The idea would be that the user could store the training data with snapshot_bot, then drive one or more additional routes over roughly the same area and the estimated headings etc. could be examined offline, to see where the robot is likely to make mistakes.
@aphilippides Did this remain as an action item at the end of the meeting or are we happy with the analysis tools we have?
As discussed on the phone, I think it is still an action but I think for the offline analysis, it would involve comparing images taken from the training route as we don't have ground truth
The idea would be that the user could store the training data with
snapshot_bot
, then drive one or more additional routes over roughly the same area and the estimated headings etc. could be examined offline, to see where the robot is likely to make mistakes.@aphilippides Did this remain as an action item at the end of the meeting or are we happy with the analysis tools we have?