History-Aware Free Space Detection For Efficient Autonomous Exploration Using Aerial Robots
2019 IEEE AEROSPACE CONFERENCE(2019)
摘要
In this work, we present an approach for the detection of the direction of free space in order to improve the efficiency of robotic exploration by exploiting the history of free space calculations. As a motivational example, we consider the case of exploration of subterranean environments where the length of corridors can exceed the range of most sensors, multibranched geometry may lead to ambiguity with respect to the most efficient direction of exploration, or sensor degradation can shorten the effective depth range. The proposed method can be used to assist a path planner by determining the directions of probable free space for efficient exploration. The algorithm was evaluated using point clouds from two types of sensors, namely sparse long-range sensors such as a LiDAR and dense shortrange sensors such as direct depth RGBD sensors. Furthermore, evaluation took place against a variety of environments using handheld and aerial robotic data in urban and subterranean environments. During each of the tests, the algorithm has shown to be capable of consistently and reliably finding the directions of probable unobserved free space in real-time. As a final evaluation step, the proposed algorithm was integrated as part of the path planning functionality on-board an autonomous aerial robot and the relevant mine exploration field results are shown. Analysis of computational efficiency is further presented. The code for this method is open-sourced and accompanies this paper submission.
更多查看译文
关键词
history-aware free space detection,free space calculations,subterranean environments,multibranched geometry,sensor degradation,effective depth range,path planner,probable free space,long-range sensors,dense short-range sensors,direct depth RGBD sensors,aerial robotic data,probable unobserved free space,autonomous aerial robot,computational efficiency,mine exploration field,path planning functionality
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络