Science of Sensor System Software

传感器无处不在，有助于实时决策和执行，并为政策选择提供信息。但从传感器数据中提取信息远不是一帆风顺的：传感器噪音大，容易去校准，可能会被放错位置、移动、损坏，而且随着时间的推移通常会降级。面对无处不在的不确定性，我们对编程问题知之甚少，但传感器驱动的系统本质上给设计师带来了无法通过工程设计消除的不确定性。此外，不确定性是一种多层次的现象，在这种现象中，部署中的错误可能会传播到位置不正确的读数，然后传播到糟糕的决定；当暴露出不确定性时，系统分层就会崩溃。我们如何才能确保传感器系统在一系列动态环境中做我们想要做的事情，我们如何才能使一个系统“更智能”？目前我们无法回答这些问题，因为我们缺少一门关于传感器系统软件的科学。我们将发展缺失的科学，使我们能够针对现实世界系统中固有的不确定性进行工程设计。我们将为开发和部署可验证的、可靠的、自主的传感器系统提供新的原则和技术，这些系统在不确定的、多个和多尺度的环境中运行。这项科学将由最终用户和实验应用程序来驱动和验证。

Sensors are everywhere, facilitating real-time decision making and actuation, and informing policy choices. But extracting information from sensor data is far from straightforward: sensors are noisy, prone to decalibrate, and may be misplaced, moved, compromised, and generally degraded over time. We understand very little about the issues of programming in the face of pervasive uncertainty, yet sensor-driven systems essentially present the designer with uncertainty that cannot be engineered away. Moreover uncertainty is a multi-level phenomenon in which errors in deployment can propagate through to incorrectly-positioned readings and then to poor decisions; system layering breaks down when exposed touncertainty. How can we be assured a sensor system does what we intend, in a range of dynamic environments, and how can we make a system ``smarter'' ? Currently we cannot answer these questions because we are missing a science of sensor system software. We will develop the missing science that will allow us to engineer for the uncertainty inherent in real-world systems. We will deliver new principles and techniques for the development and deployment of verifiable, reliable, autonomous sensor systems that operate in uncertain, multiple and multi-scale environments. The science will be driven and validated by end-user and experimental applications.