ALBERT

Description: Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution February 2008

Description: This thesis formulates an estimation framework for Simultaneous Localization and Mapping (SLAM) that addresses the problem of scalability in large environments. We describe an estimation-theoretic algorithm that achieves significant gains in computational efficiency while maintaining consistent estimates for the vehicle pose and the map of the environment. We specifically address the feature-based SLAM problem in which the robot represents the environment as a collection of landmarks. The thesis takes a Bayesian approach whereby we maintain a joint posterior over the vehicle pose and feature states, conditioned upon measurement data. We model the distribution as Gaussian and parametrize the posterior in the canonical form, in terms of the information (inverse covariance) matrix. When sparse, this representation is amenable to computationally efficient Bayesian SLAM filtering. However, while a large majority of the elements within the normalized information matrix are very small in magnitude, it is fully populated nonetheless. Recent feature-based SLAM filters achieve the scalability benefits of a sparse parametrization by explicitly pruning these weak links in an effort to enforce sparsity. We analyze one such algorithm, the Sparse Extended Information Filter (SEIF), which has laid much of the groundwork concerning the computational benefits of the sparse canonical form. The thesis performs a detailed analysis of the process by which the SEIF approximates the sparsity of the information matrix and reveals key insights into the consequences of different sparsification strategies. We demonstrate that the SEIF yields a sparse approximation to the posterior that is inconsistent, suffering from exaggerated confidence estimates. This overconfidence has detrimental effects on important aspects of the SLAM process and affects the higher level goal of producing accurate maps for subsequent localization and path planning. This thesis proposes an alternative scalable filter that maintains sparsity while preserving the consistency of the distribution. We leverage insights into the natural structure of the feature-based canonical parametrization and derive a method that actively maintains an exactly sparse posterior. Our algorithm exploits the structure of the parametrization to achieve gains in efficiency, with a computational cost that scales linearly with the size of the map. Unlike similar techniques that sacrifice consistency for improved scalability, our algorithm performs inference over a posterior that is conservative relative to the nominal Gaussian distribution. Consequently, we preserve the consistency of the pose and map estimates and avoid the effects of an overconfident posterior. We demonstrate our filter alongside the SEIF and the standard EKF both in simulation as well as on two real-world datasets. While we maintain the computational advantages of an exactly sparse representation, the results show convincingly that our method yields conservative estimates for the robot pose and map that are nearly identical to those of the original Gaussian distribution as produced by the EKF, but at much less computational expense. The thesis concludes with an extension of our SLAM filter to a complex underwater environment. We describe a systems-level framework for localization and mapping relative to a ship hull with an Autonomous Underwater Vehicle (AUV) equipped with a forward-looking sonar. The approach utilizes our filter to fuse measurements of vehicle attitude and motion from onboard sensors with data from sonar images of the hull. We employ the system to perform three-dimensional, 6-DOF SLAM on a ship hull.

Description: Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution February 2009

Description: Self-localization of an underwater vehicle is particularly challenging due to the absence of Global Positioning System (GPS) reception or features at known positions that could otherwise have been used for position computation. Thus Autonomous Underwater Vehicle (AUV) applications typically require the pre-deployment of a set of beacons. This thesis examines the scenario in which the members of a group of AUVs exchange navigation information with one another so as to improve their individual position estimates. We describe how the underwater environment poses unique challenges to vehicle navigation not encountered in other environments in which robots operate and how cooperation can improve the performance of self-localization. As intra-vehicle communication is crucial to cooperation, we also address the constraints of the communication channel and the effect that these constraints have on the design of cooperation strategies. The classical approaches to underwater self-localization of a single vehicle, as well as more recently developed techniques are presented. We then examine how methods used for cooperating land-vehicles can be transferred to the underwater domain. An algorithm for distributed self-localization, which is designed to take the specific characteristics of the environment into account, is proposed. We also address how correlated position estimates of cooperating vehicles can lead to overconfidence in individual position estimates. Finally, key to any successful cooperative navigation strategy is the incorporation of the relative positioning between vehicles. The performance of localization algorithms with different geometries is analyzed and a distributed algorithm for the dynamic positioning of vehicles, which serve as dedicated navigation beacons for a fleet of AUVs, is proposed.

Description: This work was funded by Office of Naval Research grants N00014-97-1-0202, N00014-05-1-0255, N00014-02-C-0210, N00014-07-1-1102 and the ASAP MURI program led by Naomi Leonard of Princeton University.

