Executing blind versus watching and adjusting
Open-loop control means executing a pre-planned sequence of commands with no feedback about whether it's actually working. A robot arm running open-loop might send out a pre-computed sequence of torques, calculated entirely from the dynamics model in advance, and simply play that sequence back regardless of what's actually happening to the arm. If nothing ever computes an error signal, nothing can be corrected while the motion is in progress.
Closed-loop control instead continuously senses the robot's actual state and adjusts on the fly. It closes the loop from output back to input, the robot's real position feeds back into the controller, which recalculates its command, which affects the robot, which is sensed again, over and over. This is the sense-compare-correct cycle from the previous lesson, running in real time throughout the motion rather than being planned once beforehand.