Simulation Framework for Executing Component and …

1 Simulation Framework for Executing Componentand Connector Models of Self-Driving VehiclesFilippo Grazioli, Evgeny Kusmenko, Alexander Roth, Bernhard Rumpe, Michael von WencksternSoftware Engineering, RWTH Aachen University, GermanyAbstract Software for self-driving vehicles requires intensivetesting to avoid fatal accidents and to allow correct operationin real-world environments. Simulation frameworks allow toimitate the behaviour of complex systems such as autonomousvehicles using simplified models of the real world. Hence, theyare important tools allowing to extend Component and functionaltests to address interconnections between sensors, actuators,and controllers in virtual and predefined environments. Existingsimulators can be separated into high-level and low-level are designed for very specific scenarios and are not suitablefor addressing all driving situations.

2 While high-level simulatorsare suitable for mastering large testing environments such ascities, they lack fine-grained Simulation capabilities, , turningof wheels. In contrast, low-level simulators provide a high level ofdetail with realistic motion profiles. This is usually only possiblein small testing environments. In this paper, we present anapproach that combines the benefits of both high-level and low-level simulators to execute Component and connector and traffic engineers can choose the most suitable level ofdetail for their application and integrate real-world environmentdata from OpenStreetMap. Moreover, the simulator allows foradaptations and extensions of the physical vehicle configurationincluding new sensors, actuators and control systems.

3 Anotherfeature of our simulator is its automated testing support and itsability to visualize 3D simulations in a Terms Simulation Framework , self-driving vehicles, Component and connector models, executable modelingI. INTRODUCTIONA utonomous vehicles are complex software and hardwaresystems providing a wide spectrum of safety-relevant functions[1] which have to be tested and validated extensively beforethey can be made available for series production. Real worldtests are not only dangerous but also time consuming andcost-intensive and therefore unfeasible in early developmentstages. This leads to the importance of Simulation frameworksfor self-driving vehicles enabling safe and efficient testing inpredefined , a homogeneous and versatile solution would cover alarge variety of scenarios starting from fine grained sensor andactuator Simulation through to high level traffic scenarios.

4 Asa consequence such a versatile simulator would need to meeta broad set of requirements. First of all, it should combine thefeatures of high-level and low-level simulators. In other words,it should be able to simulate large-scale environments but, atthe same time, guarantee a high level of detail. Furthermore,it should allow the detailed Simulation of real-world situationssuch as city areas. Moreover, it needs to support an easyintegration and comparison of different sensors, actuators andcontrollers. In addition, automated regression testing is of greatimportance to enable agile development. Meeting all theserequirements represents the challenge that we want to tacklein this significant number of Simulation frameworks alreadyexists. In Section III, we compare the most popular andpowerful ones, such as Gazebo, ptv vissim , SUMO, and Car-Sim.

5 Each of the investigated Simulation frameworks exhibitsits particular benefits and drawbacks. For example, Gazebo,which is a highly versatile open-source Simulation tool fea-turing multiple physics engines, does not support environmentdata imports from OpenStreetMap. Conversely, PTV Vissimcan import OpenStreetMap data but does not have a physicsengine and is relatively expensive. To our knowledge nosimulation Framework supports native execution of Component & Connector (C&C) models (Simulink, Modelica, LabView,EmbeddedMontiArc, and SysML are popular C&C representa-tives) in realistic city environments. In fact, C&C modeling ispossible in MathWorks Simulink with Animation3D. However,no realistic city environment is provided by this , most simulators do not explicitly address automatedtesting though further work is still needed, the simulationframework introduced in this paper aims to combine all desir-able features of the considered frameworks and to simultane-ously eliminate their drawbacks.

6 We propose a MontiCAR [13] Simulation Framework named MontiSim that meets the pre-viously highlighted requirements, thereby supporting model-based software engineering (MBSE), test driven development(TDD), evolution, and execution of autonomous vehicle func-tions. Our Framework has its own extensible physics engineand is compatible with OpenStreetMap enabling a realisticsimulation of large-scale environments such as cities. At thesame time, highly-detailed low-level simulations of vehicledynamics and accidents are possible. A further feature is thecontinuous integration and regression testing support allowingto perform unit, integration and acceptance paper is structured as follows: First, we introduce a run-ning example and requirements on the simulator in Section IIin order to demonstrate the capabilities of the developed frame-work.

7 Second, we compare existing Simulation frameworkswith our proposed solution in Section III. Third, in Section IVwe present our first contribution, namely, themodular archi-tecture of our Simulation Framework allowing to exchangeand adapt C&C controllers, sensors and actuators invehicle models as well as maps and environment datain Simulation models. Our second contribution presented inSection V areautomatically executable testing concepts forC&C models such as regression tests with and without theenvironment. Finally, Section VI concludes this RUNNINGEXAMPLE ANDREQUIREMENTSB ased on previous research projects with industry partners[21], [19], [3], [4] and three labs on autonomous vehiclemodeling, we identified a precise set of requirements for asimulation Framework :(R1)Import and reuse of existing realworld environment data.

8 (R2)Capability to simulate large-scale everyday scenarios, , different traffic densities, lightand weather conditions.(R3)Support for realistic and exten-sible car models with sensors, controllers and actuators.(R4)Multi-platform and portable devices support.(R5)Automatedsupport for continuous integration and regression testing.(R6)Simulator should contain a physics engine.(R7)3D visualiza-tion for demonstration purposes. These requirements must befulfilled by a versatile autonomous driving simulator in orderto execute and validate previously developed C&C the remainder of this section, the usage workflow ofour proposed Simulation Framework is demonstrated based ona small realistic example, where two cars follow a straightstreet; videos showing our Simulation Framework in action areavailable at: , , and (1) First, the map of interest is downloaded ; based on this information the simulatoraccurately recreates streets, buildings and traffic signs.

9 (2)Next, a Simulation model as shown in Figure 1 is contains an environment description as well as simula-tion parameters such as location, time, resolution, weatherconditions, and available vehicles. (3) New vehicle modelscan be created as shown in Figure 2 if needed. Here, onespecific car is assembled by assigning its physical and visualproperties as well as its controlling unit, sensors and each input port of the controller, our Framework deliversthree default virtual sensors with different noise models. (4)The logic of the self-driving vehicle is modeled as a C&Ccontroller mapping the sensor inputs to actuator , we use the MontiCAR C&C modeling languagefamily together with its embedded math expression languagefor behavior definitions [13].

10 Figure 4 depicts a simple lanekeeping control system for straight streets (road curvature =0 ) inspired by [15]; due to better readability, the body oftheLaneKeepingControlleris represented graphicallyin Figure 5 instead of showing the textual syntax. The inputport of the lane keeping control systemd[4]represents adistance array to the road markers (see Figure 5 for a detailedexplanation); the output portsrepresents the desired steeringangle of the wheels. The controller contains a simple cor-rection Component to automatically adapt steering errors dueto imperfect sensor inputs or actuator delays (see differenceinMeasured car state (at t=1s)andCalculated car state fort=1s (at t=0s)in Figure 5). (5) Finally, new sensors andactuator models can be created.