|dc.description.abstract||Current underwater robotic platforms rely upon waypoint-based scripted missions which
are described by the operator a-priori. This renders systems incapable of reacting to
the unexpected. In this thesis, we claim that the ability to autonomously adapt the
decision making process is the key to facilitating the change over from human intervention
to intelligent autonomy. We identify goal-based declarative mission planning
as an attractive solution to autonomous adaptability because it combines autonomous
decision making with higher levels of human interaction.
Goal-based mission planning requires the use of abstract knowledge representation
and situation awareness to link the prior knowledge provided by the operator with
the information coming from the processed sensor data. To achieve this, we propose
a semantic-based knowledge representation framework that allows this integration of
prior and processed information among all different agents available in the platform.
In order to evaluate adaptive mission planning techniques, we also introduce a novel
metric which measures the proximity between plans. We demonstrate that this metric
is better informed than previous metrics for measuring the adaptation process.
In this thesis we implement three different approaches to goal-based mission planning
in order to investigate which approach is most appropriate under different circumstances.
The first approach, continuous mission planning, focusses on long-term
deployment. This approach is based on a continuous re-assessment of the status of
the mission environment. Using our proximity metric, we evaluated this approach
and show that there is a high degree of similarity between our approach and the humanly
driven adaptation, both in a known static environment and in a partially-known
dynamic discoverable environment. The second, service-oriented mission planning,
makes use of the semantic framework to provide autonomous mission planning for
the dynamic discovery of the services published by the different agents in the system.
This allows platform independence, easing the manual creation of mission plans, and
robustness to changes. We show that this approach produces the same plans as the
baseline which was explicitly provided with the platform configuration. The last approach,
mission plan repair, handles the scenario where small changes occur in the
mission environment and there are limited resources for planning. We develop and
deploy a mission plan repair approach within a semantic-based autonomous planning
system in a real underwater vehicle. Experiments demonstrate that the integrated system
is capable of providing mission adaptation for maintaining the operability of the
host platform in the face of unexpected events.||en_US