Bharath Kalyan

A novel formulation of the Bayes recursion for single-cluster filtering

In this paper we address the problem of tracking several moving targets with a sensor whose location and orientation are uncertain. This is a generalization of the well-known problem of feature-based simultaneous localization and mapping (SLAM). It is also a generalization of multitarget tracking (MTT) in general, and related to sensor bias estimation. We address such problems from the perspective of finite set statistics (FISST) and point process theory, and develop general expressions for the posterior multiobject density, as represented by probability-generating functionals (p.g.fl.s). We discuss how this general solution relates to approximative solutions previously suggested in the literature, and we also discuss how the p.g.fl. should be defined for such problems. To the best of our knowledge, this is the first paper to outline a FISST-based treatment of explicit data association for SLAM and related problems.

A feasibility analysis on using bathymetry for navigation of autonomous underwater vehicles

Bathymetric terrain maps generated from acoustic data offer an attractive alternative for reducing the submerged pose error estimates for autonomous underwater vehicles (AUVs). The goal of this work is to determine the extent of improvement in the navigational accuracy of an AUV equipped with an echo sounder for near-seafloor, shallow water applications. Given bathymetric variations of a certain terrain, this paper analyzes the best achievable positioning accuracy for AUVs. To counter for the strong non-linearity and the non-Gaussian nature of the problem, an optimal Bayesian estimator is initially derived. The fundamental limitations in the pose uncertainty using this approach is encompassed by the Posterior Cramér-Rao bound (PCRB), that is interpreted in terms of the sonar sensor accuracy and the bathymetric variations. The PCRB on the position error covariance is determined and it is shown that the Bayesian Bootstrap filter closely follows this bound using real inferometric sonar data.

Spatial Modeling of Deep-Sea Ferromanganese Nodules With Limited Data Using Neural Networks

Deep-sea ferromanganese nodules found in the Clarion–Clipperton zone (CCZ) in the Pacific ocean are a large potential source of metals such as nickel, cobalt, and manganese. Spatial modeling of these nodules is essential to obtain a better scientific understanding about their formation and distribution, and conduct feasibility studies on their exploitation. However, data on the quantitative and qualitative distribution of nodules in CCZ are sparse and often not divulged, and the accuracy of conventional spatial modeling techniques is limited by this scarcity of data. We present an approach based on artificial neural networks for modeling nodule parameters in the CCZ using the limited data available in the open domain. Our models predictions are comparable to benchmark predictions from the International Seabed Authority which used a more extensive data set. Moreover, our model can predict small as well as large-scale variations of nodules, which are essential in making evaluations for deep-sea harvesting. We discuss the contribution of each factor in the modeling, and show that small-scale nodule parameter variations can be effectively predicted by incorporating the local topography.

Polymetallic nodules abundance estimation using sidescan sonar: A quantitative approach using artificial neural network

There is a high abundance of polymetallic nodules (PMN) scattered across the vast Clarion and Clipperton Fracture Zone (CCFZ) in the Pacific Ocean. These nodules possess high economic potential as they are rich in minerals such as manganese, nickel, copper and rare earth elements. Quantification of nodule coverage is important for economic feasibility studies and planning of effective exploitation strategies. Traditional methods for nodule quantification are highly labour and time intensive as they rely on freefall box corer measurements and/or image processing of seabed photographs. Using sidescan sonar data and geotagged photographs collected from an autonomous underwater vehicle (AUV) in our region of interest at CCFZ, we propose a novel technique based on artificial neural network (ANN) to estimate PMN abundance using texture variations from sidescan sonar data. Compared to an optical camera, the sidescan sonar provides a much larger area of coverage, which in effect can drastically increase the area surveyed by an AUV in a given amount of time. Till date, this is the first known published work to elaborate on a data-driven approach in estimating PMN abundance using sidescan sonar backscatter data. Our network yielded a test accuracy of 84%, which shows that it can be used as an effective tool in estimating nodule abundance from sidescan sonar. This approach allows faster evaluation of nodule abundance for future exploration without the need for an underwater camera.

A clutter rejection filter for sonar feature based navigation in marine environments

This paper examines the detection of landmarks in the presence of false measurements from a blazed array sonar using random finite set models. A clutter rejection filter that is based on the fusion of the moment-approximation of the posterior density, also known as probability hypothesis density (PHD), within the random finite set framework with the conventional Extended Kalman Filter simultaneous localization and mapping (EKF-SLAM) framework is presented. The PHD clutter filter is effectively used to reduce false measurements, thereby feeding mainly landmark originated measurements to the EKF based navigational filter. This effectively simplifies the data association technique needed within the navigation filter framework and completely obviates the need for external map/feature management strategies. The efficacy of the proposed approach is demonstrated by controlled field experiments in marine environments using an autonomous surface craft (ASC) equipped with the navigational sensory suite, viz., GPS, triple axis gyroscope, Doppler Velocity Log (DVL) and a payload in form of an underwater blazed array sonar.