The P6 project within i-CAVE is responsible for the functional architecture and safety of autonomous and cooperative driving vehicles.
Existing prototype demonstrators of autonomous and cooperative driving vehicles are built on top of current car systems, adding various functional systems (e.g. sensing, localization, perception, and high intelligence systems), mostly by considering the underlying vehicular system as a black box. This enabled fast prototyping and demonstration, at the cost of risking a lack of perceivable overall architecture, which is undesirable from a software engineering viewpoint; it can make the software ineffective and hard to reuse.
Definition and analysis of software architecture enables early prediction of system’s qualities, which can decrease development cost and help avoid software defects that can be costly or even endanger lives. Ensuring software quality is thus critical and a real-time fault detection mechanism unavoidable, especially in the context of autonomous and cooperative driving vehicles.
Software systems are checked against quality attributes in a quality assessment stage of the software development life cycle. This checking is based on quality models, which traditionally do not scale up to autonomous and cooperative driving vehicles with their complex systems and intertwined interactions between systems and environments. Although the quality issues, especially safety, will be tackled in each i-CAVE project in isolation, the integrated quality aspects e.g. safety cases of the systems of each project need to be formulated and evaluated. Therefore, this project investigates automotive architecture, a safety-driven quality model, and means to monitor and ensure safety for autonomous and cooperative driving vehicles.
We are currently focusing on architectural models and quality standards, and on ensuring functional safety at runtime. Current quality standards for automotive software are governed by the mandatory adherence to ISO 26262. However, new developments in co-operative and autonomous driving outreach the boundaries of ISO 26262 and require new paradigms. In this project we also study new safety and quality concerns that can extend current models to include robustness of intelligent algorithms, verification of probabilistic systems and secure communication between vehicles. In order to ensure safety of autonomous and cooperative driving vehicles at runtime, we are currently investigating ways to derive online, real-time, safety mechanisms out of historical data automatically, i.e. to learn behaviors. The learned models can be used to provide context-aware runtime monitoring as a means to ensure functional safety at runtime.