to use it in a loosely collimated manner, and tight collimation requires precise tracking and aiming. I'd tend toward a less selective approach, especially as swarm techniques are being developed to overwhelm defenses. It shouldn't be too hard to economically assemble a DIY version, especially given the deep learning machine vision capabilities a $99 mini AI supercomputer can provide. Have it perform classification for target identification then regression for target tracking. As I was shown a demo in September, the data needed in the form of images of drones flying can be synthesized in volume from a much smaller set of images, and performance depends strongly on the amount of data provided for training.