure 7.3 shows the structure of the unified scenario generation on the basis of the 4-level model, which is described in the following.
7.4.1 Level 1: Road Network
On the first level of the unified scenario generation, the road network has to be defined. Therefore, the geometry and topology of the roads have to be specified. For this, the basic elements for road design straight lines, curves, and clothoids are used. Straight lines are described only by a length. Curves are defined by a constant radius or curvature, which is unequal to zero, and a length. To combine straight lines with curves, clothoids are used. Clothoids are defined by a start curvature, an end curvature, and a length. Therefore, clothoids represent transition elements between straight lanes and curves or curves with different curvatures.
The geometries of the road network are extracted from current standards and guidelines for the creation of highways, rural roads, and urban roads (FGSV1980, 1995, 1996, 2006, 2008). These geometric parameters can be, for example, the allowed minimal and maximal curvature and/or lateral and longitudinal elevation profiles. Table7.1shows some exemplary values from the construction guidelines of a highway in Germany.
If the road cannot be described with these three basic elements, it is also possible to define different splines to describe the road. Therefore, it is also possible to create and define non-standard roads.
After defining the geometries of the road, the topology elements of the road are described. Therefore, the number of lanes has to be defined, along with their width Fig. 7.3 The 4-level model
for a unified scenario generation
and markings. The lane markings have to be specified by a width, color, and also a style, like solid or broken. Additionally, the state of the road surface has to be specified, like wetness, dirt, or the abrasion of the road. In the standards and guidelines, different typical cross sections are available for the road topology.
These describe the range of allowed lane widths in relation to the road type, for instance. Table7.2gives an overview of some typical cross sections in Germany.
Figure7.4shows the route of an example road, consisting of lines, curves, and clothoids with two typical cross sections for a highway. The figure also shows a Table 7.1 Overview of design parameters of different highways in Germany (FGSV2008)
EKA 1A EKA 1b EKA 2 EKA 3
Road function Highway Over regional
highway
Highway similar road
Urban highway Maximal length of a straight
line [m]
2000 – – –
Minimal curve radius [m] 900 720 470 280
Minimal clothoid parameter
Aẳ ffiffiffiffiffiffiffiffiffiffi
RL p [m]
300 240 160 90
EKA in German: “Entwurfsklasse” (draft class) RRadius,LLength of the element
Table 7.2 Overview of typical cross sections in Germany (FGSV2008) Typical
cross section
Border [m]
Hard shoulder [m]
Margin strip [m]
Driving lane [m] Hard
shoulder [m]
Central reserve First Second Third Fourth [m]
RQ25 1.5 2 0.5 3.5 3.25 0.5 2.5
RQ28 1.5 2.5 0.75 3.5 3.5 0.75 3.0
RQ31 1.5 3 0.75 3.75 3.75 0.75 4.0
RQ36 1.5 2.5 0.5 3.75 3.5 3.5 0.75 4.0
RQ43.5 1.5 2.5 0.5 3.75 3.75 3.5 3.5 0.75 4.0
RQ: German: “Regelquerschnitt” (cross section)
Fig. 7.4 An exemplary basic road, consisting of a straight line, clothoid, curve, and a transition between the typical cross section RQ36 and RQ31 on the straight line
transition between the two different typical cross sections, which is also defined in the standards and guidelines.
The basic roads can also be combined to intersections. Therefore, the standards and guidelines for the creation of urban roads are used, where default intersection topologies are defined (FGSV2006). To describe an intersection, the number of incoming and outgoing roads, and the connecting lanes between these roads have to be defined. Furthermore, the number of lanes for left and right turning has to be specified, along with their lane markings.
7.4.2 Level 2: Adaption of the Road for Special Situations
On the second level, situation-dependent adaptions, which are required for special applications or advanced driver assistance systems with special requirements, are added to the basic road.
These adaptions are for example the surroundings of the road or static ele- ments on the road. If it is important for the test that the basic roads or inter- sections are in an urban environment, surrounding elements like houses, street lights, and traffic lights have to be added to the basic roads to generate an urban environment.
