A swarm of drones flying through heavy forest–while staying in formation | Hacker News Reader

A swarm of drones flying through heavy forest–while staying in formation

sciencemag.org

Github: https://github.com/ZJU-FAST-Lab/ego-planner-swarm

Longer Video: https://youtu.be/K5WKg8meb94

Plenty of negativity in the comments here. Used properly, a system like this could be a boon to the forestry industry, invasive species management, and forest fire fighting and prevention, among other things.

A friend did his PhD in lidar mapping of trees and forests, with a focus on identifying risk trees. He primarily focused on roadside trees that were a danger to drivers, but that was in part due to the ease of lidar mapping from a vehicle. He did trials with drone imaging, but they had to be manually controlled, which made them no better than people.

I live near a national park, and they average several lost hikers and roughly one death a year. The majority of those are due to poor planning and injury (FFS always check in at a ranger station, tell someone your schedule/route, and stay on the trial if you are injured!). A system like this with thermal cameras could find injured people much faster than a search party every could.

Half the battle of fighting a forest fire is understanding its location and boundaries, and tracking/predicting its movement. A system like this could be augmented with smoke particulate sensors and thermal cameras.

Many ecosystems are being devastated by invasive species. Wild hogs, anacondas, airborne pests. Early attempts to track 'murder' hornets were hampered because the scientists could not follow their transponders effectively.

The main issue there is flight time. Most small drones have very limited endurance.

That said SAR already use drones to help search when weather permits.

For autonomous systems one thing I find interesting is how far they have to go in this sort of domain versus a good human pilot for example: https://youtu.be/bFVvUeP_oqY?t=36

Here is one example of an automated drone charging base: http://www.edronic.com/

Others are working on drone charging systems that are faster than the unit's flight time: https://dronelife.com/2020/08/01/a-drone-battery-that-charge...

To the best of my knowledge, all SAR drones operate above the tree canopy. The forest I live by has, even at peak occupancy, more people sized animals than people. It makes thermal imaging through a dense canopy difficult at best. A good friend spent 2 years as a coast guard pilot stationed near me, and he spent the majority of his time doing SAR training. He was the one to describe the issues with thermal imaging above the canopy. We discussed using drones at length, and his main issue was the risk that they might randomly enter his flight path and damage a rotor. Operating below the canopy would eliminate this risk.

Well I live in Iceland and we famously have no trees. :D

People are working on that!

https://phys.org/news/2019-07-iceland-trees-razed-vikings.ht...

Speaking of SAR, well the other SAR (Synthetic-aperture radar) it was discussed yesterday that a radar reflector can be used to bounce SAR signals back, which would highlight your location, such as it's used for marine applications.

Perhaps carrying one of these folded up ready to deploy in an emergency wouldn't be a terrible idea as the availability of SAR imaging goes up. Much cheaper to carry that than some of the GPS/beacon devices sold to hikers. Image detection is already a common ML tool for Earth observation analysis so building in something that looks for that marker in a known area of forest sounds possible.

https://www.westmarine.com/WestAdvisor/Selecting-a-Radar-Ref...

The traditional version of that is a bright orange square or signal mirror. They pack down really small and are extremely visible. Similarly little IR and retroreflective glint squares are handy for spotting people at night.

That said if you're doing anything serious away from people carrying a GPS and EPIRB/SPOT is way more effective. The latter working even in the most remote locations. It's no good being able to signal if no one is even looking for you.

There is already a product on the market call RECCO. It is a two part system very similar to what you described. The reflector is seen directly into coat, boots, you name it. SAR then uses a device that sends out a signal that bounces off the reflectors.

https://recco.com/technology/

I think some sort of hybrid balloon/drone could be more suited for certain tasks. Still, the demonstration shows that we could have a fleet of whatever navigate a forest.

Wow, I really want to see a photogrammetry system that can turn that video into to a reconstruction of the area.

The work of your friend sounds fascinating. Can you link to any published material? My professional work relates to tree risk and spatial analysis. Would love to see some of the higher level research.

