M. Seto – Marine Robot Autonomy (2013)

Автор: M. Seto
Название книги: Marine Robot Autonomy (2013)
Формат: PDF
Жанр: Водный транспорт
Страницы: 390
Качество: Изначально компьютерное, E-book

Autonomy for Marine Robots provides a timely and insightful overview of intelligent autonomy in marine robots. A brief history of this emerging field is provided, along with a discussion of the challenges unique to the underwater environment and their impact on the level of intelligent autonomy required. Topics covered at length examine advanced frameworks, path-planning, fault tolerance, machine learning, and cooperation as relevant to marine robots that need intelligent autonomy.

This book provides an update to select underwater autonomy areas. It is intended
for researchers and engineers who are new to the field of marine robot autonomy, at
the same time, appealing to those with more experience. This book was inspired by:
(1) quite a few researchers looking for a reference for graduate courses in marine
autonomous robotics with emphasis on autonomy and (2) researchers and engineers
who are new to the area with little or no formal training or experience in the area. It
is hoped that the extensive references deliberately compiled by each chapter author
provide a valuable starting point for further study.
The introductory chapter sets the background and provides definitions for
subsequent chapters. It starts with a motivation for why autonomy is necessary and
timely for marine robots. Then, it briefly reviews existing metrics and standards
for marine autonomy and the components that exist in many intelligent autonomy
architectures. The autonomy requirements of military and oceanographic users are
touched upon and briefly contrasted to give the reader an appreciation for two
different users of marine robots. Then, the fundamentals of what limits marine
autonomy specifically, the underwater medium, underwater navigation, and critical
enablers like energy are introduced. In-depth exploration is left to subsequent
chapters; however, tie-ins and developments related to the following chapters are
noted.
To start, two different intelligent autonomy architectures are highlighted, the
behaviour-based MOOS-IvP (Benjamin et al., Chap. 2) and the deliberative T-REX
(Rajan et al., Chap. 3). These architectures are implemented on actual systems and
have received a fair amount of interest from the military and scientific users.
One of the basic functionalities of an intelligent autonomy architecture (for
all environments) is motion or path planning. This occurs once a map of the
environment exists. Plans optimized around constraints have to be generated. For
reactive applications as in obstacle avoidance, plans are generated in near real time.
For more deliberative search applications where a detailed digital map of the area
exists this can take longer. Path planning is discussed in Chap. 4 (Paull et al.).
UUVs on long deployment require fault tolerance as the UUV itself will change
if the mission is long enough. When unexpected hardware failures occur, the
intelligent autonomy should allow the UUV to reconfigure itself to use alternative
combinations of the remaining functioning resources (Lane et al., Chap. 5). This
has been termed “autonomous embedded recoverability.” Chapter 5 describes work
on a declarative goal-based solution for adaptive mission planning that builds in the
ability to adapt and recover from failures.
The ability to scan or sense a wider area and to work cooperatively has the
potential to vastly improve the efficiency and effectiveness of mission operations.
However, given the complexity and difficulty of the underwater environment,
cooperation between underwater vehicles faces many challenges. This is covered
in Chap. 6 (Redfield).
In the very dynamic ocean environment where operators work with little or
no a priori information the value of machine learning emerges. Reinforcement
learning is a methodology in robot learning where a scalar evaluation (reward) of
the performance of the algorithm is available from interaction with the environment.
The objective in reinforcement learning is to maximize the expected reward through
adjusting a value function. The role of machine learning in an intelligent autonomy
architecture is highlighted in Chap. 7 (Carreras et al.).
SLAM is an example of a truly autonomous capability with little or no human
intervention. With SLAM, a spatial map of the UUV’s environment is built for
navigation purposes. The UUV uses its sensors (sonars, bathymetric sensors, etc.)
to perceive the environment. The sensors are modelled as real with errors and finite
ranges. The sensor measurements are assembled to create the map. There is no a
priori environment information for the UUV to work with. With SLAM, beacons
and networks do not have to be deployed or used making the technique useful
in GPS-denied, underwater, or under-ice environments. Chapter 8 showcases this
capability through three case studies.
I would like to acknowledge the efforts of the many reviewers who have given
generously their time. Each chapter was meticulously peer reviewed. Lastly, I would
like to thank all the authors who have contributed to this book. They are among the
researchers who are expanding the envelope on the state of the art in autonomy for
marine robots.