【计算机科学】【2017.06】室内智能体定位中的传感器融合与深度学习

本文为美国罗切斯特理工学院（作者：Jacob F. Lauzon）的硕士论文，共124页。

随着科技化的时代到来，自主的、自动导航的智能体越来越受欢迎。从汽车到吸尘器，自动导航的应用范围非常广泛，涉及到生活的许多不同领域和方面。

随着对这些自主机器人的需求不断增加，对创新功能、健壮行为和低成本硬件的需求也在不断增加。一个需要不断改进的特定领域是定位问题，即智能体确定其在环境中位置的能力。无论智能体所处的环境是室内还是室外、密集还是稀疏、静态还是动态，智能体都必须能够知道其当前的位置。今天，存在许多不同的定位技术，每种技术都有其优缺点。尽管目前已经有许多不同的技术，但仍有改进的余地。本研究提出一种新的室内定位演算法，以较低的成本融合多个感测器的信息。受深度学习和粒子滤波的启发，设计了一个快速、健壮、精确的自主定位系统。研究结果表明，该系统具有实时性和健壮特性。

Autonomous, self-navigating agents have been rising inpopularity due to a push for a more technologically aided future. From cars tovacuum cleaners, the applications of self-navigating agents are vast and spanmany different fields and aspects of life.

As the demand for these autonomous robotic agents has beenincreasing, so has the demand for innovative features, robust behavior, andlower cost hardware. One particular area with a constant demand for improvementis localization, or an agent’sability to determine where it is located within its environment. Whether theagent’senvironment is primarily indoor or outdoor, dense or sparse, static or dynamic,an agent must be able to have knowledge of its location. Many differenttechniques exist today for localization, each having its strengths and weaknesses.Despite the abundance of different techniques, there is still room forimprovement. This research presents a novel indoor localization algorithm thatfuses data from multiple sensors for a relatively low cost. Inspired by recentinnovations in deep learning and particle filters, a fast, robust, and accurateautonomous localization system has been created. Results demonstrate that theproposed system is both real-time and robust against changing conditions withinthe environment.

1． 引言
2. 项目背景
3. 研究平台
4. 研究方法
5. 研究结果
6. 结论与展望
附录A 部分硬件设计图

更多精彩文章请关注公众号：