A work cell and method for automatically separating objects disposed in 3D clusters includes dispensing the objects onto a horizontal conveying surface to form a 2D array, reforming the 2D array into a 1D stream in which the objects move in single-file in a predefined moving direction, utilizing a vision-based or other stationary sensing system to identify a selected target object in the 1D stream as the target object passes through an image capture (sensing) region, calculating trajectory data defining the target object’s time-based position in the 1D stream, and then utilizing the trajectory data to control a robot arm or other object removal mechanism such that only the selected object is forcibly removed (eg, swiped or picked-up) from the horizontal conveying surface. A continuous-loop-type conveying mechanism includes two parallel conveyor-belt-type conveying structures and associated belt-switching structures. An Al-powered vision system identifies new object shapes during preliminary learning phases.