A Robot Web for Distributed Many-Device Localisation
Transactions on Robotics, 2024
1. Imperial College London, 2. Toronto Metropolitan University
Abstract
We show that a distributed network of robots or other devices which make measurements of each other can
collaborate to globally localise via efficient ad-hoc peer to peer communication. Our Robot Web solution is
based
on Gaussian Belief Propagation on the fundamental non-linear factor graph describing the probabilistic
structure
of all of the observations robots make internally or of each other, and is flexible for any type of robot,
motion
or sensor. We define a simple and efficient communication protocol which can be implemented by the
publishing and
reading of web pages or other asynchronous communication technologies. We show in simulations with up to
1000
robots interacting in arbitrary patterns that our solution convergently achieves global accuracy as accurate
as a
centralised non-linear factor graph solver while operating with high distributed efficiency of computation
and
communication. Via the use of robust factors in GBP, our method is tolerant to a high percentage of faults
in
sensor measurements or dropped communication packets.
Summary Video
Simulation Results
Joining Robot Web
Robot Web is extremely dynamic and allows for the addition of new robots at any
time. The robots have no prior information about their initial poses but are still able to
accurately localise.
Outlier Measurements
Robot Web can handle extreme outliers using robust factors.
Here, 30% of the sensor measurements are garbage.
Robots automatically reject the garbage measurements (blue lines) and downweights them in the
optimisation.
Real-world Results
With landmarks
Without landmarks
We evaluate Robot Web running on real robots. Here, we demonstrate that Robot Web can relocalise the
robots
even
if we pick and place them in a random place.
BibTex
@article{Murai:etal:TRO2024,
title={A {R}obot {W}eb for {D}istributed {M}any-{D}evice {L}ocalisation},
author={Murai, Riku and Ortiz, Joseph and Saeedi, Sajad and Kelly, Paul HJ and Davison, Andrew J},
journal={IEEE Transactions on Robotics},
year={2024},
volume={40},
number={},
pages={121-138},
publisher={IEEE}
}
Acknowledgements
We are grateful to many researchers with whom we have discussed some of the ideas in this paper, especially
from
the Dyson Robotics Lab and Robot Vision Group at Imperial College, and SLAMcore. We would particularly like
to
thank Aalok Patwardhan, Marwan Taher, Hussein Ali Jaafar, Stefan Leutenegger, Raluca Scona, Talfan Evans,
Eric
Dexheimer, Seth Nabarro, Mark van der Wilk, Owen Nicholson, Rob Deaves, Pablo Alcantarilla, Jacek
Zienkiewicz,
Amanda Prorok, Mac Schwager, Frank Dellaert, Michael Kaess, Tim Barfoot, Richard Newcombe.
This work was partially supported by the EPSRC (EP/K008730/1 and EP/P010040/1).