Visual robot navigation

点击翻译按钮获取中文摘要

The objective of this research program is to develop methods for robots to navigate in unstructured environments.  Unstructured environments are those which have not been specifically designed for robot operation.  The approach taken here is partially inspired by biological models.  The current view of navigation in insects such as bees, ants, and wasps is that these insects capture "snapshot images" throughout their environment.  They then use a procedure called visual homing to move between snapshot positions.  This research will apply a similar methodology to robots. Robot navigation is a difficult problem for two fundamental reasons:  (1) Using odometric information to estimate position incurs a cumulative error.  In other words, the further a robot travels, the more unsure it becomes of its position.  (2) Places that have the same appearance cannot be distinguished.  Interestingly, most existing methods can address one of the above difficulties, but seldom both. This research program will utilize visual homing to allow a robot to travel between places where it had previously captured snapshot images.  We will begin with an existing model for navigation based on homing, and incorporate additional insights so as to resolve the problems described above.  The desired end product of this research will be a set of methods which allow a robot to travel robustly between known places.  The challenge is to make these methods truly robust so that available visual cues are utilized while noise is ignored. The applications for robustly navigating robots are both numerous and diverse.  Some example applications are autonomous driving, planetary exploration, search and rescue, surveillance, remote sensing, and household cleaning.