How to use pose for `move_base` to produce `cmd_vel`?

edit

From the Wiki page of move_base, it seems that move_base obtains the pose estimation of a robot by a tf from map to odom, which is often published by amcl. However, in my application, I use a motion predictor to predict the future pose of the robot, and I want to use that future pose as a basis for my move_base computation (i.e. use the future pose instead of the current pose estimated by amcl).

The message of my future pose is geometry_msgs/PoseWithCovarianceStamped, which is not a tf. How do I control the robot using geometry_msgs/PoseWithCovarianceStamped instead of tf for move_base?

There are several solutions that I have tried:

1. One alternative is to replace the tf publisher in amcl with a custom tf publisher (from map to odom) using the future pose, but it causes the laser scan misaligned.

2. Another alternative is to keep the original amcl code unchanged, and write a tf publisher (from map to odom_future frame) using the future pose . This approach seems realistic, because it does not affect localization and sensor readings, but I could not find the pieces of code in move_base to expect the frame odom_future instead of odom.

Could somebody provide me some guidance?