As initially commented here: https://discourse.ros.org/t/introducting-behaviortrees-cpp-who-needs-finite-state-machines-anymore/6899/7?u=v-lopez

Once you are reusing components (ideally even components you have not developed yourself), there will be some name clashing, which can be potentially dangerous (ie: mix the targetPose for foot placement with the targetPose for grasping).

We have developed a FSM implementation and faced this issues, and approached them in the same way ROS1 does (remapping and namespaces).

When a FSM includes another FSM, it can remap some of its identifiers (targetPose:=walkingTargetPose), or the sub state machine can be put inside a namespace, and the one including it can perform the required remappings.