Where finite state machines start to strain
A finite state machine (FSM) represents a robot's behavior as a set of named states, like Idle, Searching, or Grasping, connected by explicit transitions that fire on specific events. FSMs are easy to reason about when a robot has a handful of states, but every new state tends to require new transitions to and from many of the existing states. A robot with a dozen states can easily end up with dozens of hand-written transition edges, and adding one more capability, like a low-battery interrupt, can mean touching transitions all over the diagram.
That transition explosion is the core scaling problem. Because a state's behavior and its transitions are tightly coupled to the specific states around it, an FSM state is rarely reusable on its own. Copying a 'search for object' state into a different robot's FSM usually means also copying and adapting a web of transition logic that was written for the first robot's specific state set.