Description: Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution June 2009

Description: This research presents a parallel hypothesis method for autonomous underwater vehicle navigation that enables a vehicle to expand the operating envelope of existing long baseline acoustic navigation systems by incorporating information that is not normally used. The parallel hypothesis method allows the in-situ identification of acoustic multipath time-of-flight measurements between a vehicle and an external transponder and uses them in real-time to augment the navigation algorithm during periods when direct-path time-of-flight measurements are not available. A proof of concept was conducted using real-world data obtained by the Woods Hole Oceanographic Institution Deep Submergence Lab's Autonomous Benthic Explorer (ABE) and Sentry autonomous underwater vehicles during operations on the Juan de Fuca Ridge. This algorithm uses a nested architecture to break the navigation solution down into basic building blocks for each type of available external information. The algorithm classifies external information as either line of position or gridded observations. For any line of position observation, the algorithm generates a multi-modal block of parallel position estimate hypotheses. The multimodal hypotheses are input into an arbiter which produces a single unimodal output. If a priori maps of gridded information are available, they are used within the arbiter structure to aid in the elimination of false hypotheses. For the proof of concept, this research uses ranges from a single external acoustic transponder in the hypothesis generation process and grids of low-resolution bathymetric data from a ship-based multibeam sonar in the arbitration process. The major contributions of this research include the in-situ identification of acoustic multipath time-of-flight measurements, the multiscale utilization of a priori low-resolution bathymetric data in a high-resolution navigation algorithm, and the design of a navigation algorithm with a exible architecture. This flexible architecture allows the incorporation of multimodal beliefs without requiring a complex mechanism for real-time hypothesis generation and culling, and it allows the real-time incorporation of multiple types of external information as they become available in situ into the overall navigation solution.

Description: Submitted in partial fulfillment of the requirements for the degree of Doctor of Science at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution January 1983

Description: A simple model for the bottom boundary layer on the continental shelf is presented. The governing equations are developed for a stratified, turbulent Ekman layer in a combined wave and current flow over a moveable sediment bed. An eddy diffusivity closure scheme that includes the effect of suspended sediment, temperature, and salinity induced stratification on the vertical turbulent diffusion of mass and momentum couples the resulting unsteady conservation equations for fluid momentum, fluid mass, and suspended sediment mass. The wave velocity, current velocity, and suspended sediment concentration profiles predicted by the simultaneous solution of the conservation equations require the physical bottom roughness and a sediment reference concentrati on to be specified as boundary conditions. The physical bottom roughness associated with biologically generated bedforms, wave generated ripples, and near bed sediment transport are calculated as functions of the flow and sediment conditions. Using expressions for the height of sediment transporting layer and the sediment velocity, an expression for the sediment reference concentration is developed by matching laboratory measurements of sediment transport rates in oscillatory flow. The model predicts that the bottom flow field is highly dependent on (1) the nonlinear wave and current interaction, which increases the boundary shear stress and enhances vertical turbulent diffusion, (2) the effect of the boundary shear stress on a moveable sediment bed, which determines the physical bottom roughness and the amount of sediment in suspension, and (3) the effect of stable stratification, which inhibits vertical turbulent transport and couples the flow to the suspended sediment and fluid density profiles. The validity of the theoretical approach is supported by model predictions that are in excellent agreement with high quality data collected during two continental shelf bottom boundary layer experiments for a wide range of flow and bottom conditions.

Description: Funding for the work resulting in this Thesis has been provided by the American Gas Association (Project No. PR-153-126), the National Science Foundation (Grant No. OCE~8014930), and NOAA-Sea Grant (NA-79AA-D-0010l; NA 79AA-D-00102).

Description: Author Posting. © Annual Reviews, 2006. This article is posted here by permission of Annual Reviews for personal use, not for redistribution. The definitive version was published in Annual Review of Fluid Mechanics 38 (2006): 395-425, doi:10.1146/annurev.fluid.38.050304.092129.

Description: Over the past four decades, the combination of in situ and remote sensing observations has demonstrated that long nonlinear internal solitary-like waves are ubiquitous features of coastal oceans. The following provides an overview of the properties of steady internal solitary waves and the transient processes of wave generation and evolution, primarily from the point of view of weakly nonlinear theory, of which the Korteweg-de Vries equation is the most frequently used example. However, the oceanographically important processes of wave instability and breaking, generally inaccessible with these models, are also discussed. Furthermore, observations often show strongly nonlinear waves whose properties can only be explained with fully nonlinear models.

Description: KRH acknowledges support from NSF and ONR and an Independent Study Award from the Woods Hole Oceanographic Institution. WKM acknowledges support from NSF and ONR, which has made his work in this area possible, in close collaboration with former graduate students at Scripps Institution of Oceanography and MIT.

Description: Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution June 2006

Description: In this thesis, an innovative architecture for real-time adaptive and cooperative control of autonomous sensor platforms in a marine sensor network is described in the context of the autonomous oceanographic network scenario. This architecture has three major components, an intelligent, logical sensor that provides high-level environmental state information to a behavior-based autonomous vehicle control system, a new approach to behavior-based control of autonomous vehicles using multiple objective functions that allows reactive control in complex environments with multiple constraints, and an approach to cooperative robotics that is a hybrid between the swarm cooperation and intentional cooperation approaches. The mobility of the sensor platforms is a key advantage of this strategy, allowing dynamic optimization of the sensor locations with respect to the classification or localization of a process of interest including processes which can be time varying, not spatially isotropic and for which action is required in real-time. Experimental results are presented for a 2-D target tracking application in which fully autonomous surface craft using simulated bearing sensors acquire and track a moving target in open water. In the first example, a single sensor vehicle adaptively tracks a target while simultaneously relaying the estimated track to a second vehicle acting as a classification platform. In the second example, two spatially distributed sensor vehicles adaptively track a moving target by fusing their sensor information to form a single target track estimate. In both cases the goal is to adapt the platform motion to minimize the uncertainty of the target track parameter estimates. The link between the sensor platform motion and the target track estimate uncertainty is fully derived and this information is used to develop the behaviors for the sensor platform control system. The experimental results clearly illustrate the significant processing gain that spatially distributed sensors can achieve over a single sensor when observing a dynamic phenomenon as well as the viability of behavior-based control for dealing with uncertainty in complex situations in marine sensor networks.

