visual servoing

Visual servoing means steering a robot's movement using a live camera feed as the constant feedback, correcting the motion many times a second based on what the robot sees right now. It is the difference between glancing at a target, closing your eyes, and reaching blindly, versus keeping your eyes open and adjusting your hand the whole way there. In visual servoing the robot keeps looking, compares where things appear against where it wants them to appear, and nudges its motors to shrink that gap — over and over, in a tight loop, until the picture matches the goal.

The trick is the closed loop. An open-loop reach computes one plan up front and commits to it, so any error — a slightly wrong measurement, an object that shifted, a joint that drooped — goes uncorrected and the robot misses. Visual servoing closes the loop by feeding the camera's view back into the controller continuously: see the error, move a little to reduce it, look again, move again. Because every cycle uses fresh vision, the robot naturally copes with things it did not perfectly know in advance, much as you can thread a needle by watching and adjusting rather than by calculating the thread's path once.

This is what lets a robot do delicate, vision-guided tasks in a changing world: a welding arm that tracks a moving seam, a drone that holds station over a drifting boat, a manipulator that aligns a plug with a socket, or a camera-equipped gripper that keeps centering on fruit even as the branch sways. The cost is that the controller must process images and react fast enough that the loop stays stable rather than overshooting and oscillating.