Discuss Proper Method for Fusing Pose Data
NDT matching as implemented by @cvasfi is very time-sensitive and the Autoware.Auto architecture is designed to be data-driven, not time-driven. However, the EKF and UKF filter nodes from robot_localization
, which we are currently using to produce the odom
->base_link
transform estimate, produce their pose estimates at a steady frequency and add a new timestamp to the estimate. Since NDT matching is receiving (essentially) raw lidar scans which are stamped based on the data source and transform information which is stamped at regular intervals using the ROS clock, inconsistencies can (and do if any lag is introduced) occur. #431 (closed) was proposed as a solution but it essentially does away with the data-driven nature of the lidar scan data and "shoehorns" the scans into a new time reference frame.
The purpose of this discussion is to find a long-term solution to this problem. Once a solution has been reached, this issue will also track implementation.