search

carologistics / fawkes-robotino

Fawkes for the RCLL Carologistics Flavor
https://www.carologistics.org
9 stars 0 forks source link

conveyor-pose: remember tag->conveyor TF #60

Open vmatare opened 5 years ago

vmatare commented 5 years ago

We might be able to significantly speed up conveyor gripping by saving the ICP pose of the conveyor relative to the tag. That way, it might be possible to just grip "blindly" relative to the tag after we have detected the conveyor once.

nlimpert commented 5 years ago

Great idea! I guess it would also simply suffice to just store the transform relative to the tag.

This would also allow us to already move the gripper while we are approaching the MPS. :+1:

Obviously we would have to make sure that the tag remains in sight even when we are close to the MPS (see https://github.com/carologistics/hardware/issues/3).

timn commented 5 years ago

So you essentially propose to make your approach a calibration tool, a nice one, which can be used either before or during the game. That is perfectly reasonable if it is too slow to run it online.

However, consider for a moment why we started the conveyor recognition in the first place: inaccuracies in the marker localization, and in the lateral measurement based on laser line data alone. So, a question that should come up is whether the marker localization (you are suggesting to use only that in your proposal) is reliable and accurate enough, and if I remember correctly those two properties were not independent of each other in the past (depending on the position a detection might have a wrong orientation or some offset, a lot of work in development and during tournaments went into tuning the respective skill parameters in the past).

All this might have changed, since small motions can now be done by means of the gripper axes alone and do not require the rather imprecise movement of the whole base. Still, the lateral placement tolerance of the conveyors has not increased.

Just a few thoughts and maybe a warning, that this is something that might require more development and testing than anticipated (I would have loved to read a more elaborate proposal where some of the benefits and risks would have been analyzed). My two cents.

morxa commented 4 years ago

Another idea would be to save the transform and use it for the initial estimate of the next run.