IEEE Robotics & Automation Magazine - June 2019 - 98

currently deployed embedded autonomy solutions are still
brittle and often require support from the operator. Specifically, operators have to deal with complexity in controlling multiobjective, multivehicle missions. They
simultaneously face uncertainty over the current status
and safety of several remote high-value assets. This is due
to the limited bandwidth and poor reliability of the communication channel. True autonomy solutions will thus
not be adopted by operators unless shared-autonomy
solutions are developed in which vehicles can explain
their decisions to operators and ask for help when needed,
and operators can query vehicles about their intentions
and support them [36]. This is especially critical for
remote operations, where long delays in communications
force the embedded autonomy to be in control of all realtime operations. In such cases, the remote supervisor can
only trigger behaviors to support the mission. If these
problems can be solved, long-term, persistent autonomy
will become possible-and, we believe, routine-for a
wide range of applications.
Manipulation
Underwater manipulation is a challenging problem addressed
in the literature from a control perspective [37]. Typical tasks
include valve-turning, pick-and-place, and plug-connector
operations. The state of the art at the industrial level is still
confined to master-slave robotic systems with the operator
constantly in the loop. Due to the specific difficulties in working in the marine environment, such as high pressure and
currents, this task remains tedious and expensive.
As mentioned previously, progress has been made since
the early 1990s in manipulation capabilities, advancing
from 1 DoF to current dual-arm free-floating manipulation. Operations such as valve turning or plugging in a connector have been demonstrated in laboratory experiments
[38]. Haptic-based control of a dual-arm system has been
implemented in the Ocean One project [13]; here, two
hands helped the operator in recovering an object from the
sea floor. From a research perspective, in underwater

manipulation, the trend is to provide the robotic system
with a sufficient level of autonomy while keeping the
human in the loop for high-abstraction-level decisions.
This has been achieved by the European project Dexterous
ROV [14], which used a satellite-based control architecture
able to handle intermittent or low-frequency communication. Its proof of concept used an architecture composed of
an on-shore learning system, able to recognize the user's
input instructed by an exoskeleton, and an off-shore twin
cognitive node in charge of reconstructing the desired endeffector movement [39].
Having fully autonomous manipulation in demanding
environments is one of the challenges that remain. To solve it,
the future of intervention robots needs to merge control with
issues coming from the communication and perception systems. This could provide the underwater system with sufficient autonomy to safely execute the operator's high-level
commands even in nonideal conditions.
Short Summary
As shown, the capabilities of UUVs have increased over the
years at the same time that their autonomy has grown. To
briefly summarize the differences among the various types
of vehicles and their evolution, we compare three popular
systems in Table 1. Early ROV systems were mostly teleoperated, with a low degree of autonomy and using black-andwhite cameras, but they included manipulation capabilities
that made them popular. Remote Environmental Monitoring UnitS (REMUS) AUV, originally built by the Woods
Hole Institute (like the Jason ROV), has been used as a lowcost vehicle designed for environmental monitoring since
1997, later becoming a commercial vehicle. In this AUV,
manipulation is not present, and the standard sensor suite is
related to the specific application. The AUV mission could
be changed in real time using acoustic communications.
More recently, I-AUVs such as the Girona 500 introduced
autonomous, free-floating manipulation while using simpler
deployment localization systems (USBL instead of LBL) and
a reconfigurable payload suite.

Table 1. The evolution of UUVs and their capabilities.

Autonomy and
Planning

Vehicle

Year

Sensing

Navigation

Communications

Jason ROV

1988

Side-scan sonar,
altimeter, black-andwhite camera

LBL, INS,
dynamic
positioning

Optical-fiber tether

Remotely controlled
and preplanned
track following

Teleoperated

REMUS AUV

1997

Acoustic Doppler Current Profiler, side-scan
sonar, conductivity
temperature profiler,
light scattering sensor

DVL, INS, LBL

Acoustic modem

Preplanned
missions and
acoustic commands

No manipulation
capabilities

Girona 500 I-AUV

2011

Profiler sonar,
side-scan sonar,
video camera

DVL, attitude
and heading
reference system, USBL

Acoustic modem
and optional
tether

Preplanned
missions and
autonomous
inspection

Autonomous
free-floating
manipulation

IEEE ROBOTICS & AUTOMATION MAGAZINE

JUNE 2019

Manipulation

IEEE Robotics & Automation Magazine - June 2019

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2019