On 10/05/2012 05:20 PM, Thomas Moulard wrote:
> Just one note: I would not have called this "humanoid" as it seems
> that the component is totally
> generic and could be used in other context (i.e. drones, for instance).
It could be used, but only with a few modifications since there are a 
some assumptions about humanoids (which you mentioned below). The core 
is certainly generic and could be turned into a general 6D localization 
approach. It could then end up in a new, more general package. I didn't 
want to promise too much with a too generic package name and no robot 
available to test except our humanoids.

>
> On the opposite, one thing that would be interesting and humanoid
> specifics would be to consider that at all point of time, the robot is
> not falling, i.e. a planar contact with the floor occurs and prevent
> some d.o.f. from drifting (z, rx and ry). I am not sure however, how
> to integrate this with your current work...
That's already considered, although not enforcing it strictly but as 
probabilistic measurements in the filter (z from kinematics, roll and 
pitch from odometry & IMU). All of these can have errors, and the floor 
level can also be imprecise and suffer from discretization effects etc. 
For strictly enforcing it, the standard deviation of the noise can 
simply be set to zero.

Cheers,
Armin

-- 
Armin Hornung
Humanoid Robots Lab, Albert-Ludwigs-Universität Freiburg
Contact: http://www.informatik.uni-freiburg.de/~hornunga