Machine Learning is the study of methods for programming computers to learn. Computers are applied to a wide range of tasks, and for most of these it is relatively easy for programmers to design and implement the necessary software. However, there are many tasks for which this is difficult or impossible. These can be divided into four general categories. First, there are problems for which there exist no human experts. For example, in modern automated manufacturing facilities, there is a need to predict machine failures before they occur by analyzing sensor readings. Because the machines are new, there are no human experts who can be interviewed by a programmer to provide the knowledge necessary to build a computer system. A machine learning system can study recorded data and subsequent machine failures and learn prediction rules. Second, there are problems where human experts exist, but where they are unable to explain their expertise. This is the case in many perceptual tasks, such as speech recognition, hand-writing recognition, and natural language understanding. Virtually all humans exhibit expert-level abilities on these tasks, but none of them can describe the detailed steps that they follow as they perform them. Fortunately, humans can provide machines with examples of the inputs and correct outputs for these tasks, so machine learning algorithms can learn to map the inputs to the outputs. Third, there are problems where phenomena are changing rapidly. In finance, for example, people would like to predict the future behavior of the stock market, of consumer purchases, or of exchange rates. These behaviors change frequently, so that even if a programmer could construct a good predictive computer program, it would need to be rewritten frequently. A learning program can relieve the programmer of this burden by constantly modifying and tuning a set of learned prediction rules. Fourth, there are applications that need to be customized for each computer user separately. Consider, for example, a program to filter unwanted electronic mail messages. Different users will need different filters. It is unreasonable to expect each user to program his or her own rules, and it is infeasible to provide every user with a software engineer to keep the rules up-to-date. A machine learning system can learn which mail messages the user rejects and maintain the filtering rules automatically. Machine learning addresses many of the same research questions as the fields of statistics, data mining, and psychology, but with differences of emphasis. Statistics focuses on understanding the phenomena that have generated the data, often with the goal of testing different hypotheses about those phenomena. Data mining seeks to find patterns in the data that are understandable by people. Psychological studies of human learning aspire to understand the mechanisms underlying the various learning behaviors exhibited by people (concept learning, skill acquisition, strategy change, etc.).

Machine learning

Doing qualitative research: a practical handbook, by David Silverman, Los Angeles, Sage, 2010, 456 pp., AU$65.00, ISBN 978-1-84860-033-1, ISBN 978-1-94960-034-8. Available in Australia and New Zeal...

Doing qualitative research: a practical handbook

A scheme is developed for classifying the types of motion perceived by a humanlike robot. It is assumed that the robot receives visual images of the scene using a perspective system model. Equations, theorems, concepts, clues, etc., relating the objects, their positions, and their motion to their images on the focal plane are presented. >

http://ieeexplore.ieee.org/iel2/815/3148/00100651.pdf

Robot vision

The concept of a miss-distance, or error, between a reference quantity and its estimated/controlled value, plays a fundamental role in any filtering/control problem. Yet there is no satisfactory notion of a miss-distance in the well-established field of multi-object filtering. In this paper, we outline the inconsistencies of existing metrics in the context of multi-object miss-distances for performance evaluation. We then propose a new mathematically and intuitively consistent metric that addresses the drawbacks of current multi-object performance evaluation metrics.

/pdf/a-consistent-metric-for-performance-evaluation-of-multi-1ejcrdyq6y.pdf

A Consistent Metric for Performance Evaluation of Multi-Object Filters

Image change detectio algorithms : A systematic survey

Several physico-mechanical designs evolved in fish are currently inspiring robotic devices for propulsion and maneuvering purposes in underwater vehicles. Considering the potential benefits involved, this paper presents an overview of the swimming mechanisms employed by fish. The motivation is to provide a relevant and useful introduction to the existing literature for engineers with an interest in the emerging area of aquatic biomechanisms. The fish swimming types are presented, following the well-established classification scheme and nomenclature originally proposed by Breder. Fish swim either by body and/or caudal fin (BCF) movements or using median and/or paired fin (MPF) propulsion. The latter is generally employed at slow speeds, offering greater maneuverability and better propulsive efficiency, while BCF movements can achieve greater thrust and accelerations. For both BCF and MPF locomotion, specific swimming modes are identified, based on the propulsor and the type of movements (oscillatory or undulatory) employed for thrust generation. Along with general descriptions and kinematic data, the analytical approaches developed to study each swimming mode are also introduced. Particular reference is made to lunate tail propulsion, undulating fins, and labriform (oscillatory pectoral fin) swimming mechanisms, identified as having the greatest potential for exploitation in artificial systems.

