搭載光達之移動機器人的自主探索以利於大規模室內環境高品質重建

Abstract

自動化移動機器人在搜索救援、工業檢查和場景重建中扮演重要的角色，通過運用探索規劃演算法在未知環境中導航和收集資訊。高質量的3D重建對於自動駕駛、AR/VR 和智慧城市等應用至關重要。傳統的方法以手持感測器掃描往往會導致重建模型不完整和對齊錯誤。儘管自動化探索可以改善這些結果，但通常會犧牲品質換取速度。完整性和 RMSE 這樣的指標常用於重建品質檢測，但很難用於自動化探索中實時檢測，因此可能會導致重要細節的丟失。 機械式光達因其在範圍和不受光照影響方面的優勢，時常被廣泛採用於自動化探索，但其成本高且點雲數據稀疏。另一方面，固態光達提供了更精細的細節，但覆蓋範圍有限。另外，在大型環境中生成邊界、視點和路徑可能會超出計算資源的承受能力，導致降低規劃頻率和響應速度。因此，在大規模環境中進行自主探索期間的實時品質評估仍然是一個重要挑戰。 為了解決上述問題，我們提出的系統通過自動化探索，利用配備二自由度旋轉平台的固態激光雷達的移動機器人，進行大規模室內環境的高質量三維重建。此系統不僅控制移動機器人探索未知地點和重建質量不足的區域，還具備高頻率、實時的探索路徑規劃。最後本文包括各種模擬和真實世界實驗，以驗證我們系統的有效性。

Autonomous mobile robots play a vital role in search and rescue, industrial inspection, and scene reconstruction by using exploration planning algorithms to navigate and gather information in unknown environments. High-quality 3D reconstruction is essential for applications such as autonomous driving, AR/VR, and smart cities. Traditional methods using handheld sensors often result in incomplete and misaligned models. While autonomous exploration can improve these results, it often sacrifices quality for speed. Metrics like completeness and RMSE are often used for quality evaluation; however, they are inadequate for real-time assessment of high-quality 3D reconstruction, risking the loss of important details. LiDAR sensors, especially mechanical LiDARs, are often used for autonomous exploration because of their range and independence from lighting conditions, despite their cost and data sparsity. On the other hand, solid-state LiDARs offer finer detail but have limited coverage. Besides, generating frontiers, viewpoints, and paths in large environments can overwhelm computational resources, reducing planning frequency and responsiveness. Consequently, real-time quality evaluation during autonomous exploration in large-scale environments remains a significant challenge. To address the aforementioned issues, we propose a system for high-quality 3D reconstruction of large-scale indoor environments through autonomous exploration using a mobile robot equipped with a solid-state LiDAR on a 2-degree-of-freedom gimbal. This system not only controls the mobile robot to visit unknown places and areas with insufficient reconstruction quality but also facilitates high-frequency, real-time exploration path planning. This thesis concludes with various simulations and real-world experiments to validate the effectiveness of our system.