Intelligent Mapping for Autonomous Robotic Survey

Abstract

In general today? planetary exploration robots do not travel beyond the previous day? imagery. However, advances in autonomous navigation will soon permit traverses of multiple kilometers. These long traverses present new challenges to science-driven exploration. Rovers will travel over their local horizon so that scientists will not be able to specify targets in advance. Moreover, energy and time shortages will continue to limit the number of measurements so that sampling density will decrease as mobility improves. Finally, constraints on communications bandwidth will preclude transmitting most of the collected data. These issues raise the question: is it possible to explore efficiently, with long traverses and sparse sampling, without sacrificing our understanding of the visited terrain? ?cience autonomy?addresses the optimal sampling problem through onboard data understanding. Pattern recognition, learning and planning technologies can enable robots to place instruments and take measurements without human supervision. These robots can autonomously choose the most important features to observe and transmit. This document argues that these agents should learn and exploit structure in the explored environment. In other words, they must be mapmakers. We advocate an intelligent mapping approach in which onboard predictive models represent spatial structure (similarities from one locale to the next) and cross-sensor structure (correlations between different sensing scales). These models guide the agent? exploration to informative areas while minimizing redundant sampling. The generative model allows us to formulate the exploration problem in terms of established principles of experimental design. Spatial experimental design criteria guide exploration decisions and suggest the best data products for downlink. In addition the map itself functions as a bandwidth-efficient representation of data gathered during the traverse. This work bridges the gap between Bayesian experimental design, robotic mapping and their application in autonomous surficial geology. We develop generative data models that are appropriate for geologic mapping and site survey by planetary rovers. We present algorithms for learning map parameters on the fly and leveraging these contextual cues to choose optimal data collection and return actions. Finally we implement and test adaptive exploration schemes for kilometer-scale site survey tasks.