This should be a good start: https://www.mdpi.com/journal/forests/special_issues/LiDAR_In...

Keep in mind that timber cruising is an industry, so there are also plenty of focus on software and hardware to speed up inventories: https://www.fs.fed.us/forestmanagement/products/measurement/...

You're hitting the nail on the head. Forestry is probably one of the most underserved markets. The opportunities here are abound even in the most fundamental areas like forest management.

Forests are "managed" plenty and dying at a massive rate. Local rednecks use the uniform and new equipment to track nature loving stoners. Wild mega-fauna who are true citizens of this place are the target of weaponized tech and have been for millenia, resulting in their massive over hunting to extinction. The narrative of "keeping hikers safe" and "clearing the roads" takes into account none of this, and aids some of it -- wilderness road building.

Sorry, not sorry

> A system like this with thermal cameras could find injured people much faster than a search party every could.

This is done a lot, you are not used to the idea of 'Swarm/Formation' so you are using it to cure everything.

One long range drone with thermal cameras (and other wavelengths) is all you need. No formation is needed or wanted.

I still rarely see even one drone used. We need to start with one, if its useful.

I can give you many locations with invasive species, swarms won't solve that. Unless you give the swarms guns.

But one drone with a gun would kill pests fine. I think even with hunting humans most killing wouldn't need formations.

> Plenty of negativity in the comments here.

Mostly fear. There are a lot of people suffering out there right now and the risk of that leading to a war is real. This is a salient reminder that militaries can send a drone anywhere and mow people down.

The forestry industry is nice and all, but tech like this is going to literally kill people. That will be the most interesting application in this decade.

Consider drones/AI-tech similar to nuclear weapons in terms of its deterrace as well as assault capabitities.

- What if adversaries develop full AI-tech with weapons mounted on them? How can a nation defend against such a threat?

- What if democracies around the world get wiped out by such a tech which has no public opposition in communist/ authoritarian states? They will and are developing such technology.

- We just saw the use of (non-AI) drone military tech in the Armenia-Azerbaijan war. Being defenseless is not an option.

Yes! Let's vote to make more killing machines instead of funding schools and hospitals!

Exactly, the autocratic CCP is heavily investing in military drone capabilities, we really have no choice but to remain one step ahead. If we lived in a world without such regimes then that wouldn't be an unfortunate pragmatic necessity.

What a completely ethnocentric and asinine perspective. I am not defending China here - obviously they are up to a lot of problematic atuff.

If it wasn't the CCP, it would be Russia again, or if not Russia, then another emerging power. In 10 or 20 years it could be India, or the fragmented parts of European Union that combine to form authoritarian regimes if the swings toward autocracy continue.

Every military force is investing in drone technology because any military force in the world is investing in drone technology, and while the US was the leader for a long time, the push to outsource manufacturing and globalization led to a broad distribution of the skills, tools, and manufacturing base to build modern drones.

China is no more the root cause of the arms race than Thog was when he hit Ugh with a stick some time in ancient history. Its humans and tribalism all the way down.

China is suppressing its population to great extend. More than other countries save North Korea. All school books are filled with propaganda and children need to recite xi's virtues every day. It's a brainwash machine and totalitarian. So far they are not a worldwide military power and the US cannot be conquered but this technology could change that. Deploy them in millions. Never think your opponent has the same moral values as you.

Precisely. His line of thinking would've played out to a disastrous end in the 1930s-1940s era, "oh if it wasn't for the Japanese or Germans it'd be X". Some people and some regimes are psychopathic in nature, that's the unfortunate reality, and such regimes, especially those with expansionist desires (ever heard the CCP's rhetoric about Taiwan?) have every incentive to invest in military capabilities irrespective of whether others are doing so too.

That's a rather unnecessarily aggressive comment. What I said is not ethnocentric (you're presuming I'm American or white?) or asinine.

You're not considering the historical consequences of your perspective. And I made no such claim that China is the root cause of the arms race, that's a strawman.

Historical counterfactual: What would've happened in WW2 in Asia if the US had no aircraft carriers because some pacifists convinced the administration that investing in arms was just tit for tat tribalism and contributing to an arms race?

