Underwater localization and mapping: observability analysis and experimental results

Abstract

Purpose-We investigate the observability properties of the process of simultaneous localization and mapping of an Autonomous Underwater Vehicle (AUV), a challenging and important problem in marine robotics, and illustrate the derived results through computer simulations and experimental results with a real AUV. Design/methodology/approach-We address the single/multiple beacon observability analysis of the process of simultaneous localization and mapping of an Autonomous Underwater Vehicle (AUV) by deriving the nonlinear mathematical model that describes the process; then applying a suitable coordinate transformation, and subsequently a time-scaling transformation to obtain a Linear Time Varying (LTV) system. The AUV considered is equipped with a set of inertial sensors, a depth sensor, and an acoustic ranging device that provides relative range measurements to a set of stationary beacons. The location of the beacons do not need to be necessarily known and in that case, we are also interested to localize them. Numerical tests and experimental results illustrate the derived theoretical results. Findings-We show that, if either the position of one of the beacons or the initial position of the AUV is known, then the system is at least locally weakly observable, in the sense that the set of indistinguishable states from a given initial configuration contains a finite set of isolated points. The simulations and experiments results illustrate the findings. Originality/value-In the single and multiple beacon case and for manoeuvres with constant linear and angular velocities both expressed in the body-frame, known as trimming or steady-state trajectories, we derive conditions under which it is possible to infer the state of the resulting system (and in particular the position of the AUV). We also describe implementation of an advanced continuous time constrained minimum energy observer combined with multiple model techniques. Numerical tests and experimental results illustrate the derived theoretical results.

Robust Huber-Based Iterated Divided Difference Filtering with Application to Cooperative Localization of Autonomous Underwater Vehicles

“…In this scheme, only a single surface leader was used; only range information can be observed and low update frequency for cooperative localization. All the cooperative conditions above will result in weak observability for an AUV state estimate; this has been acknowledged and proven by many early studies [ 5 , 12 , 14 , 35 , 36 ]. Although the performance can be improved by using the observed information iteratively, the outlier measurements will cause the estimate divergent.…”

“…The Huber method [ 28 , 29 ] is a recursive algorithm, in which the actual measurements and the state correction attained takes the form of a linear regression problem between the predicted state and the observed quantity. Using this technique, some robust filtering approaches [ 26 , 32 , 36 – 38 ] have been developed and successfully applied to elliptical orbit design and tracking problems. To apply this method in the DDF, it is first required to recast the measurement update as a regression problem between the observed quantity and the state prediction.…”

“…Also note that as γ → 0, the problem reduces to the absolute value estimator. This blend of estimation techniques provides the HDDF estimator robustness against deviations from Gaussian distributed random measurements errors [ 26 , 36 , 40 ]. This robustness arises from the ψ matrix and that all residuals are not weighted equally.…”

“…In virtue of the proving method of the advantage in weak observability and large initial errors of iterated Unscented Kalman Filter in [ 22 ], we can prove the advantage in tracking accuracy and convergence speed of the IDDF algorithm in AUV cooperative localization. The robustness of the Huber-based DDF method to the contaminated outlier measurement has been proved in [ 36 , 40 ]. The proposed HIDDF filter switches between the IDDF and the Huber-based DDF through switch mode function and error threshold, which does not affect the character of the two methods and combines the advantages of both, so the HIDDF method does not only achieve a more accurate position estimate in weak observability and faster convergence speed in large initial error, but also an improved robustness to the outlier measurements.…”