Autonomy: Risk Assessment 24.3 Risk of Collision 533
Part B | 24.3
hazards are usually well known, and fixed. The onus
is on the autonomous vehicle operator to plan missions with sufficient contingency and allowance for
navigational error such that a vehicle does not infringe the exclusion zones around these structures.
Coastal structures such as breakwaters, or moles:
While there may be no exclusion zones around these
hazards, their positions are fixed, and appropriate
prior planning should be adequate to mitigate the
risks.
24.3.2 Risks of Collision Underwater
Apart from the universal case of collision with the
seabed, the other risks of collision when underwater depend on the environment of the missions. Near-shore
missions may encounter obstacles such as piers and
other fixed structures, vessels at anchor, and the submerged parts of vessels underway. In the open ocean,
apart from with the seabed, collision when submerged
is likely to be a rare event. Under ice, collision is likely
with ice at the sea surface, with projecting keels of sea
ice, and with icebergs. While grounded icebergs, ice
shelves and glaciers and sea ice that is locked to the
shore or a shoal may be stationary, other forms of ice
should be considered as moving obstacles.
Collision, or obstacle, avoidance for AUVs is a wellstudied topic, combining as it does the challenges of
sensing, interpretation, and action. The effectiveness of
obstacle avoidance is one indicator of the degree of autonomy of a vehicle. Horner et al. [24.22] summarized
these challenges, and provided examples of how a forward-look sonar array can be integrated with an autopilot control algorithm that minimizes the cross-track error from the desired path. That is, the system provides
for adequate, but minimum, deviation from the desired
path to avoid the obstacle. The interpretation of the obstacle characteristics in this implementation uses image
information from the sonar array, borrows techniques
from computer vision, and can be made robust against
false targets. This approach is relevant for avoiding the
seabed, but is also sufficiently powerful to provide obstacle avoidance in more complex situations, such as
near piers and other man-made structures.
There are situations where a simpler approach may
give sufficient capability, for example, where the seabed
is the only obstacle likely to be encountered. Using a single beam mechanically scanned sonar and
a horizon-tracking algorithm, McPhail et al. [24.23]
showed that effective seabed avoidance could be delivered when terrain following at altitudes down to 10 m
on the flanks of a seamount, and down to 3 m on the
flatter, but rocky, summit. Vehicle reactions included
simple avoidance in the vertical plane, and turn-around
and retry.
Under ice, information is needed on obstacles
ahead, below, and above the vehicle. One approach,
devised for the DEPTHX (deep phreatic thermal explorer) vehicle [24.24] proposed a suite of 24 narrow
beam sonars and 30 imaging sonars. In the subsequent
Endurance vehicle, 64 pencil beam sonars provided
a 3:5 steradian view of the environment within icecovered Lake Bonney in Antarctica. A far simpler
approach is to use for obstacle avoidance purposes,
sonar information from other instruments on the vehicle. Pebody has described how, for an under ice application, a forward-looking sonar may be augmented using
range to reflector information from the four beams of
an upward-looking and a downward-looking acoustic
Doppler velocity log [24.25]. This system was tested
using an air curtain behind a ship, to simulate an iceberg, and used under fast ice off Greenland, and under
Fimbulisen and Pine Island Glacier in Antarctica to
good effect.
As a backup to collision avoidance, some operators include rubber fenders or other protection on the
nose and undersides of their vehicles. Incidents where
collision avoidance has failed include repercussions of
hardware failures and failures of the control systems to
properly appreciate the form of the environment and
hence failure to take proper action. Drawing on examples with Autosub AUVs, a sternplane fault led to
the vehicle diving far too steeply, such that its bottom avoidance sonar was at the incorrect angle, the
sonar could not see the seabed, the vehicle nose collided with the seabed but the vehicle continued to make
slow forward progress until the nose became full of
sediment. For an example of inappropriate response to
an unexpected environment, when terrain following at
low altitude up a steep cliff off Sicily, Autosub2 encountered an overhang. The appropriate response would
have been to reverse out, but this behavior was not preprogrammed, and instead the vehicle dropped its abort
weight and had to be retrieved using an remotely operated vehicle. These lessons also serve to show that not
all collisions are fatal to a vehicle.
