High latency on Pyrene

[jmirabel]

The experiment yesterday revealed a latency of 1 second of the latest transform of the object wrt the hand, as measured in the image, in TF.

This latency should be investigated. Here is a summary of the information flow.

• image is taken at timestamp T,
• agimus-vision processes the images and returns a result. The result has the timestamp of the image, thus T,
• agimus-demos/scripts/group_of_tags.py gets all the estimated tag poses of one body and publishes an estimate of its pose in TF. The timestamp should still be T.
• SoT call TF lookup_transform to get the latest transform.

At the moment, a transform is discarded if it is older than 300ms. On my computer, visual servoing works so I guess the latency is lower than this. Here are some possible reasons:

• network delays,
• slow TF lookups,
• compilation with wrong flags,
• merge agimus-demos/scripts/group_of_tags.py into agimus-vision.

I will increase the list when I see new possible reasons.