IEEE Robotics & Automation Magazine - June 2019 - 96

contact with the inspected structure. For more complete
inspection or object recognition, 3D sensing can be achieved
using stereo camera systems and active structured light [15]
at short ranges.
Acoustic sensors are the other main class of payload sensors that can increase mission safety. For instance, echo
sounders can measure altitude from the bottom (using the
time-of-flight principle) and so help make sure UUVs keep
at a safe depth. Imaging sonars can be used for many applications and vary greatly in range and resolution. The synthetic
aperture sonar can reach
centimeter resolution, an
order of magnitude greatNavigation capabilities
er than conventional
sonars. It can be used for
can also be improved with
seabed mapping, mine
detection [16], or shipexteroceptive sensors
wreck surveying, as instrumental to the UUV
(optical or sonar) that
mission. Side-scan sonar,
also a down-looking sonar,
identify specific landmarks is used for similar applications but provides a
in the environment and use lower resolution. The multibeam forward-looking
them to localize the UUV.
sonars (FLSs) can be used
for obstacle avoidance,
object detection and recognition, and navigation, increasing mission safety. FLSs can
reach ranges of hundreds of meters and thus can help AUVs
avoid obstacles at a greater range than optical cameras. In
recent years, FLSs at higher frequencies (up to 3 MHz) have
produced images with millimeter resolution for very small
ranges (fewer than 10 m). These sonars can be used to safely
inspect ship hulls and chains or to detect divers or structural
damages at a close distance. The market now includes multibeam 3D high-resolution sonars. The most advanced 3D
sonar can provide real-time 3D images at 20 Hz [51]. These
sensors introduce very-high-resolution inspection possibilities at a safer distance, although their cost is still too high to
popularize them.
The challenges in sensing are mostly correlated to the
physical medium and, thus, hard to solve. Optical cameras
suffer from light attenuation and water turbidity, preventing
their use (without artificial lighting) below 30-40-m depth.

Artificial lighting systems can be used to improve their
usability at the cost of extra disturbances due to nonhomogeneous lighting. One way of solving these challenges is to
combine optical cameras with other sensors. For instance,
pairing them with lidar is becoming popular for high-resolution mapping and 3D reconstruction. Combining optical
cameras with sonars is another way of solving the optical
cameras' intrinsic challenges. Sonars, however, present their
own challenges. In particular, they need to be correctly calibrated, as sound propagation in water depends on temperature and salinity. Moreover, their resolution is typically lower
than that of optical cameras (except for short ranges),
although their ranges are typically greater. Acoustic shadows
and multipath artifacts are two other common issues, but
they can be mitigated and even explored for object detection.
Nonetheless, the fusion of optical and acoustic cameras
(high-resolution imaging sonars) is a way to solve the inherent challenges of each sensory modality. It can be used for
improving navigation (thus allowing safer operation), 3D
object reconstruction, inspection, and augmented reality
[17]. Nonetheless, it is a complex challenge still to be fully
solved, and a good calibration of the heterogeneous stereo
systems is required.
Another way of alleviating optical cameras' challenges is
the recent trend of using hyperspectral imaging sensors.
While these sensors have been used for many years in airplanes and satellites, only recently have they entered the
marine domain. These cameras can sense both within and
beyond the visible spectrum. Although they suffer absorption and scattering issues similar to those of optical cameras,
they provide better spectral resolution (with hundreds of
bands instead of a three-channel red-green-blue). This leads
to improved object identification and more efficient mapping [18].

Navigation, Guidance, and Control
A classical control architecture for UUVs is presented in Figure 3. Navigation uses the sensor data (and possibly external
input) to estimate position and velocity and feed the guidance
and control modules. Currently, UUVs largely rely on proprioceptive sensors, such as an inertial navigation system
(INS) integrated with Doppler velocity logs (DVLs). This is
because electromagnetic waves do not propagate well underwater, and, thus, global navigation satellite systems are not an
option. However, INS and DVL are subject to drift and biases,
leading to growing position uncertainty. Moreover, DVL usage is limited by
its range. The state of the art includes a
Desired Position
combination of internal sensors with
and/or Velocity
Forces
Mission
Guidance
Control
acoustic positioning systems from long
baseline (LBL) to short baseline (SBL)
Estimated Position
and ultrashort baseline (USBL).
Sensor
and Velocity
Data
Acoustic positioning systems, though,
Navigation
require careful calibration of the sound
velocity, as they suffer from multipath
Figure 3. The scheme of a navigation, guidance, and control architecture for a UUV.
Doppler effects and susceptibility to

IEEE ROBOTICS & AUTOMATION MAGAZINE

JUNE 2019

IEEE Robotics & Automation Magazine - June 2019

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