遠端操控自動駕駛車輛系統中支援向量機之研究

Abstract

遠端駕駛是一種嚴重依賴遠程操作員與車輛自動駕駛系統 之間可靠通信鏈來運作的系統。雖然變道和加速/減速等細節的操作（低級動作）由 ADS 執行，但路線規劃（高級指令）是通過使用從沿路的攝像頭或傳感器收集的信息來遠程完成的。在論文中，討論了支持向量機的理論，它在中有著核心的作用。作為一種基於機器學習的技術， 必須經過訓練才能用於執行分類任務。因此，我們將說明如何在中找到優化的支持超平面的推導，包括如何使用序列最小優化。

A tele-operated vehicle is a system relying heavily on a reliable communication link between a remote operator and an automated driving system (ADS) at a vehicle. While detailed maneuvers (low level actions) such as lane changes and accelerations/decelerations are performed by the ADS, route planning ( high level instructions) is remotely done by using gathered information from cameras or sensors along the road. In the thesis, the theory of the support vector machine (SVM) is discussed, which plays a central role in an ADS. As a machine-learning based technique, a SVM has to be trained before it can be used to do classification tasks. Therefore, the derivations of how an optimized supporting hyperplane in SVM can be found are illustrated, including the use of the sequential minimal optimization (SMO).