Revision history [back]

Ok, may be very specific but for my case, it was because of delays in my laserscan pipeline:

• A first delay introduced by a Message Filter that cached some laserscan message to sync them in my scan merger. So when a new scan was received, the previous scan was typically outputed (so 1 sample of delay)
• The struct of my laser_odometry package was so that the the TF odom <-> base_link was computed and timestamped with the latest laserscan, but was only published 100ms later (a structure with a process() then a spin_some() that processed obsolete data)

Solving this two issues and reducing the delay, so that the timestamp of the TF remains close of now(), solved this behaviour for me.