Received: 26 March 2018 Revised: 18 December 2018 Accepted: 31 January 2019 DOI: 10.1002/asjc.2107 SPECIAL ISSUE A data-driven particle filter for terrain based navigation of sensor-limited autonomous underwater vehicles José Melo 1 Aníbal Matos 1,2 1 INESC TEC - INESC Technology and Science, Porto, Portugal 2 FEUP - Faculty of Engineering, University of Porto, Porto, Portugal Correspondence José Melo, INESC TEC - INESC Technology and Science, Porto, Portugal. Email: jose.melo@fe.up.pt Abstract In this article a new Data-Driven formulation of the Particle Filter framework is proposed. The new formulation is able to learn an approximate proposal distribution from previous data. By doing so, the need to explicitly model all the disturbances that might affect the system is relaxed. Such characteristics are particularly suited for Terrain Based Navigation for sensor-limited AUVs, where typical scenarios often include non-negligible sources of noise affecting the system, which are unknown and hard to model. Numerical results are pre- sented that demonstrate the superior accuracy, robustness and efficiency of the proposed Data-Driven approach. KEYWORDS autonomous underwater vehicles, data-driven learning, particle filter, terrain based navigation, underwater navigation 1 INTRODUCTION Terrain Based Navigation (TBN) is a term used to refer to a class of algorithms that takes advantage of variations of the terrain to obtain navigation position fixes, in a pro- cess that is similar to what happens for instance with the use of Global Navigation Satellite Systems (GNSS). In fact, information about the terrain, or bottom topography, can be very powerful not only for the case of TBN, but also, for example, for proximity navigation relative to drifting iceberg, as suggested in [1]. Underwater TBN is a fairly recent topic, with the first body of work on the topic dating from the early 1990s, with the initial approaches focused on using dense sen- sors, able to map large areas of terrain within a single measurement acquisition step. The experimental valida- tion of such approaches was also consistently coupled with the use of high-grade INS. However, recently, the study of TBN for sensor-limited Autonomous Underwater Vehicles (AUVs) has been reported by several authors, for example [2,3]. Sensor-limited systems refer to a class of vehicles equipped with low-information sonar like Doppler Velocity Log (DVL) or altimeters, but also low accuracy inertial measurement units (IMU). The use of low grade IMUs motivates a tightly-coupled integration between all the sensors, but also requires an online estimation of critical sensor errors 4. Accurate modelling of such errors will definitely yield better per- formance in terms of navigation accuracy of the system. However this requires a thorough understanding of the underlying physical properties of the system, but also a detailed modelling of those sources of error, which is not always easy or even possible to achieve. For a complete and up to date review of state of the art TBN algorithms for AUVs, the reader is referred to [5]. This article presents a novel data-driven approach to underwater TBN for sensor-limited systems. Data-driven methods do not depend on an explicit and detailed model of the environment. Instead, these methods are based on statistical models or machine learning techniques, which ............................................................................................................................................................... © 2019 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd Asian J Control. 2019;1–12. wileyonlinelibrary.com/journal/asjc 1