Drones, Swarms, and Storms of Drones
There is no situation when I want to encounter a drone of unknown origin, and in the future we'll be seeing hundreds.
In 2016 the number of consumer drones purchased was 2.4million; which was up from 1.1million in 2015 [source, source]. Continuing that trend (pessimistically) has about 10-20million drones sold in 2020. The consumerization of small drones opens up big questions in the future of stable society and warfare.
Don’t believe me? Click below for a quick introduction.
Drones are cheap, versatile, and difficult to track. If you think, “hmm, maybe people won’t exploit them,” here is the US Military practice releasing a micro-aerial-vehicle (MAV) swarm. The classified content must be more extreme.
And the clincher, Turkey has suicide bombing drones it could export (designed in Isreal). The drone, called the Kargu is named after hawk in Turkish. Read some technology specifications:
The 15-pound Kargu-2 can fly at up to 90 miles per hour and can remain airborne for up to 30 minutes. It has a line-of-sight control link with a range of around six miles.
The drones can carry one of three different types of warheads, including a high-explosive fragmentation one for engaging personnel and other unarmored targets in the open, a thermobaric type good for targets in confined spaces such as buildings or caves, and a shaped charge for attacking lightly armored threats.
The hawk naming now should be abundantly clear - it has the ability to eliminate any visible target. The seller claims that this drone is for anti-drone and asymmetric, modern warfare. It is only a matter of time until these technologies trickle down into even more controversial hands. The US Central Command Chief already declared:
“US troops can't keep up with the flood of cheap drones downrange.”
This is the storm of drones I am writing about. 10s, 100s, or 1000s of drones attacking our troops abroad or leaders at home. How do we defend against this?
The definition of a drone and a swarm is not always clear - especially swarm.
Google default definition (after some noises, and bees):
a remote-controlled pilotless aircraft or missile.
I think Google’s definition will rapidly be outdated as drones are controlled onboard/locally (autonomous).
Google’s default definition is actually not too bad (better then many researchers I encounter using the word):
a large or dense group of insects, especially flying ones
The word that I think is imperative is dense. A swarm is a dense object, where controlling one member differently may go totally unnoticed. In this end, a swarm is governed by a) emergent behavior or b) central control.
Does a party of robots become a swarm at 10, 50, 100, or 1000 agents? Swarm-control research in the 2000s primarily focused on optimal-control for the order of 10 robots. The research is focused on a few robots, and each individual is still crucial. In my circles, we don’t consider that a swarm, but we like to think about where that boundary lies. A comment on swarm definitions from my advisor Kris Pister:
How do people do [swarm control]? Somehow we have centralized hierarchical planning with distributed decision making at all levels as the communication breaks down or time constants of the dynamics are shorter than the communication/decision-making latency. Military operations are a common example [of a swarm], but running a household is similar - usually you don't run out of coffee, and usually two people don't buy coffee at the same time. (Households are not a swarm by Nathan's definition, but militaries are, and so is public health).
We are interested in swarms of micro-robots to assist in emergency response, agriculture, distributed sensing, and more. The distinction is that micro-robots can be made at an immense scale (1000s), so we have a totally different magnitude of concern.
How do we wrangle a true aggressive swarm?
Here is what I have found (and done) in the space of anti-drone work.
This was in a slew of emails about drones in the defense industry. The drone (equipped with a surface designed to ram other drones and break rotors and motors) is named Anduril (also known as the flame of the west), and is impressive, but I’m not sure how well it’ll work versus an expectant target.
Click below for a demo of how an anti-drone drone behaves.
Smarter anti-drone drones?
I don’t love this one because it seems fairly obvious that the technology used to improve defenders could improve attackers. We are researching how a team of defensive drones could try to prevent many attackers from entering a highlighted area. The thing to learn is how to position each drone relative to the attacker so that it can launch the net at the attacker (destroying it) while remaining airborne. I suspect we may have drones to do this around airports and many other critical, public areas.
Click below for a video illustrating a defensive swarm (red) fending off invading drones (blue) from a base area. This is based on Boids flocking algorithm, and an active area of research.
Brainstorming anti-drone techniques results in some common answers [see more]:
Drones themselves: a drone equipped with a net can easily take down another drone. Drones also can crash into a target to destroy fragile flying pieces (see below, Drone smashers). A one-on-one trade of a military object for a consumer level drone is hilariously expensive (I estimate 1000x).
Electromagnetic pulse (EMP): EMP is hailed because it’ll destroy the electronics and render the electric-motor driven vehicle (and its control electronics) useless. The issue is that EMP pulses generally have severe effects on things other than the target, and development of mobile / handheld EMPs is pricey and slow.
Lasers / other ground based defense: I’m not sure how useful these are, but there are military-vehicles equipped with high power lasers designed to melt pieces of oncoming drones. The issue again is that this defense is way more costly than the offensive (potentially consumer drones).
What is the ethical line of working on anti-drone technologies?
We are simulating “drone dogfighting” in my research group, and I wonder if we should be. It’s not clear how to control the dissemination of work, but how to nations close the gap from offense to defense in technology, without making the world more dangerous (see arms race). History shows that this is hard to avoid.
I have a future post about the problem with accelerating technological progress and the difference between offensive and defensive capabilities. I think we need treaties and active cyber awareness. With drone’s it is difficult to study because consumers love drones! People want drones to deliver their pizza robustly. How do we stop progress and availability if consumers drive the market? We don’t.
Other storms brewing
Bunches of drones forming and moving over Colorado and Nebraska prairies at nighttime. Not practically traceable, an invasion of privacy, and causing some state-level responses.
The US Air Force testing and utilizing unmanned drones for undisclosed missions. This is unsurprising.
The US Air Force testing a pilot-less fighter. They hope the advantage in sustained g-forces will help, but honestly the processing speed for modern autonomous systems is still years behind. In a few years, yes this will be the reality.
How many of you have actually listened to a bunch of quadrotor drones flying nearby? It sucks. A lot. To the point where even considering using quadrotors extensively for local delivery sounds so nonsensical to me.
In my research group, we are developing a silent drone, but it is still so many steps from practical that we do not worry about militarization (yes, we discuss the ethics of it). You can watch the video below.
What’s new with me
Happy solstice! I am reading (newsletters/blogs):
A funny read about a women in China playing the flight insurance market to literally make money.
Human Compatible: Artificial Intelligence and the Problem of Control - Stuart Russell: Why we need more future-conscious AI designers.
Waiting for my copy of Race After Technology: Abolitionist Tools for the New Jim Code - Ruha Benjamin, to show up. Part of an anti-racist technology reading group at UC Berkeley Graduates for Engaged and Extended Scholarship in Computing and Engineering (GEESE).
I am listening to / watching:
Sam Harris’s calculated exposition of current civil unrest. If you want an informed, and calm opinion on the state of racism and police violence, I encourage it greatly.
Two weeks ago I participated in #ShutdownSTEM to gather attention for racial disparities in academia. Brett Weinstein and Joe Rogan discuss a higher level lack of rationality in the shutdown (I don’t entirely agree).
A fantastic robotics talk from a graduating student at UC Berkeley, Justin Yim. If you want to know the process of precisely controlling a novel robot, watch.
Want another AI researcher to follow? Check out Lex Friedman, who volunteered on twitter to do 20k pushups (or other exercises) a day for a new challenge.
13th on Netflix. Documents the systematic crime-centric attack on our own African-American population.