Examples for two generated intersections in an urban environment are illustrated in Fig.7.5.
Possible adaptions are additional roads, which are required (e.g. to define road works). These adaptions are required for a test of the constriction assist, which is developed in the research project UR:BAN (Scholl2015). The road works are also generated according to the current standards and guidelines, like the RSA in Germany (FGSV 2009). Figure 7.6illustrates a roadwork zone according to the guidelines.
The left lane of the road is closed through a lateral barrier. The barrier is specified by a length and a lateral offset. The ratio between the length and the offset is defined for highways, rural roads, and urban roads. According to the offset of the lateral barrier, the right lane has a remaining width of the delta of the road width with two lanes and the lateral offset. The standards and guidelines provide a minimal width of the right lane in such roadwork zones. For protecting the roadwork zone, additional lane markings with different colors and static elements, like traffic cones, traffic beacons, or concrete barriers are added to the roadwork zone. The distance of the roadwork elements in lateral and longitudinal direction can be varied. The basic road in the first level of the 4-level model is under the adapted road, which means that its geometry is still existent and visible, but may not be used in the construction zone. Thus, the lane markings of the basic road can be used as inconsistent information for systems under test.
Fig. 7.5 Examples of two intersections defined on the first and second level of the model
7.4.3 Level 3: Dynamic Elements
After defining the basic track and optional adaptions, the quantity and the behaviour of the dynamic elements are defined on the third level. Figure7.7illustrates the integration of the first two levels of the 4-level model in the context of a test scenario.
The dynamic elements can be defined on various abstractions levels. On the top level, it should be possible to define dynamic elements with information about the traffic flow. Thereby, it is possible to generate random traffic situations on a defined road. For the test of advanced driver assistance systems, this abstraction layer is useful to generate random scenarios without any detailed information about test cases. The traffic can be described by little information, like the number of other road-users. However, on this level it is not possible to create reproducible traffic scenarios. To generate reproducible scenarios, a lower abstraction level of the definition of dynamic elements is also provided in the unified scenario generation.
Fig. 7.6 A roadwork zone according to the current guideline on an urban road
Fig. 7.7 Structure of the control for the dynamic elements and the first and second level of the unified scenario generation in the context of the scenario
On the lower abstraction level, it is possible to define the traffic on a higher specific level, like following a vehicle, parallel driving vehicles, or lane changing vehicles, based on different parameters.
To generate reproducible scenarios, it is important that the scene as perceived by the ego-vehicle is still the same. This means that the other vehicles in the scenario have to react to the actions of the ego-vehicle. For example, if the ego-vehicle is driving with different velocities in the scenario, the other vehicles have to adapt their own behavior to the velocity of the ego-vehicle. Only if this requirement is fulfilled, reproducible scenarios can be generated and efficient test cases with dynamic elements can be executed.
To implement this in the simulation, the scenario generation uses a special control of the dynamic elements. Figure7.7illustrates the structure of the control for the dynamic elements in the context of the scenario.
On the right side of the figure, the stationary elements of the scenario are shown, which are also called scenery (Geyer et al.2014). These elements are described on the first two levels of the unified scenario generation. The left side of the figure shows the dynamic elements, which are the ego-vehicle or the other driving vehicles for example. To control the dynamic elements, different maneuvers are defined in the scenario, which are applied to the dynamic elements by a scenario scheduler. The scenario scheduler has also the option to control the environmental conditions, which are described in the next sub-chapter. With the information about the ego-vehicle, the scheduler has the opportunity to adjust the maneuver to the behavior of the ego-vehicle to generate reproducible scenarios from the viewpoint of the ego-vehicle.
Every maneuver has also an internal structure, which is illustrated in Fig.7.8.
A maneuver has one or more start conditions. When one of these conditions is fulfilled, the scheduler of the scenario starts the maneuver. For instance, the start conditions are a relative distance or relative velocity to the ego-vehicle, or a specific location on the road, which can be a curve with a defined curvature. Thereby, it is possible to generate special situations at different locations, which can be critical situations for the system under test.
