Estimate the pose, then drive toward it
Position-Based Visual Servoing takes the opposite strategy. Rather than working directly in pixels, PBVS first uses the camera image, together with a model of the target and the camera's calibration, to estimate the target's actual 3D pose relative to the robot. Once that pose estimate exists, the rest of the control problem looks like ordinary Cartesian motion control: compute the error between current and desired 3D pose, and drive the robot along a sensible path in real-world space to close that gap.
This produces exactly the kind of clean, predictable, intuitive trajectory that IBVS can struggle to guarantee. Because the controller is reasoning in the same physical space the robot actually moves through, the resulting path tends to look like what a person would plan by hand, a direct, efficient approach to the target.