/pdf/review-of-fish-swimming-modes-for-aquatic-locomotion-12dkkh88vr.pdf

Review of fish swimming modes for aquatic locomotion

Efficient path-planning algorithms are a crucial issue for modern autonomous underwater vehicles. Classical path-planning algorithms in artificial intelligence are not designed to deal with wide continuous environments prone to currents. We present a novel Fast Marching (FM)-based approach to address the following issues. First, we develop an algorithm we call FM* to efficiently extract a 2-D continuous path from a discrete representation of the environment. Second, we take underwater currents into account thanks to an anisotropic extension of the original FM algorithm. Third, the vehicle turning radius is introduced as a constraint on the optimal path curvature for both isotropic and anisotropic media. Finally, a multiresolution method is introduced to speed up the overall path-planning process

/pdf/path-planning-for-autonomous-underwater-vehicles-17pc2wefaj.pdf

Path Planning for Autonomous Underwater Vehicles

This paper describes a new framework for segmentation of sonar images, tracking of underwater objects and motion estimation. This framework is applied to the design of an obstacle avoidance and path planning system for underwater vehicles based on a multi-beam forward looking sonar sensor. The real-time data flow (acoustic images) at the input of the system is first segmented and relevant features are extracted. We also take advantage of the real-time data stream to track the obstacles in following frames to obtain their dynamic characteristics. This allows us to optimize the preprocessing phases in segmenting only the relevant part of the images. Once the static (size and shape) as well as dynamic characteristics (velocity, acceleration,...) of the obstacles have been computed, we create a representation of the vehicle's workspace based on these features. This representation uses constructive solid geometry (CSG) to create a convex set of obstacles defining the workspace. The tracking takes also into account obstacles which are no longer in the field of view of the sonar in the path planning phase. A well-proven nonlinear search (sequential quadratic programming) is then employed, where obstacles are expressed as constraints in the search space. This approach is less affected by local minima than classical methods using potential fields. The proposed system is not only capable of obstacle avoidance but also of path planning in complex environments which include fast moving obstacles. Results obtained on real sonar data are shown and discussed. Possible applications to sonar servoing and real-time motion estimation are also discussed.

/pdf/underwater-vehicle-obstacle-avoidance-and-path-planning-2wtx04hicu.pdf

Underwater vehicle obstacle avoidance and path planning using a multi-beam forward looking sonar

An Intervention-AUV (or I-AUV), is a hover capable AUV whose primary role is direct contact with subsea structures for measurement or physical manipulation of components. The aim of the ALIVE project is to develop an Intervention-AUV capable of docking to a subsea structure which has not been specifically modified for AUV use. This paper describes the modular structure of the ALIVE AUV, including its distributed software architecture and in particular the ADS (Autonomous Docking System). It then outlines the sonar and video sensor processing techniques used for real-time control of the AUV to perform tracking and 3D pose reconstruction. In addition, details of the system tests and practical trials used in the development process are described.

Autonomous docking for Intervention-AUVs using sonar and video-based real-time 3D pose estimation

This paper describes and evaluates a concurrent mapping and localization (CML) algorithm suitable for localizing an autonomous underwater vehicle. The proposed CML algorithm uses a sidescan sonar to sense the environment. The returns from the sonar are used to detect landmarks in the vehicle's vicinity. These landmarks are used, in conjunction with a vehicle model, by the CML algorithm to concurrently build an absolute map of the environment and to localize the vehicle in absolute coordinates. As the vehicle moves forward, the areas covered by a forward-look sonar overlap, whereas little or no overlap occurs when using sidescan sonar. It has been demonstrated that numerous reobservations by a forward-look sonar of the landmarks can be used to perform CML. Multipass missions, such as sets of parallel and regularly spaced linear tracks, allow a few reobservations of each landmark with sidescan sonar. An evaluation of the CML algorithm using sidescan sonar is made on this type of trajectory. The estimated trajectory provided by the CML algorithm shows significant jerks in the positions and heading brought about by the corrections that occur when a landmark is reobserved. Thus, this trajectory is not useful to mosaic the sea bed. This paper proposes the implementation of an optimal smoother on the CML solution. A forward stochastic map is used in conjunction with a backward Rauch-Tung-Striebel filter to provide the smoothed trajectory. This paper presents simulation and real results and shows that the smoothed CML solution helps to produce a more accurate navigation solution and a smooth navigation trajectory. This paper also shows that the qualitative value of the mosaics produced using CML is far superior to those that do not use it.

David M. Lane

Papers

Review of fish swimming modes for aquatic locomotion

Path Planning for Autonomous Underwater Vehicles

Underwater vehicle obstacle avoidance and path planning using a multi-beam forward looking sonar

Autonomous docking for Intervention-AUVs using sonar and video-based real-time 3D pose estimation

Concurrent Mapping and Localisation using Side-scan Sonar