Towards a 3-tier Architecture for Connected Vehicles Christian Prehofer * , Konstantin Schorp * , Stefan Kugele † , Daniel Clarke * , Markus Duchon * * fortiss GmbH, {prehofer, schorp, clarke, duchon}@fortiss.org † Institut f¨ ur Informatik, Technische Universit¨ at M¨ unchen, kugele@in.tum.de Abstract—In this paper, we consider scenarios, requirements, and architectures for future connected vehicles. Regarding sce- narios, we discuss new features like autonomous driving, which has very strong requirements on sensor data fusion, real-time processing, and decision making. We also address a major trend in automotive architectures, which is the aggregation of computing inside the vehicle. Furthermore, we consider connected vehicles, where functionality can be taken over by the cloud. This leads to new challenges with respect to allocation of real-time data processing and control as well as the distribution of the strategy layer across the tiers. To address the challenges of these novel scenarios, we propose a 3-tier architecture for future connected vehicles. I. I NTRODUCTION The goal of this paper is to consider scenarios, require- ments, and architectures for future connected vehicles. Regard- ing scenarios, we consider new features like autonomous driv- ing, which has very strong requirements on sensor data fusion, real-time processing, and decision making. Furthermore, we consider connected vehicles, where functionality can be taken by the cloud. Based on this, we derive high-level requirements and discuss the distribution of computing and control in such scenarios. We also address a major trend in automotive architectures, which is the aggregation of computing inside the vehicle. Putting this together, we propose a 3-tier architecture for future connected vehicles. II. SCENARIOS AND REQUIREMENTS In the following, we briefly discuss key scenarios for future vehicles and derive requirements for the overall architecture. 1) Autonomous Driving: In the last decades, technical evo- lution of sensing and processing technologies have contributed to a significant reduction of the architectural footprint with respect to size, weight, cost, and power. The subsequent pro- liferation of these technologies led to a number of significant possibilities within the automotive domain, particularly in the application fields of advanced driver assistance systems (ADAS) and autonomous driving (AD). These possibilities include the replacement of tightly calibrated and expensive hardware components with low cost hardware and effective software processing [1]. Improved sensing and processing capabilities are being exploited in ADAS and AD applications, where the driver is becoming increasingly relieved of the vehicle’s control [2]. Currently, autonomous features are limited to a subset of specific automotive functions. However, a potential time-line for the development of fully autonomous cars is given in [3]. In particular relevant is multi-sensor data fusion, which aims at combining sensory information from sources with disparate characteristics (either spectral or spatial) in order to provide an improved estimate of the state of an object. In order to support a multi-sensor data fusion system, an effective archi- tecture is required which provides both the communication and processing capabilities. Hence, future autonomous driving enabled systems must have high processing capacity along with high communication bandwidth. In summary, at a minimum, the general requirements are: (i) a comprehensive suite of sensors, (ii) multi-sensor data fusion architectures, (iii) algorithms capable of providing high precision situational awareness (both for the machine and the driver), and (iv) a centralised ICT architecture to support the inter-connected information transfer and processing around the vehicle (and with appropriate entities in infrastructure). 2) Connected Vehicles and Cloud-based Control: In the following, we consider the new possibilities of cloud-based services for external control and guidance of vehicles. This is different from today’s backend connections for instance for traffic information or hotel booking. A simple example would be a remotely operated cruise control, which automatically adapts the speed to the speed limit (following some user preferences). In this case, the car adapts automatically to slower or faster speed limits, but the driver has to make sure that this is actually possible. In a weaker form, the system may only slow down to lower speed limits, not automatically speeding up in case of a higher speed limit. Remotely guided autonomous driving is another important example here. Due to limited latency and reliability of the communication medium, the car has to support autonomous driving up to a level where it is at least capable of resolving critical situations such as emergency breaking without driver intervention. Local control components have to perform low level control and driving tasks, while the strategy is determined by the remote controller in the cloud. Another example, based on autonomous driving functions, is platooning or “convoy driving” (see e.g. [4]), where vehicles are controlled by some external service. This can be cloud-controlled, too. The main requirements from this trend towards external cloud-based control are as follows: (i) We need suitable network connection, which is sufficiently fast and reliable. (ii) We need to separate a higher strategy or control layer from the local execution in the vehicle. III. A 3- TIER ARCHITECTURE FOR CONNECTED VEHICLES In the following, we discuss new architecture concepts for future connected vehicles based on the above. There is currently considerable research on aggregating computing and control centrally in a vehicle, e.g. [3]. However, we need to address the computational needs of both sensor data processing