同時定位與地圖構建（SLAM）系統分為傳統方法和基於機器學習的方法。傳統的 SLAM 系統採用幾何和概率模型在靜態環境中實現高精度，但在動態環境中面臨計算複雜性和適應性的挑戰。基於機器學習的 SLAM 系統利用深度學習，擅長處理非結構化數據和動態場景，但需要大量訓練數據，並且通常缺乏可解釋性。
我們的目標是通過結合深度學習的模型來增強傳統的 SLAM 系統，使傳統的
SLAM 系統更加強大和全面。本論文在 R3LIVE 框架下優化和加速了點雲渲染過程，並利用深度學習模型解決了因光達特性導致的建圖隙縫與漏洞。;Simultaneous Localization and Mapping (SLAM) systems are divided into traditional and machine learning-based methods. Traditional SLAM employs geometric and probabilistic models to achieve high precision in static environments but faces challenges with computational complexity and adaptability in dynamic environments.Machine learning-based SLAM, utilizing deep learning, excels in handling unstructured data and dynamic scenarios but requires substantial training data and often lacks interpretability.
We aim to enhance traditional SLAM systems by incorporating the advantages of deep learning model, making traditional SLAM systems more robust and comprehensive. In this paper, we optimize and accelerate the point cloud rendering process under the R3LIVE framework and use a deep learning model to solve the mapping gaps caused by the characteristics of LiDAR
