Thank you for your contribution, I happened to find this exactly when I am in the beginning of my project for having a path planner for an ackermann rover to avoid obstacles while moving towards the goal point.
I am using ROS2 Jazzy with MAVROS and Gazebo Harmonic. I cloned your fork and was testing offboard mode control with the ackermann rover. I am publishing position setpoints on the MAVROS topic /mavros/setpoint_trajectory/local. Upon providing a single MultiDOFJointTrajectoryPoint in the MultiDOFJointTrajectory with the position specified, the rover moves to this position.
What I experience is that even though the rover reaches the desired trajectory point, after some time, it begins to circle around this point (while my ros node continues to publish the same trajectory point), resulting in this:

This was with the NAV_ACC_RAD set to 0.5 m. Upon increasing this to 1 m, the number of circles around this trajectory point reduces. But is there a solution to avoid this?
From what I understood, the current local pose estimate keeps drifting over time and eventually causes the rover to think that it is not at the desired trajectory point and tries to move towards it and then ends up steering around it.
I wish to integrate a path planner (maybe using the ROS2 Navigation stack) and would like to get some advice on what would be the best course of action.

I switched to using the Micro XRCE-DDS Agent & Client instead of Mavros to send the position setpoints on the TrajectorySetpoint message. I was using https://github.com/Jaeyoung-Lim/px4-offboard/tree/master and wanted to see the rover move in a circle.

I want to get some advice on how to improve the trajectory tracking by the rover, so that the final trajectory looks like: