Safe and Efficient Runtime Resource Management in Heterogeneous Systems for Automated Driving

摘要

In this paper, we present a novel runtime resource management approach that obeys automotive safety constraints. We specifically target emerging heterogeneous embedded plat-forms which promise potential to ease the ever-growing gap between demanded processing power and feasible efficient em-bedded realization of modern assistance systems by allowing both, hardware and software implementations of automotive driver assistance tasks. Our approach proposes runtime concepts that are mandatory for efficiently utilizing those heterogeneous architectures, specifically taking into account hard automotive safety requirements. Our dynamic management is complemented by a fail-operational scheme that ensures permanent safe vehicle operation. For evaluation, we implement a modern heterogeneous embedded platform as both, an in-vehicle prototype platform using a near-series CMOS sensor and as hardware-in-the-loop prototype, concurrently executing two complex assistance applications, a traffic light recognition and a traffic sign recognition, demonstrating the feasibility of our approach.