Fig. 7.8 Structure of a maneuver with events and actions
Every maneuver has its own scheduler, which has the task to control different events. The events are required to realize the desired behavior of the dynamic element. Possible events are the appearance of a dynamic element, the change of the velocity, or a lane change. Every event has also one or more start conditions. The maneuver scheduler uses these conditions to start the events to execute the maneu- ver successfully. The events can trigger one or more actions to fulfill the event.
The executer can execute the actions on different levels. If the simulation offers the direct execution of an action, the executer will use this option. In Fig.7.8, this is demonstrated through the blocksimulation control. If it is possible to command a lane change to a dynamic vehicle through an interface of the simulation software, the executer uses this interface to command the lane change for instance.
If such an interface does not exist, the executer has the option to command the action to dynamic elements on a lower level. This could be the driver commands to control the dynamic vehicle. For the example of the lane change, this means that the maneuver is more configurable and a trajectory can be defined for the lane changing vehicle. Thereby, the distance between the dynamic element and the ego-vehicle during the lane change can be chosen. Thereby, different kinds of lane change maneuvers for the dynamic vehicles can be simulated.
If the action cannot be executed with driver commands, the executer has also the opportunity to realize the action by setting the state of the dynamic element in every frame of the simulation. Thereby, it is possible to execute actions, which are not covered by the vehicle dynamics model of the simulation environment, a spinning vehicle for example.
With this control of the dynamic elements, the unified scenario generation has the possibility to generate static as well as dynamic scenarios with different elements. Furthermore, it is possible to generate reproducible scenarios from the perspective of the ego-vehicle.
7.4.4 Level 4: Environmental Conditions
On the first three levels of the model, the road, possible adaptions to the road, as well as the behavior of dynamic elements are described. On the fourth level of the unified scenario generation, it is possible to vary the environmental conditions, which are the daytime and the weather in the scenario for example. Possible states for the daytime are dawn, day, dusk, and night. Therefore, advanced driver assistance systems can be tested under different lighting conditions, which is mainly relevant for vision-based systems. Another option is to vary the weather conditions. The following states are possible:sunny,cloudy,rainy, andsnowy.
These weather states are also influencing the lighting conditions and the road conditions.
7.4.5 Summary of the 4-Level Model of the Scenario Generation for Advanced Driver Assistance Systems
The last section described the structure of the unified scenario generation based on the 4-level model. On the four levels, the road geometry and topology, possible adaption of the road for special situations, the dynamic elements, and environmental condi- tions can be defined. Figure7.9illustrates an example of a generated scenario based on the 4-level model for a constriction assist. The basic road is described by a straight line and a following curve. The road has four driving lanes, each with a width of 3.5 m. The second level describes an adaption by a roadwork zone. Two dynamic vehicles are defined, which are driving in front of the ego-vehicle (white vehicle in Fig.7.9) through the road work zone. On the level of the environmental conditions, the weather is defined as rainy and the daytime as day.
On every level an amount of parameters and their values can be varied. A single scenario with five impact parameters and six discrete values for each parameter generates 65ẳ7776 test cases. By adding new parameters, the number of test cases increases exponentially with each parameter added and additionally with each new discrete parameter value. Thus, it is not possible to execute all test cases, if many new parameters are added due to time and economic limitations, especially when all parameter combinations have to be tested. This is the reason why new methods have to be found to reduce the number of test cases. This problem is already known and addressed in the research area of software-testing. One option is the usage of combinatorial algorithms. These algorithms reduce the number of test cases on the basis of combinatorial considerations. Kuhn et al. (2004) show in a case study that a large number of faults can be found with combinatorial testing. In this case
Fig. 7.9 Example of a generated scenario on the basis of the 4-level model for the constriction assist
study, different kinds of systems are tested. Thereby, all systems have to be tested by a large number of input parameters and parameter values. The testing of advanced driver assistance systems has similar requirements to the test method.
For this reason, the efficient, systematic test case generation also uses combinatorial algorithms to reduce the number of test cases. The single steps for reducing the number of test cases are explained in the following part.