Anduril Industries, whose mission statement is “Transforming defense capabilities with advanced technology,” recently released video
of their Altius 700M autonomous drone unerringly crashing into a target
at the Army’s Dugway Proving Grounds in Utah, igniting same in a
suitably impressive fireball. A press report gave the demonstration a
glowing review, with the headline “Anduril attack drone deemed ‘accurate and effective’ in Dugway trials.”
Producing
dramatic test results is an art, sometimes requiring expert finesse.
Back in the days when the Army was trying to ram through a lethally
vulnerable version of the Bradley Fighting Vehicle, it famously filled
ammunition compartments of a test vehicle with water. This ensured the
vehicle would not explode in an embarrassing conflagration when hit, as
was otherwise normally the case. Anduril’s drone triumph therefore
requires careful scrutiny, especially in light of another
of the company’s drone demonstration videos a few months back, in which
a jet-powered Roadrunner drone returned to earth, tail first,
apparently without kicking up a (normally inevitable) cloud of dust.
Thus,
in viewing the demonstration of the Altius 770 M, which is an upscaled
version of smaller Altius models, with a larger 33 lb warhead, keen-eye
observers note that the target, a mock, SA-17 surface-to-air missile
system, is parked against an open expanse of desert, thus presenting a
conveniently high contrast image. As one such observer, with a lifetime
experience in calling out such shenanigans, commented to me, “Target =
black cross on white desert? Nice to know that some things never
change.”
In
the real world of combat, of course, targets are almost never so
conveniently highlighted, with a straight, light colored roadway right
beside it just to be extra helpful for the drone’s automatic target
recognition (ATR) system. Nor are they necessarily stationary. This
issue is essential to understanding the reality of autonomous weapons
systems, for all the excited commentary they elicit. Simply put, such
systems have to be trained to recognize a target. An analyst with deep
experience of ATR development in the real world put it to me this way:
“Let's pick any old legacy type of SAM system. Let's say the SA 10, and
if you had a picture of every single SA 10 in the world, you might have
30% of 'em look exactly the same and the rest would not. They aren't
going to look exactly the same. So Silicon Valley is saying to people
‘this is just a problem of having enough data and giving the machine
enough time to practice with the data that it will solve the problem.’”
But, he explained, the variations in the way a target, as observed by
the system’s sensors, can appear are almost infinite. “You have to
provide enough images at enough many, many, many different angles, light
conditions, altitudes, ranges of a given thing and inform the
processing around all of that information so that the image that's
actually coming into a processor that has something in its library that
it can use to say, ‘ah, I've got a match.’” Furthermore, the system must
be able not only to recognize a tank, and distinguish it from, for
example, a truck, but also a friendly tank from an enemy tank. As my
informant stated, “that’s not an insignificant problem.” (In Ukraine,
just to compound the problem, both sides make frequent use of the same
Chinese-made Mavic quadcopter drones, which troops on the ground cannot
distinguish even with their own eyes, let alone computer processors.)
Additional difficulties for ATR occur when the other side takes steps to
confuse the system. Infra-red sensors, which pick up heat signatures,
can be spoofed with fake hot spots that the system cannot distinguish
from the real thing. Even five years ago, U.S.Marines were learning to
hide from I.R. equipped drones by covering themselves with their tarps.
Nevertheless, such realities are unlikely to affect the steaming
embrace between the Pentagon and Silicon Valley “innovators” like
Anduril (you know it’s innovative because the boss, Palmer Luckey,
always wears a Hawaian shirt and shorts, just like Sam Bankman-Fried.)
“Tech,” which was birthed by lavish Pentagon research grants and
contracts, is coming home to mother, who had abandoned none of the bad
ideas so profitably nurtured over so many years.