實時語義同步定位與地圖建構融合系統開發與在動態環境中移動機器人的應用

Abstract

保持機器人建圖和定位的高精度是極具挑戰性的工作，尤其是在複雜多變的環境中，如動態環境、缺少明顯特徵的室內環境以及不平整的室外環境。傳統的單感測器SLAM（同步建圖與定位）方法面臨諸多困難，需要採用更為先進和多樣化的技術來應對這些挑戰。 本論文致力於提升機器人SLAM的強健性、穩定性和準確性。首先，我們建立了搭載深度相機和雷射雷達的智慧型機器人平臺，並分別應用於防疫消毒和協助失智症老人的照護。其次，針對動態環境，我們設計並實現了支援多目標跟蹤的動態環境語義SLAM系統，該系統能夠在未知環境中進行機器人狀態、環境特徵和多目標狀態的聯合估計。隨後，我們對室內外視覺SLAM和雷射雷達SLAM的定位精度、建圖效果和性能進行了綜合評估。最後，我們提出了基於相機與雷射雷達融合的SLAM方法，充分利用兩種儀器的互補優勢。 這些演算法均經過公開資料集或實際場景的測試，結果表明，本論文提出的SLAM系統具有顯著的強健性、穩定性和準確性。

Maintaining high accuracy in robotic mapping and localization is a highly challenging task, especially in complex and dynamic environments, such as those with moving objects, featureless indoor spaces, and uneven outdoor terrains. Traditional single-sensor SLAM (Simultaneous Localization and Mapping) methods face numerous difficulties and require more advanced and diverse technologies to address these challenges. This research aims to enhance the robustness, stability, and accuracy of SLAM by developing an intelligent robotic platform with depth cameras and 3D LiDAR, and by applying it to both pandemic disinfection and dementia elderly care. The platform implements a specially designed dynamic-environment semantic SLAM system which supports multi-object tracking and concurrent estimation of environmental features, multiple object states in unknown environments, and the robot’s own state. In addition, a comprehensive evaluation is made of the localization accuracy, mapping quality, and performance of indoor and outdoor visual SLAM and LiDAR SLAM. A novel SLAM method is then proposed based on the fusion of camera and LiDAR data, leveraging the complementary advantages of both sensors. These algorithms are tested both on public datasets and in real-world scenarios. The results demonstrate that this new SLAM system exhibits significant robustness, stability, and accuracy.