A Modular Five-Layered V-Shaped Architecture for Autonomous Vehicles

The real challenge in autonomous vehicles is to have them as robust to the unexpected edge conditions as human drivers under all possible scenarios with varying conditions even when some of the components are not behaving adequately. This requires a paradigm shift on the architecture design so that redundancy for safety-critical operations are utilized based on self-diagnostics as well as the (probabilistic) success measures based on the key performance indices. In this paper, we propose an inherently-safe and modular software architecture based on automotive standards and software design principles. The main advantages of the proposed architecture are i) a modular structure that enables multiple components to be hot-swapped during execution, ii) a layered-based design that enables quick validation of the requirements defined for each layer and therefore guarantee traceability, iii) inherent diagnostics capabilities that continuously feed a central control unit and iv) distinct responsibilities assigned to each separate module tiff abstraction.