Benyang Tang

Towards learned traversability for robot navigation: From underfoot to the far field

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 limited lookahead range of three-dimensional (3D) sensors and by the difficulty of heuristically programming systems to understand the traversability of the wide variety of terrain 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 vegetated terrain and learning to extend the lookahead range of the vision system.

Learning long-range terrain classification for autonomous navigation

This paper describes a method for learning the terrain classification of long-range appearance data from short- range, stereo-based geometry, along with a map representation for utilizing this data to improve autonomous off-road navigation. The continuous, online learning method allows the system to constantly adapt to changing terrain and environmental conditions, while the polar-perspective map representation allows the system to effectively plan with stereo data at long ranges. Various evaluations of the long-range classification and improvements in system performance are described, including results from an independent third-party testing team.

On-board analysis of uncalibrated data for a spacecraft at mars

Analyzing data on-board a spacecraft as it is collected enables several advanced spacecraft capabilities, such as prioritizing observations to make the best use of limited bandwidth and reacting to dynamic events as they happen. In this paper, we describe how we addressed the unique challenges associated with on-board mining of data as it is collected: uncalibrated data, noisy observations, and severe limitations on computational and memory resources. The goal of this effort, which falls into the emerging application area of spacecraft-based data mining, was to study three specific science phenomena on Mars. Following previous work that used a linear support vector machine (SVM) on-board the Earth Observing 1 (EO-1)spacecraft, we developed three data mining techniques for use on-board the Mars Odyssey spacecraft. These methods range from simple thresholding to state-of-the-art reduced-set SVM technology. We tested these algorithms on archived data in a flight software testbed. We also describe a significant, serendipitous science discovery of this data mining effort: the confirmation of a water ice annulus around the north polar cap of Mars. We conclude with a discussion on lessons learned in developing algorithms for use on-board a spacecraft.

Statistical Approaches to Detecting Transient Signals in GPS: Results from the 2009–2011 Transient Detection Exercise

We present the results of our participation in four phases of the Southern California Earthquake Center (SCEC) transient detection exercise (Lohman and Murray, 2013). In each phase, a blind test was conducted in which sets of synthetic Global Positioning Systems (GPS) data were released and a deadline set for submission of detection results. For each data set, the presence or absence of transient events was to be determined, and the location and time of each specified. After all submissions were received, the ground‐truth information about any transient signals in the data was released. The synthetic data sets were generated by FAKENET, a software package that produces realistic GPS time series that include secular motion and seasonal signals as well as realistic noise and distributions of missing data (Agnew, 2013). Station locations in the synthetic data set were a subset of GPS installations in the western United States. In this work, we pursue a purely data‐driven approach to transient detection, rather than one based on an assumption of an underlying physical model. We view this approach as having two important advantages. First, it facilitates the detection of events, which might happen through a previously unknown, but geophysically interesting, physical process or on a previously unknown fault or structure. Second, it enables the detection of events, which have nothing to do with the solid earth but, although not the subject of the SCEC exercise, have scientific or practical merit. These include signals such as those due to atmospheric phenomena as well as signals, which result from hardware faults or failures and software processing glitches. The four phases of the exercise were conducted over approximately two years. As a result, our approach evolved as we learned from experiences in the initial phases as well as from other ongoing work during that period. The …