If you can't explain why the US should not invest in drones today but they should've invested in aircraft carriers in the 1930s and 1940s, then you're holding a sack filled with little more than misplaced confidence.

Combine that with this: https://www.youtube.com/watch?v=HipTO_7mUOw

Do you really need to get thermal cameras below the canopy to see forest fires? I think even satellites can see fires on thermal. Drones flying above the trees should work too.

These whining rotors will be the last sound some poor freedom fighters hear in 10 years. And of course 10 years later the black market will have caught up, and these automated killers will be deployed by anyone from anywhere with near impunity.

More accurate than a roadside bomb, probably less risk as well.

I would be more worried about the fact that a government would want to kill me rather than the method it chooses to kill me. I don't need to be a freedom fighter to get shot by the police in the USA yet nobody cares as much unless it is done by some stupid toy that is never going to be a practical weapon.

I think there is an interplay between the means of killing and the willingness to do so that you shouldn't ignore. The U.S. elevated extralegal killing of brown people to a full-time pursuit under the Obama administration with the proliferation of the drone program.

It's not as easy to blow up a wedding with an F18 as it is to do so from a trailer in Arizona, for some reason.

Every president since we've had drones has killed more people with drones than the last guy. Bit early to call it a trend I guess, since we only armed the things toward the end of Bush's last term, but still seems worrying.

> Bit early to call it a trend

Disagree.

These programs, from the President's perspective, are as autonomous as the machines. Barely any interaction, if any, is required from the W.H. to keep the bombs dropping.

the extrajudicial execution of American's abroad by using drones was by far the most glaring change in US policy during that Administration

I'm not sure why this is downvoted. Obama established the precedent that it's totally kosher to use flying death robots to murder US citizens without any semblance of due process. You just have to claim (but not actually produce any evidence or convince a jury beyond reasonable doubt) that they posed an imminent threat to the US.

That’s been the case for a very long time. Even ignoring the civil war, US citizens for example fought on both sides of WWII both before and after it entered the war. As soon as you start talking millions of people you end up with some extreme outliers.

The only thing that’s changed is we have a slightly better idea who we are shooting over time.

PS: Somewhat more bizarrely, letters of marque and reprisal mean private citizens can get the rights to kill other Americans who happen to be working on non military foreign ships.

> some stupid toy that is never going to be a practical weapon

Please watch: https://www.youtube.com/watch?v=9CO6M2HsoIA

I think the real world examples are better than a hypothetical.

https://old.reddit.com/r/CombatFootage/comments/jgmou6/azerb...

Warning: This contains real world footage with examples including troops in a foxhole being precisely targeted by a drone that you can see on film moving into the foxhole with them that then explodes, numerous cases of troops in an open field spotting a drone, running from it and then exploding. Also contains numerous examples of tank and vehicle columns as well as fixed emplacements exploding.

I think we're there. As of 2020, and in particular the Nagarno-Karabakh war it's clear that having more drones than the other side is the key. Number of troops, tanks and gun emplacements isn't as relevant anymore. Mid-sized drones such as the Harop drone trump all of those.

> stupid toy that is never going to be a practical weapon.

Have you not heard of the recent Azeri invasion of Artsakh? It was a major war between modern armies that was won by small drone warfare, just a few weeks ago.

Yes, watch any of the Azerbaijan drone videos on Youtube. They decimated the conventional (tanks, vehicles, bunkers, artillery,) Armenian forces with "relatively" inexpensive Turkish and Israeli drones. A real object lesson about the future of warfare and the need for airborne countermeasures. A swarm of inexpensive drones like in this video, each one carrying an explosive charge, could easily overwhelm a traditional military position; in a forest, in a bunker or cave, behind fences, inside buildings. Hardly a toy.

As a combat Infamtryman (2007) the drone footage gave me chills watching. Im glad I don't have to fight in a war where drones are the norm. When I was in our drones were still very much in the early stages and practical use seemed very far off. There is definitely some serious amount of money for anyone who can field technology to defend tanks, infantry, and artillery from this sort of onslaught.

