以雲端運算為基礎之增強型同時定位與建圖

Abstract

FastSLAM演算法常常被用來解決同時定位與建圖問題。雖然FastSLAM2.0的運算效率比EKF-SLAM來的高，但是隨著地標數目增加的時候，FastSLAM2.0會因為需要多次比對量測資訊與粒子內存的地標資訊，而降低運算效率。因此，本論文提出一改良作法，稱之為「增強型同時定位與建圖演算法(ESLAM)」，避免只用里程計資訊預測機器人位置，也使用環境資訊更新機器人預測位置，並選擇與量測資訊相似性最高的地標資訊先更新機器人位置後，再更新地標位置。模擬結果顯示，我們所提出的演算法相較於FastSLAM2.0具有較高的運算效率，且具有較良好的定位與建圖準確度，而相較於CESLAM雖然犧牲了些許運算效率，但提升了準確度。由於SLAM演算法常需要複雜計算，使得執行效率低落，無法達成即時處理的目標。因此，我們提出一雲端運算架構，將計算密集的任務卸載至雲端運算平台，運用雲端的快速運算以提升演算法之效能，其作法係利用RPC傳輸協定搭配雲端平行化架構進行以雲端為基礎之增強型同時定位與建圖。實驗結果證明，本方法可以確保定位與建圖的準確度之外，並運用雲端運算提升同時定位與建圖之執行效率。

FastSLAM is currently the most common solution to SLAM problems. Although the processing speed of FastSLAM2.0 is already faster than the EKF-SLAM, it could become slower under the circumstances of too many landmarks existence, where comparison measurements needed to be taken many times and would lower the calculating effectiveness. Therefore, this thesis proposes an improved version, Enhanced SLAM, which avoids using the odometer information only but also include the sensor measurements to estimate the robot’s pose. We used the landmark information that has the largest likelihood to update the robot’s pose first and then update the landmarks’ location. Compared to the FastSLAM2.0, our algorithm improved both the accuracy and the efficiency. Compared to the CESLAM, we improved the accuracy of locating and mapping but sacrificed some calculating effectiveness. The calculation consumes too much time and thus fails to achieve the goal of instant processing, hence, we utilized the high-speed of the cloud computing based on the combination of RPC Transfer Protocol and cloud parallel system to process ESLAM. The experiment results showed that this solution we proposed can improve the accuracy as well as the effectiveness of locating and mapping.