I understand that odom_frame is only locally accurate, and that the transformation map -> odom -> base_link is what provides accurate pose information if it is ever needed. However, map -> odom -> base_link is subject to discrete jumps, so odom -> base_link is what is used when you need smooth data. This makes sense, but why would you ever need smooth, completely incorrect pose information? One possibility that I have thought of is that you never just use odom -> base link, but instead, when you need smooth pose for whatever reason, you use a single map -> odom -> base_link transformation at first, and then from there you ADD the odom -> base_link transformation to the single, static transformation you took earlier, and therefore essentially "reset" the odom from being completely inaccurate to once again being on the mark, after which it slowly diverges again. Is this the norm?
↧