Open Magnusgaertner opened 7 years ago
Hi Magnus,
No, you didn't miss anything.
It's simply that I never had the motivation/time to implement landmarks-based rbpf-slam. It was on my "wish list" during a long time, but "state of the art" in SLAM quickly moved forward and attention for this kind of solutions dropped.
From what I recall, the optimal approach would be implementing the "FASTSLAM 2.0" algorithm (paper) as the "optimal PF algorithm".
However, a simpler working solution would be possible: just add support for insertObservation()
and computeObservationLikelihood()
in the landmarks map for the range-bearing observation.
As you can see in the observations/map matrix table, CObservationBearingRange
is shamelessly ignored by all metric maps right now (!!).
Implementing both methods above, the particle filter algorithm named "Standard Proposal" will work, despite not being optimal.
Cheers!
Hey Magnaes Gater,
Did you implement this ? :)))
nope, sorry...
Hi, the ekf_slam_2d wrapper already supports landmarks (2d points, no image landmarks, but identified objects with a large distance ~3meters from each other) as input, but other wrappers e.g rbpf_slam do not. How can I modify the wrappers to get landmarks working? I added a subscriber and callback for landmarks:
changed the config file to:
but now I am stuck at figuring out which map actually holds the landmarks, I tried:
However m_landmarksMap does not seem to be the right map as it does not process the landmarks at all. So my question is basically, what do I miss/ which map could I use to use the rbpf slam with non identifyable landmarks?
Best Regards Magnus