應用於辦公室機器人服務之基於異質特徵的語意地圖建置

Abstract

為了讓機器人能在日常環境中執行各種任務，我們需要提高機器人對環境的理解能力。在本論文中，我們提出使用彩色與深度相機(RGB-D camera)建立語意地圖的演算法，在建立地圖的同時進行物體辨識，並建立物體之間的關係。藉由讓機器人認識環境中的各種物體，機器人可以執行更多高階行為，這些知識也能讓機器人聽懂人類的語音命令進行更自然的人機互動。另外，藉由整合不同種類的特徵資訊，機器人可以在多種環境中建立地圖、導航，及執行任務，不會受限於環境的種類。 在我們的系統中，藉由整合彩色與深度相機及雷射測距儀的資訊，機器人可以進行門口辨識、在擁擠的房間中計算出可移動的區域，並根據簡單的語意命令來執行尋物並取物的任務。

Semantic mapping has become a popular and essential research topic for intelligent robots nowadays. In this thesis, we present a semantic mapping framework using heterogeneous features, and its realization in office environment. Instead of considering only dataset-based or dataset-free semantic information like previous researches, our system incorporates heterogeneous features so that it is applicable to different environments. Besides its flexibility, object recognition is performed during the online mapping process and relations representing the inter-relationships among objects in the environment are labeled and updated through each new measurement. In addition, by defining semantic commands, our robot can understand semantic meaning and perform high-level tasks. Finally, a mechanism of object finding task is proposed in office environment. By combining the information from Laser Range Finder and RGB-D camera, our robot can detect doors and navigate in crowded environment, which are quite typical in an office building. To validate our framework, several experiments are conducted, and it shows that our system can indeed build the semantic map and carry out some high-level tasks in office environment successfully.