When light itself is the bottleneck
Every remotely operated system faces some lag between command and execution, but space pushes that lag into an entirely different regime. Radio signals travel at the speed of light, which is fast, but distances in space are enormous enough that the delay becomes operationally significant. A signal to Mars can take several minutes to arrive one way, meaning a round trip of a question and its answer can take the better part of a quarter hour or more depending on planetary positions. During that gap, nothing on Earth can influence what the spacecraft is doing in real time. Compare this to the underwater course's ROV model, where a tethered vehicle receives near-instant commands from a pilot and the entire approach to control assumes a human is watching a live feed and reacting within a fraction of a second. Space simply cannot offer that.
This isn't a minor inconvenience to be engineered around with a slightly longer wait, it forces a completely different philosophy of control. Real-time teleoperation, the backbone of underwater and many terrestrial remote operations, is not physically possible for a Mars rover encountering a rock in its path. By the time a human operator sees the hazard, decides on a response, and sends the command, several more minutes have passed and the situation on the ground may have already changed, or worse, damage may already be done. Space robots therefore cannot be built as extensions of a human operator's hands, they have to be built as independent decision makers that only check in periodically.