the simulated IMU sends relatively bad data (e.g. the z acceleration isn't fixed at 9.81 but fluctuates wildely around that number). however, so far the covariance in the IMU data was always set to 0.0 which meant that the ekf_localization_node thought that the data was perfect and used it as the primary source, thus leading to wrong position estimates for the odom frame.
by setting the gaussianNoise it now sets a covariance and the localization algorithm gives the IMU data slightly less weight.
the number 0.5 was arrived at experimentally, it seems to give a good result: this allows mapping the whole map in gazebo without any noticeable drift.
the simulated IMU sends relatively bad data (e.g. the z acceleration isn't fixed at 9.81 but fluctuates wildely around that number). however, so far the covariance in the IMU data was always set to 0.0 which meant that the
ekf_localization_nodethought that the data was perfect and used it as the primary source, thus leading to wrong position estimates for theodomframe. by setting thegaussianNoiseit now sets a covariance and the localization algorithm gives the IMU data slightly less weight.the number 0.5 was arrived at experimentally, it seems to give a good result: this allows mapping the whole map in gazebo without any noticeable drift.