A Complete Navigation System for Goal Acquisition in Unknown Environments

Abstract

Autonomous outdoor navigation has broad application in mining, construction, planetary exploration, and military reconnaissance. To date, most of the work tested on actual robots has centered on local navigation tasks such as avoiding obstacles or following roads. Global navigation has been limited to simple wandering, path tracking, straight-line goal seeking behaviors, or executing a sequence of scripted local behaviors. The problem of global navigation in outdoor environments has been addressed in the literature, but it is generally assumed that the world exhibits coarse topological structure, most of which is known, and that sensors and position estimation systems provide highly-accurate data. These assumptions break down for real robots in highly unstructured and unknown environments. With every image, the sensors provide new information about the world that can impact the robot's path to the goal. Some of the information is real, some arises from noise, and some arises from aliasing due to robot position error. Replanning may be needed for every image, and it may be nontrivial due to the unstructured nature of the environment. To address these problems, we have developed a complete system that integrates local and global navigation. This system is capable of finding goal given no a priori map of the environment. It is robust to noise, vehicle position error, and is able to replan in real-time. We describe the system and present the results of experiments performed using a real robot.