Dozens of virtual ECUs simulate car-to-car communication
In the future, the aim for driver assistance systems is to also draw on sensor data from vehicles in their close proximity. To enable this, wireless ad-hoc networks must ensure reliable data exchange between constantly changing communication partners. The software of the vehicle ECUs involved plays a key role here. In order to realistically test these interactions, researchers at the Technical University of Munich are using ETAS virtual ECUs to model highly dynamic car-to-car networking.
Camera and radar sensors have a limited field of vision. They cannot detect what lies around bends or over the brows of hills, nor can they spot vehicles emerging from out-of-view side streets or the sudden braking of vehicles ahead. Wireless communication between vehicles is set to minimize these dangers.
In the future, vehicles will connect with each other via the automotive WLAN standard IEEE 802.11p to form vehicular ad-hoc networks (VANETs). This will let them exchange their current coordinates and sensor data. Using direct WLAN connections means data does not need to pass via a backend, thus minimizing latency and connection costs. Such interplay between all-round sensor systems and car-to-car communication would enable advanced driver assistance systems (ADAS) to offer even greater protection against accidents, optimize traffic flows, and – insofar as drivers want it – take control over the vehicle.
Testing based on models
Road tests in real traffic are not a suitable way to test this kind of communication between vehicle systems. On test tracks, the time, material, and cost involved in such testing would be excessive, and the tests would be scarcely reproducible. The solution is virtualization. Precisely, in order to explore the bases of VANETs, to sound out their potential right down to limit ranges, to recognize problems, and to move existing technical limits toward production readiness, virtual testing is essential – because errors here do not lead directly to accidents.
However, virtually testing the highly dynamic VANETs requires tools that can handle the complexity involved. This complexity is based on several factors:
- Oncoming and crossing vehicles often join the wireless ad-hoc networks for only a few seconds.
- Communication is bidirectional, with every vehicle involved both transmitting and receiving data that their IT systems are simultaneously generating and processing in highly dynamic processes.
- Interference with the wireless connection can occur at any time. Non-connected road users may also be present in the VANET zone. Both these situations require the systems to have high fault tolerance.
- Despite heterogeneous software and hardware in the vehicles, communication must be smooth and trouble-free – including at rush hour, when there are many, rapidly changing participants.
To be able to realistically model this variety in all its breadth, holistic strategies such as multi-resolution modeling (MRM) are needed. The VANET simulation/emulation should offer rapid switching between detailed views and system overviews as well as high flexibility as regards the number of vehicles integrated.
Up to 40 virtual ECUs in use
A team led by Prof. Alois Knoll and Manuel Schiller at the Chair for Robotics and Embedded Systems at the Technical University of Munich works with VANET emulations in which they realistically simulate data exchange in VANET zones ranging from 100 to 500 m wide. They want to determine how bidirectional communication and, above all, the ECU software executing the communication in the vehicles can be tested with regard to deployment in series production contexts.
At the heart of their modeling work is ETAS ISOLAR-EVE (ETAS Virtual ECU). This tool generates virtual ECUs that can be run on ordinary PCs. Up to 40 of these virtual ECUs execute ECU code during their emulations, which are carried out while the network behavior and interferences, the traffic flow, and the vehicle behavior are also simulated. “The goal is to deliver a realistic overall scenario,” explains Schiller, “so that we can integrate real and simulated code in our network emulations and test the code realistically.” One of the objectives is to evaluate how the ECU software responds to open and closed-loop control interventions addressed to them via VANET. Fallback routines in the event of an interrupted connection or increased latency are also to be investigated.
The key to this is ISOLAR-EVE, which allows the researchers to integrate dozens of virtual ECUs into their overall simulation simultaneously on a PC. In this way, they can test and analyze the system behavior of the highly dynamic VANETs at various scales. It is always possible to further scale the system by operating several computers in parallel. What is more, the virtual ECUs run production-ready basic and application software of variable maturity – and they do so in direct connection with ordinary network stacks. “We consciously decided not to run any preproduction application software on the virtual ECUs yet, because the point is specifically to develop a test and evaluation setup for embedded software in VANETs,” explains Schiller. Nonetheless, he adds, ISOLAR-EVE’s compliance with AUTOSAR ensures that application software can be used, provided it is conform to AUTOSAR.
Collaboration between the Technical University of Munich and ETAS
Soon after Schiller started the project in 2015, he came across ISOLAR-EVE while searching for solutions for the virtual testing of VANETs. “Although we could also have used other virtualization solutions,” he explains, “we actually needed a flexibly scalable solution based on AUTOSAR for the automotive environment.” The fact that ISOLAR-EVE permitted dozens of virtual ECUs to be operated synchronously on a single PC was ideal for testing embedded software in VANETs. “Its high performance enables us to study large-scale scenarios with many simulated network participants. The option to integrate preproduction basic software is essential for future deployments of the developed methodology in production contexts. On top of that, the time synchronization of the ECUs and the complete control over the ECUs using their APIs, are a great help in our complex test setup,” he says. And the same goes for the concrete support from ETAS experts, added Schiller, which also helped them to master the huge complexity.