Background
Mapping Without Odometery
Robotic odometery is not 100% reliable. In this study, I take it to
the next level and simulate a multi-robot mapping strategy with robots
that have no odometery sensors at all. The general and most intuitive
approach to multi-robot mapping without odometery is to have a one
robot act as a static observer that knows its pose, and use sensors to
detect its moving & scouting counterparts and update their pose. Once
a scout leaves the sensor range of the observer, it becomes an
observer with its known pose and updates the other robot as it catches
up.
The main challenge of this project is determining another robot's
relative pose and relating that to world coordinates given the
observing robot's pose.
Mapping Without Odometery with 3+ Robots
This can be scaled with multiple robots with a number of inter-robot
behavior strategies. We could do any of the following:
- Have multiple scouts per static observer - this could
increase the speed of mapping, but if the scouts do not coordinate,
they could spread out too far and the observer would not know which
one to follow.
- Team up (even number of robots) in a buddy system - this
would work well in a divide & conquer strategy and make the two teams
completely independent, but it requires even numbers of robots, which
isn't always available.
- Have a dynamic scout, a static observer, and any number of
hybrid robots in between - the Gravy Train approach, I like
to call it. This allows more robust behavior, since [ideally] we can
span more space between the first scout and the last observer and get
greater reach without having to use explicit catch-ups. Moreover,
once we reach a corner of the map, we call a turnaround and flip roles
& ordering to conquer another part of themap.
The Gravy Train approach intrigued me the most, though it is
the most difficult to implement. I saved my work after coming up with
a functional & roboust 2-robot strategy.
All
material & intellectual property on this site
was gathered by
Scott Jason Friedman for CS553
at Washington University in
St. Louis, 2004.
|