Revision history [back]

Information about the LIDAR attitude has to be available. If the platform experiences roll/pitch motion, using a IMU and tf is a straightforward way to supply this information. Of course, if the LIDAR is mounted to a heavy platform traveling on flat ground, the roll/pitch estimate can be considered fixed and no IMU is needed.

Low cost IMUs as commonly used in robotics (as opposed to military grade/aerospace ones) are hard to use for odometry, as the double integration of accelerometer data leads to very high translational errors within seconds. For this reason, IMU data generally has to be filtered and fused with other sources to be useful for (translational) motion estimation. A EKF-based filter for this is available in the hector_localization stack. Unfortunately we didn't get around to writing a proper tutorial for that so far. This state estimation filter estimates the 6DOF pose of the vehicle (as well as IMU sensor biases). The tf generated by the filter can be used as a start estimate for scan matching in hector_mapping, while the 2D pose estimate by the latter in turn can be used to update the EKF solution again.