Issues in Autonomous Mobile Robot Navigation

1 The author would like to express his gratitude to everybody who read preliminary versions of this report and gave invaluable feedback; especially Chris Brown, who spent innumerable hours poring over this paper. It would not have been possible without him. Abstract Three main problems facing outdoor autonomous mobile robot navigation are un-structured environments, moving obstacles, and multiple sensors. Each of these leads to uncertainties that usually cannot be resolved using techniques popular for indoor navigation. However, by modularizing the task of navigation and treating it as three diierent subtasks (robot localization, goal recognition, and path planning), we can use some of the techniques in combination to build a mobile robot system than can work in the outdoor world. We present and summarize an extant body of literature dealing withthe unstructured dynamic environments observed via multiple sensing modalities. An in-depth survey of the current research literature in the eld shows the lack of research eeorts directed at generating generalized autonomous mobile robot navigation systems. We believe that while a single technique cannot be used to create a robust and ee-cient generalized autonomous mobile robot navigation system, a combination of some of the diierent methods in a probabilistic framework can achieve this goal. We propose a probabilistic framework that is based on the generalized framework approach. We use a modiied version of the occupancy grids, called dynamic occupancy grids as our representation of the environment and a modiication of Bayesian networks, termed dynamic Bayesian networks as our tool for updating the dynamic occupancy grid representation. We use an asynchronous update methodology that uses dynamic Bayesian networks to create new probability density functions (PDF) that are then used to update the information stored in the occupancy grids. Whenever a robot needs to perform some new motion task, it discretizes the PDFs currently associated with the dynamic occupancy grids to create a uniied view of the current state of the environment. We have recently acquired two mobile wheelchair robots with limited oo-road navigation capabilities. These will serve as a test-bed for our framework in the real world. We are also in the process of building simulated virtual environments on an SGI machine for use as a simulator test-bed to model multiple complex test environments. We are currently involved in two major research projects related to the development of our proposed framework. The rst research eeort was performed as part of the Mobile Robot …