基於單眼視覺之無人飛行器自動環境探索與可行駛區域地圖之建立

本篇論文為無人飛行器的自動探索提出了一個完整的系統架構。在此架構下，五個平行運作的程序各自負責不同的功能。首先，一個獨立運行的位置控制器持續地向飛行器傳遞控制訊號。這使得飛行器的位置控制不會受到中斷，即使其他的程序正在等待各自尚未完成的運算。接著，Large Scale Direct SLAM (LSD-SLAM) 程序負責飛行器的自體定位以及半稠密的影像深度估測，而該半稠密深度經蒐集後，由Octomap 程序建立一個三圍的佔據圖(occupancy map)。利用立體空間中的深度點資訊，本系統中的導航程序便能夠處進行飛行器導航與環境探索。幾乎所有單眼影像系統，如LSD-SLAM，都需要進行尺度復原，而本系統架構中的尺度估測程序，可以在一個超聲波距離感測器的幫助下，估測出單眼影像系統的尺度。有了一個估測的尺度後，LSD-SLAM提供的自體定位便能用以進行無人飛行器導航，並且根據所建立的三圍佔據圖達到無碰撞飛行。本論文提供了在模擬環境與實際情況下的無人飛行器位置控制器，並且通過實驗進行測試。本篇論文也提出了無人飛行器的自動環境探索策略。在探索的每一步中，目前可飛行區域的所有邊界點將會被當作是下一個探索點的候選點。對於每一個候選點，與之對應的預期資訊增加量與預期的可追蹤性會被個別計算，並且從中選出最適合的候選點，做為下一個探索點。如此，無人飛行器便能飛往能夠提供最多新資訊的位置，並且能夠持續進行自體定位。最後，通過模擬實驗以及實際實驗，本篇論文對所提出的系統架構以及方法進行了測試與驗證。

關鍵字

自動探索；自動導航；單眼視覺同步定位與地圖建構；即時系統；路徑規劃；機器人視覺；飛行器控制

並列摘要

In this thesis, a system structure for Unmanned Aerial Vehicle (UAV) quadrotor navigation and exploration is proposed. The system is composed of five nodes running in parallel, each responsible for different tasks. First of all, a pose controller continuously sends control commands to the quadrotor. This enables the quadrotor to continue in pose controlling even when other nodes are waiting for their computations to complete. The Large Scale Direct SLAM (LSD-SLAM) provides self-localization of the quadrotor and semi-dense depth map estimation. The collected depth maps are then integrated into a three-dimensional occupancy map with the Octomap, and the occupancy information of the currently built map is utilized by the Navigation node, dealing with the navigation of the quadrotor and the exploration of the environment. For the scale recovery of monocular vision-based systems, such as the LSD-SLAM system, the Scale Estimation node provides an estimate of the scale with the aid of an ultrasound sensor. Given the estimated scale, the quadrotor can then be navigated with the re-scaled pose provided by the LSD-SLAM system. The pose controller for both the simulated quadrotor and the real-world quadrotor are designed and verified. Also with the estimated scale, the proposed system avoids collision of the quadrotor body with any obstacle, allowing the quadrotor to only approach to those areas that are free of collision. An exploration strategy is also proposed in this thesis, which is essentially a frontier-based Next-Best-View (NBV) exploration strategy. The next point to approach to during each step of exploration is selected by taking the extracted frontier points as NBV candidates. The expected visibility-gain as well as the expected trackability are calculated for each NBV candidate and the most qualified among all is selected, moving towards the highest information gain in the meanwhile ensuring the trackability while moving. Experiments in both simulation and the real world are provided to verify the proposed system and the proposed exploration strategy.

並列關鍵字

autonomous exploration ； autonomous navigation ； monocular SLAM system ； Real-time systems ； path planning ； robot vision ； aerospace control

參考文獻

References

Google Scholar

[1: Engel et al. 2014]

Google Scholar

J. Engel, T. Schops, and D. Cremers, “LSD-SLAM: Large-scale direct monocular SLAM,” in Proceedings of the European Conference on Computer Vision (ECCV), Zurich, Switzerland, 2014.

Google Scholar

[2: Yamauchi 1997]

Google Scholar

B. Yamauchi, “A frontier-based approach for autonomous exploration,” in Proceedings of the 1997 IEEE International symposium on computational intelligence in robotics and automation (CIRA-97), Orlando, FL, USA, pp. 146–151.

Google Scholar

國際替代計量

基於單眼視覺之無人飛行器自動環境探索與可行駛區域地圖之建立

全文下載

主題瀏覽