Revision history [back]

As long as you correctly define the transforms for those sensors, you shouldn't need a second EKF, I think.

In most applications, sensors are not mounted at the robot's origin, so you need to provide static transforms that define the offset of the sensor from the base_link frame. I think the same is true in your scenario, but the transforms would just not be static. It would require you to publish the relative transform between the two sections, but with the right sensors, that should be do-able.

While I wouldn't think independent estimation of the states of both sections would get the performance you'd want, there's nothing stopping you from doing that now (i.e., there's nothing to be worked on to support it). Just fire up a second EKF instance and fuse the sensors accordingly. Treat it as two separate robots.