How to fuse odometry and AR slam using robot_localization?

edit

I am trying to use robot_localization to fuse wheel odometry and fiducial_slam and I am a little confused by the transforms.

In the current system, we have the wheel odometry publishing odom->base_link, and fiducials publishing map->odom.

When I run the EKF in this system (with publish_tf: false) and look at the /odom/filtered topic in rviz, it looks great, the kalman filter pose seems to track the robot pose in the real world very well.

The problem comes when I turn off the map->odom publishing from fiducials, and turn on publish_tf for robot_localization. This causes the position of the robot to fly off wildly every time we move.

I think this is because the wheel odometry data is published relative to the /odom frame, but there is no way to transform this into the /map frame without the kalman filter using it's own output, and this cyclical flow is causing wild instability.

How should I go about trying to solve this issue? How do you fuse map relative and odom relative pose measurements.

Edit: localization config

frequency: 10
two_d_mode: true

map_frame: map
odom_frame: odom
base_link_frame: base_footprint
world_frame: map

publish_tf: true

#  x, y, z,
#  roll, pitch, yaw,
#  vx, vy, vz,
#  vroll, vpitch, vyaw,
#  ax, ay, az

odom0: /odom
odom0_config: [true,  true,  false,
               false, false, true,
               false, false, false,
               false, false, true,
               false, false, false]
odom0_differential: false
odom0_relative: true

#imu0: /imu
#imu0_config: [false, false, false,
#              false, false, true,
#              false, false, false,
#              false, false, true,
#              false, false, false]
#imu0_differential: true
#imu0_relative: true

pose0: /fiducial_pose
pose0_config: [true,   true,   false,
               false,  false,  true,
               false,  false,  false,
               false,  false,  false,
               false,  false,  false]

pose0_rejection_threshold: 3

#  x,    y,    z,    roll, pitch, yaw, vx,  vy,    vz,  vroll, vpitch, vyaw, ax,   ay,  az
process_noise_covariance: [0.05, 0,    0,    0,    0,    0,    0,     0,     0,    0,    0,    0,    0,    0,    0,
                           0,    0.05, 0,    0,    0,    0,    0,     0,     0,    0,    0,    0,    0,    0,    0,
                           0,    0,    0.00, 0,    0,    0,    0,     0,     0,    0,    0,    0,    0,    0,    0,
                           0,    0,    0,    0.00, 0,    0,    0,     0,     0,    0,    0,    0,    0,    0,    0,
                           0,    0,    0,    0,    0.00, 0,    0,     0,     0,    0,    0,    0,    0,    0,    0,
                           0,    0,    0,    0,    0,    0.01, 0,     0,     0,    0,    0,    0,    0,    0,    0,
                           0,    0,    0,    0,    0,    0,    0.000, 0,     0,    0,    0,    0,    0,    0,    0,
                           0,    0,    0,    0,    0,    0,    0,     0.000, 0,    0,    0,    0,    0,    0,    0,
                           0,    0,    0,    0,    0,    0,    0,     0,     0.00, 0,    0,    0,    0,    0,    0,
                           0,    0,    0,    0,    0,    0,    0,     0,     0,    0.00, 0,    0,    0,    0,    0,
                           0,    0,    0,    0,    0,    0,    0,     0,     0,    0,    0.00, 0,    0,    0,    0,
                           0,    0,    0,    0,    0,    0,    0,     0,     0,    0,    0,    0.01, 0,    0,    0,
                           0,    0,    0,    0,    0,    0,    0,     0,     0,    0,    0,    0,    0.00, 0,    0,
                           0,    0,    0 ...