Part B | 24.3
hazards are usually well known, and fixed. The onus
is on the autonomous vehicle operator to plan missions with sufficient contingency and allowance for
navigational error such that a vehicle does not infringe the exclusion zones around these structures.
Coastal structures such as breakwaters, or moles:
While there may be no exclusion zones around these
hazards, their positions are fixed, and appropriate
prior planning should be adequate to mitigate the
risks.
24.3.2 Risks of Collision Underwater
Apart from the universal case of collision with the
seabed, the other risks of collision when underwater depend on the environment of the missions. Near-shore
missions may encounter obstacles such as piers and
other fixed structures, vessels at anchor, and the submerged parts of vessels underway. In the open ocean,
apart from with the seabed, collision when submerged
is likely to be a rare event. Under ice, collision is likely
with ice at the sea surface, with projecting keels of sea
ice, and with icebergs. While grounded icebergs, ice
shelves and glaciers and sea ice that is locked to the
shore or a shoal may be stationary, other forms of ice
should be considered as moving obstacles.
Collision, or obstacle, avoidance for AUVs is a wellstudied topic, combining as it does the challenges of
sensing, interpretation, and action. The effectiveness of
obstacle avoidance is one indicator of the degree of autonomy of a vehicle. Horner et al. [24.22] summarized
these challenges, and provided examples of how a forward-look sonar array can be integrated with an autopilot control algorithm that minimizes the cross-track error from the desired path. That is, the system provides
for adequate, but minimum, deviation from the desired
path to avoid the obstacle. The interpretation of the obstacle characteristics in this implementation uses image
information from the sonar array, borrows techniques
from computer vision, and can be made robust against
false targets. This approach is relevant for avoiding the
seabed, but is also sufficiently powerful to provide obstacle avoidance in more complex situations, such as
near piers and other man-made structures.
There are situations where a simpler approach may
give sufficient capability, for example, where the seabed
is the only obstacle likely to be encountered. Using a single beam mechanically scanned sonar and
a horizon-tracking algorithm, McPhail et al. [24.23]
showed that effective seabed avoidance could be delivered when terrain following at altitudes down to 10 m
on the flanks of a seamount, and down to 3 m on the
flatter, but rocky, summit. Vehicle reactions included
simple avoidance in the vertical plane, and turn-around
and retry.
Under ice, information is needed on obstacles
ahead, below, and above the vehicle. One approach,
devised for the DEPTHX (deep phreatic thermal explorer) vehicle [24.24] proposed a suite of 24 narrow
beam sonars and 30 imaging sonars. In the subsequent
Endurance vehicle, 64 pencil beam sonars provided
a 3:5 steradian view of the environment within icecovered Lake Bonney in Antarctica. A far simpler
approach is to use for obstacle avoidance purposes,
sonar information from other instruments on the vehicle. Pebody has described how, for an under ice application, a forward-looking sonar may be augmented using
range to reflector information from the four beams of
an upward-looking and a downward-looking acoustic
Doppler velocity log [24.25]. This system was tested
using an air curtain behind a ship, to simulate an iceberg, and used under fast ice off Greenland, and under
Fimbulisen and Pine Island Glacier in Antarctica to
good effect.
As a backup to collision avoidance, some operators include rubber fenders or other protection on the
nose and undersides of their vehicles. Incidents where
collision avoidance has failed include repercussions of
hardware failures and failures of the control systems to
properly appreciate the form of the environment and
hence failure to take proper action. Drawing on examples with Autosub AUVs, a sternplane fault led to
the vehicle diving far too steeply, such that its bottom avoidance sonar was at the incorrect angle, the
sonar could not see the seabed, the vehicle nose collided with the seabed but the vehicle continued to make
slow forward progress until the nose became full of
sediment. For an example of inappropriate response to
an unexpected environment, when terrain following at
low altitude up a steep cliff off Sicily, Autosub2 encountered an overhang. The appropriate response would
have been to reverse out, but this behavior was not preprogrammed, and instead the vehicle dropped its abort
weight and had to be retrieved using an remotely operated vehicle. These lessons also serve to show that not
all collisions are fatal to a vehicle.