Description
The majority of the earth's surface is covered by the openocean. However, harbors and marine structures produce areas
that constrain the possible paths that can be traversed by a
maritime vehicle. For a vehicle to perform an efficient set of
actions in such an environment, a plan must be assembled that
respects these fundamental spatial constraints. The complexity
of the planning task can be further increased if the environment
is partially observed, requiring the fusion of online sensor data
into the planning domain.
This talk will discuss an approach for planning tasks in spatially
constrained environments based on the Delaunay Triangulation.
This model allows the construction of a planning domain based
on the fundamental group – the set of homotopy classes as
constrained by the environment.
Initially developed for the Maritime RobotX competition, this
system allows the generation of plans fusing offline spatial data
with online sensor observations, and respecting spatial
constraints. While the complexity of adding an action to the plan
is higher than a system based on a grid metric, the planning
complexity in this domain scales with the number of obstacles
rather than the size of the spatial environment. The small size
of the planning domain thus allows the generation of plans to
scale beyond what would be possible in a grid-based domain.
An overview of the potential applications along with some of
Flinders University's recent work in Autonomous Maritime
Vehicles will also be covered.
| Period | 26 May 2022 |
|---|---|
| Held at | IEEE Computing Society, South Australian Chapter, Australia |
| Degree of Recognition | Local |