機器人在感測範圍受限情形下完成室內環境定位與製圖

Abstract

本論文提出一個在有人室內環境中，基於感知能力受限機器人所提供的彩度與深度照相機與短程雷射來進行環境的地圖建製與即時定位。由於感知能力受限機器人目前的移動能力會受周圍的環境影響無法產生一個可重復使用的環境參考地圖，而身為一個室內服務型機器人，它必須要具備能在環境中定位與移動的能力。然而機器人本身的設計原因，它並不具備一般常使用於定位的雷射測距儀來做長距離、密集且精準的資料蒐集，取而代之的是是稀疏、短距離且容易被干擾的資料。於是我們需要其他的感測裝置來做輔助，也就是位於頭部的深度與彩度照相機來彌補短程雷射對環境偵測的不足。在這篇論文中，我們會先遠端取得以短程雷射所偵測到離機器人的資料進行疊代最近點演算法以修正機器人自身測距的誤差，同時以深度照相機所取得的三維資料來描述所感測到的環境資料代替遠程雷射的不足。

This thesis proposes a kind of simultaneous localization and mapping framework based on data of color camera, depth camera and short range infrared finder from the limited sensing robot in an indoor environment with human. As an indoor social robot, it needs to be equipped abilities of localization and navigation in the environment with human. However, due to the design of the limited sensing robot, it does not have a regular laser range finder which is commonly used to obtain dense, long-range and precise data in SLAM. Instead of using regular laser range finder, the limited sensing robot uses short range infrared finder to estimate the distance. The sample data of short range infrared finder are sparse, short-range and easier to be disturbed than the sample data of long range laser. To compensate the lack of information obtained from short range infrared finder, we use other sensors, RGB and depth camera, on the robot to support the perception from the environment. In this work, we will take the estimation obtained from short range infrared finder to do the Iterative Closet Point (ICP) algorithm for fixing the error of robot odometry. Then, the 3D data from the depth camera are used to make a map describing the physical situation of environment. When the robot makes the reference map, there may be some un-fixed object in the environment and we have to filter out those 3D data belonging human. Considering the limited sensing robot is a social robot and it inevitably avoids interacting with human, it should not only locate itself but also locate human in the environment.