Long-Term Exploration for Mobile Robots

点击翻译按钮获取中文摘要

Mobile robots are currently being used in challenging applications indoors, underground, underwater, in urban canyons, in forests, and on other planets. Despite recent advances in the field, creation of a unifying long-term exploration framework (comprised of localization, mapping, and planning) that can work online, in large-scale, unstructured, three-dimensional environments, without prior maps and using only onboard sensors, is an open problem in mobile robotics and is precisely the topic of this proposal.Whether for planetary exploration or office mail delivery, we may be interested in having a mobile robot carry out tasks that are too dangerous or dull for humans. For a long time in robotics, the world was considered to be static and thus a map could be built up in advance of task execution and then used by the robot to operate. Even successful techniques such as simultaneous localization and mapping (SLAM) have historically assumed that each place in the world has only a single appearance or `face'. We now know that to have robots operating in long-term service roles, we must acknowledge that the world is not static and that a single location can have a variety of appearances (e.g., different lighting conditions, varying weather, objects moving). We seek to develop navigation solutions that allow robots to operate in the real world without restrictive assumptions.