Wim, I'm not sure I understand this rationale.  Somewhere someone is 
chaining all the transforms and computing the shallow tree.  Why not 
just publish a more "natural" tree, along with the pre-computed 
transforms that go into the shallow tree?  To the user, the tree would 
make more sense, and computational efficiency would be preserved.

Cheers --Kurt

Wim Meeussen wrote:
>> Also, I noticed that, at least for the PR2 simulator, the PR2's tf
>> tree is very shallow with base_footprint being the parent frame to
>> a large number of frames. I would have thought that tf tree would
>> closely mimic the physical degree of freedom heirarchy in the PR2
>> (i.e. base->torso->upper_arm->lower_arm, etc.) Is there a reason
>> for the PR2s shallow tf tree?
> 
> The shallow tree allows you to compute the pose between any two links
>  of the pr2, by chaining only two transforms.  If the tf tree would 
> mimic the physical degree of freedom heirarchy in the pr2, you'd
> often have to chain 10+ transforms to get the pose between two links.
> So this is a performance optimization.
> 
> Wim
> 
>