Octomap results


I tested parrot and octomap under bridge. Why octomap and pose works only when the S.L.A.M.dunk is no further than ~2meters from the pillar? It doesn’t work properly.
I attach two bag files: first contains data from the raid close to the pillars and the second from the distance from them.
Can you look on the data and tell me your conclusions?




From what I can see from the bad octomap bag, the odometry never actually starts:
The “SLAM” quality stays at Unknown.
The unknown quality means that the algorithm computing the pose is starting up. In that phase, the only pose information given is orientation, no position. So it cannot build an octomap if it considers it’s always in the same point. And I believe the octomap doesn’t start while it’s in that mode anyway…

The slamdunk needs some time staying still to be able to get the pose started up, a few seconds should suffice but if it moves before it’s initialised it won’t be able to.

That being said, maybe you did let it settle for a while before taking off and the slamdunk got lost for some other reason I don’t know.

By the way, if you want to capture bag files, I would advise you to record the /tf topic (and maybe the depthmap as well) so we can play the pointCloud topics too and get the quality of the depth reconstruction…


Thanks for your answer. I realised that I didn’t record /tf topic after tests. It’s my fault. So you advice me to wait until /quality is set up? But what I should do if odometry doesn’t start after one minute?
Best regards!


I would advice to wait until quality goes up yes.

But it should not take more than, say, 15s to set up.
If it doesn’t start after a while, you could have it looking at stuff with a lot of features on it (contrast, colors,… not a plain white wall :slight_smile: ) and not too far from it (in the order of a few meters).
I guess you could use the ground for that if you’re far away from obstacles.

And, if it really doesn’t start, well, there is something wrong…

I’ve never used it outdoors so I don’t know what its behaviour should be when there are features to be seen to compute but they are far away…