Description: Supported by the Office of Naval Research, with a 3-year National Defense Science and Engineering Grant Fellowship and research assistantships through the Generic Ocean Array Technology Sonar (GOATS) project, contract N00014-97-1-0202 and contract N00014-05-G-0106 Delivery Order 008, PLUSNET: Persistent Littoral Undersea Surveillance Network.

Description: Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution June 2008

Description: Inspired by the swimming abilities of marine animals, this thesis presents "Finnegan the RoboTurtle", an autonomous underwater vehicle (AUV) powered entirely by four flapping foils. Biomimetic actuation is shown to produce dramatic improvements in AUV maneuvering at cruising speeds, while simultaneously allowing for agility at low speeds. Using control algorithms linear in the modified Rodrigues parameters to support large angle maneuvers, the vehicle is successfully controlled in banked and twisting turns, exceeding the best reported AUV turning performance by more than a factor of two; a minimum turning radius of 0.7BL, and the ability to avoid walls detected〉 1.8BL ahead, are found for cruising speeds of 0.75BL/S, with a maximum heading rate of 400 / S recorded. Observations of "Myrtle", a 250kg Green sea turtle (Chelonia mydas) at the New England Aquarium, are detailed; along with steady swimming, Myrtle is observed performing 1800 level turns and rapidly actuating pitch to control depth and speed. Limb kinematics for the level turning maneuver are replicated by Finnegan, and turning rates comparable to those of the turtle are achieved. Foil kinematics which produce approximately sinusoidal nominal angle of attack trace are shown to improve turning performance by as much as 25%; the effect is achieved despite limited knowledge of the flow field. Finally, tests with a single foil are used to demonstrate that biomimetically inspired inline motion can allow oscillating foils utilizing a power/recovery style stroke to generate as much as 90% of the thrust from a power/power stroke style motion.

Description: Submitted in partial fulfillment of the requirements for the degree of Master of Science at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution February 2003

Description: The Sentry AUV represents a radical departure from conventional AUV design, particularly with respect to actuation. The vehicle's combined foil/thruster actuators have the potential to produce a vehicle both maneuverable in the veritical plane and efficient in forward flight, well suited to survey work over rough topography. Capitalizing on this; however, requires an understanding of the vehicles dynamics. In this work, we present the development and analysis of an analytic model of the Sentry AUV. Our goals were to develop a model sufficiently accurate in terms of the mission profile to identify critical vehicle behaviors influencing successful mission completion. The analytical vehicle model was developed with structural accuracy in mind, and under the requirement that it handle a large range of vertical plane velocities, Our primary methodology for analysis was through the design of a linear controller, whose behavior was investigated in simulation and as implemented on a 1/4-scale physical model. Based on decoupled linearized models for near-horizontal flight derived from the full non-linear model, classical linear controllers were designed and validated by simulation and implementation on the physical model. Closed loop simulations conducted at high angle of attack verified the vehicle's predicted maneuverability in the vertical plane. Ultimately we determined the vehicle's input structure limited the achievable performance of a classical linear controller.

Description: Submitted in partial fulfillment of the requirements for the degree of Master of Science at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution September 2008

Description: Oceanographic applications of robotics are as varied as the undersea environment itself. As underwater robotics moves toward the study of dynamic processes with multiple vehicles, there is an increasing need to distill large volumes of data from underwater vehicles and deliver it quickly to human operators. While tethered robots are able to communicate data to surface observers instantly, communicating discoveries is more difficult for untethered vehicles. The ocean imposes severe limitations on wireless communications; light is quickly absorbed by seawater, and tradeoffs between frequency, bitrate and environmental effects result in data rates for acoustic modems that are routinely as low as tens of bits per second. These data rates usually limit telemetry to state and health information, to the exclusion of mission-specific science data. In this thesis, I present a system designed for communicating and presenting science telemetry from untethered underwater vehicles to surface observers. The system's goals are threefold: to aid human operators in understanding oceanographic processes, to enable human operators to play a role in adaptively responding to mission-specific data, and to accelerate mission planning from one vehicle dive to the next. The system uses standard lossy compression techniques to lower required data rates to those supported by commercially available acoustic modems (O(10)-O(100) bits per second). As part of the system, a method for compressing time-series science data based upon the Discrete Wavelet Transform (DWT) is explained, a number of low-bitrate image compression techniques are compared, and a novel user interface for reviewing transmitted telemetry is presented. Each component is motivated by science data from a variety of actual Autonomous Underwater Vehicle (AUV) missions performed in the last year.

Description: National Science Foundation Center for Subsurface Sensing and Imaging (CenSSIS ERC)