Towards Learned Traversability for Robot Navigation: From Underfoot to the Far Field Andrew Howard, Michael Turmon, Larry Matthies, Benyang Tang, and Anelia Angelova Jet Propulsion Laboratory California Institute of Technology Pasadena, California 91109 e-mail: andrew.howard@jpl.nasa.gov, michael.turmon@jpl.nasa.gov, larry.matthies@jpl.nasa.gov, benyang.tang@jpl.nasa.gov, anelia.angelova@jpl.nasa.gov Eric Mjolsness Department of Computer Science University of California Irvine, California 92697 e-mail: emj@uci.edu Received 1 April 2006; accepted 30 October 2006 Autonomous off-road navigation of robotic ground vehicles has important applications on Earth and in space exploration. Progress in this domain has been retarded by the lim- ited lookahead range of three-dimensional 3D sensors and by the difficulty of heuris- tically programming systems to understand the traversability of the wide variety of ter- rain they can encounter. Enabling robots to learn from experience may alleviate both of these problems. We define two paradigms for this, learning from 3D geometry and learning from proprioception, and describe initial instantiations of them as developed under DARPA and NASA programs. Field test results show promise for learning traversability of veg- etated terrain and learning to extend the lookahead range of the vision system. © 2007 Wiley Periodicals, Inc. 1. INTRODUCTION Robotic ground vehicles for outdoor applications have achieved some remarkable successes, notably in autonomous highway following Dickmanns & Mys- liwetz, 1992; Pomerleau & Jochem, 1996, planetary exploration Bapna et al., 1998; Biesiadecki et al., 2005; Leger et al., 2005; Maimone, Biesiadecki, Tun- stel, Cheng & Leger, 2006, and off-road navigation on Earth Lacaze, Murphy & DelGiorno, 2002; Boldt & Camden, 2004; Krotkov et al., 2006. Nevertheless, major challenges remain to enable reliable, high- • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • Journal of Field Robotics 23(11/12), 1005–1017 (2006) © 2007 Wiley Periodicals, Inc. Published online in Wiley InterScience (www.interscience.wiley.com). • DOI: 10.1002/rob.20168