Intelligent Maps for Autonomous Kilometer-Scale Science Survey

Abstract

We present a new approach for remote exploration by autonomous surface robots. In our method the agent synthesizes in situ measurements with remote sensing data to learn a multi-scale model of the explored environment. This "intelligent map" predicts the information value of candidate observations to guide adaptive navigation and sampling decisions. The agent learns map parameters on the fly, modifying its exploration behavior in response to novel correlations, resource constraints and execution errors. Rover tests at Amboy Crater, California demonstrate improved performance over non-adaptive strategies for a geologic site survey task.