ABSTRACT

Current Self-Adaptive Systems (SASs) such as Autonomous Vehicles (AVs) are systems able to deal with highly complex contexts. However, due to the use of static feedback loops they are not able to respond to unanticipated situations such as sensor faults. Previously, we have proposed HAFLoop (Highly Adaptive Feedback control Loop), an architecture for adaptive loops in SASs. In this paper, we incorporate HAFLoop into an AV solution that leverages machine learning techniques to determine the best monitoring strategy at runtime. We have evaluated our solution using real vehicles. Evaluation results are promising and demonstrate the great potential of our proposal.

摘要

当前的诸如自适应车辆（AV）的自适应系统（SAS）是能够处理高度复杂的环境的系统。但是，由于使用了静态反馈回路，因此它们无法响应意外情况，例如传感器故障。以前，我们已经提出了HAFLoop（高度自适应反馈控制环），一种用于SAS中自适应环的体系结构。在本文中，我们将HAFLoop合并到一个AV解决方案中，该解决方案利用机器学习技术来确定运行时的最佳监视策略。我们已经使用实际车辆评估了我们的解决方案。评估结果令人鼓舞，并证明了我们提案的巨大潜力。