Revision history [back]

if you are having issues with transforms only appearing after a certain time. The usual problem is with your publishers not the listener. You should use view_frames and tf_monitor to debug your transforms sources. Debugging tutorial I'll also note that it's uncommon to want to take callbacks for every tf update. The best approach for holding data until it's available is to use the tf::MessageFilter

Subsidiary Question: tf messages are just a message, you can subscribe to them manually. However each message is only an incremental update which the listener aggregates and computes the net transform. You are looking up the first transform in a list of published transforms without checking the name so it's quite possible that the first transform in the list of transforms has value of 0 for the x translation component.

I recommend that you go through the full set of tf tutorials to understand how tf works better. http://wiki.ros.org/tf/Tutorials

Edit answering additional questions:

Interpolation is actually more accurate than taking data from the closest timestamp as long as you can make the assumption of continuous values and a publish frequency which is above the natural bandwidth of the physical system. And you make the assumption that publishing data for every timestep is much to high bandwidth.

Interpolation is necessary because tf is aggregating information from several asyncronous sources across a distributed network. The standard models for using tf are:

Timing based - you want to know the latest available information during you update loop. To do this simply query with Time(0). Or if you're on a specific schedule you can use a specific time query.
Data source driven - you have received some data and want to transform it into a different coordinate frame. If the data is in the buffer already you can simply lookup at the timestamp of the data. Doing interpolation to evaluate as accurately as possible each link of the chain to form the transform.
- In simple cases you can simply wait for the transform to become available (waitForTransform) (useful for simple scripts which are single purpose and can block.)
- If you are in a data intensive and latency intensive application you can buffer and hold the data until the required transform is available.
- this is what the tf::MessageFilter does for you using the onUpdate callbacks internally

You're first implementation is a partial implementation of the tf::MessageFilter however it does not queue data for later callbacks because you have a hard coded query. Not only that you are simply querying at time 0 so it is going to start reporting as fast as possible when the transform becomes available. But as you mention it does not solve you lag of 20 seconds.

Your second example is not querying the data that you want and does not demonstrate that the data is available. It demonstrates that there is some message coming over the tf topic. you do not even look at the frame_id of that transform. Please follow the previous suggestions of going through the tf Tutorials especially the debugging tutorials to understand what is going on in your system. I expect that your issues are coming from the publishing side and you need to debug that.