By learning control coefficients designed for sampled-data systems rather than continuous velocity fields, you can steer large swarms efficiently in just a few control steps while respecting real hardware constraints.
This paper presents a control framework for steering large swarms with minimal updates by learning finite-window control coefficients that respect how real systems work—with intermittent control updates rather than continuous commands. The approach scales to large swarms while automatically respecting the system's dynamics and control constraints.