No need to have been in the military, these images are downright terrifying for anybody who understands what's going on... to think that thousands of young boys were killed there in a few days!

‘Small’ as in wingspans of up to 3 meters. You’d have trouble flying something that big through a forest no matter how manoeuvrable it was.

> yet nobody cares as much unless it is done by some stupid toy that is never going to be a practical weapon.

You're correct that that government wanting to kill you is a primary concern. However you're submitting this comment under a story with a video of drones flying autonomously through a forest, in a world where facial recognition is commonplace, and commercial drones can easily carry the weight of a polymer firearm.

There is value in the inherent friction a state encounters when deciding to kill someone. A lot of people have to say yes, and a person usually has to pull some kind of trigger.

Reduce that friction, reduce the amount of people involved, and it is a guarantee that coups and purges will be too tempting of an option for power seekers.

Depends. If you were high profile, you might be more concerned about the means.

For example, you might feel secure on your remote island retreat with its own airfield, security system, etc.

That any nut (not "Government") with a drone carrying an IED could surgically take you out would very likely worry that individual much more than it would worry you or me.

I recently watched the movie Angel Has Fallen on Netflix. The movie starts out with an assassination attempt on the president by a swarm of drones armed with explosives. The secret service agents were (not surprisingly) completely ineffective at defending against these drones using their hand guns. Just a fictional movie, but was still pretty terrifying to think about.

https://www.netflix.com/title/80190284

Your last statement might be the most short sighted statement I’ve read this month. In fact I don’t even think it’s short sighted, they are already incredibly practical weapons.

I'm immediately reminded of the "drones" in Half-Life 2. It really seems like the dystopia is here already.

Drones are a significant factor in present day conflicts.

https://en.wikipedia.org/wiki/2020_Nagorno-Karabakh_war#Mili...

Exactly. The Turkish military perfected that tactic earlier in the year, that's how they managed to suppress the Syrian Army's latest advance (which had been quite successful until that point). Artillery + drones proved to be too much for the Syrian Army [1].

[1] https://www.forbes.com/sites/sebastienroblin/2020/03/02/idli...

Yep, I think there is still time before they become popular black market weapons, but not that much time.

To be clear, those drones are more akin to unmanned planes than they are to quadrocopters.

Except for the one that is a quadrocopter (I guess use in the conflict I linked is not verified, but it's in use today none the less):

https://www.stm.com.tr/en/kargu-autonomous-tactical-multi-ro...

I suppose the different types of drones will cause different types of tactical shifts.

Oof some engineer is working on machine-learning to deal with the knockback of shooting a gun from a drone. It's probably an interesting as hell problem, enough to distract someone from the fact that they'll be enabling cheap robots to shoot people.

It is so sad that we, humans in general, devote so many resources to devising bigger and better means of violence.

Bigger and better actually worked pretty well w.r.t. MAD. Now we are developing smaller and better, which seems almost more dangerous since we're already using drones more flippantly than conventional military arms.

Maybe to handle knockback you could allow recoiling rotation around one axis and have the rotors automatically anticipate the force on fire.

Why do you need machine learning to "deal with the knockback of shooting a gun from a drone"?

‘cause people don’t understand this is a solvable problem any other way and of course ML can solve any problem you point it at.

well you need machine learning to coordinate these robots and they need to take into the effect of this robot shaking around a whole bunch.

It is very easy to work on these tools without even realizing it. Let us say that you are developing a video processing algorithm and publish the code on github. The developers of missile software can and will use it. At some point, they may ask you for consulting/support in exchange for a large amount of money and you will say yes, and never realize who they were. I'm not saying that this has happened several times to my lab.

In the video the drone self-destructs by exploding and producing shrapnel, I doubt knockback is a problem.

I agree with your sentiment but this already happens only that a human is piloting the drones. This is just a cost reduction, and discussing the ethics of it is more tricky.

I think it's more than a cost reduction, if a human is flying the drones then there is still someone who can rebel. If the only person who needs to actually do anything at all is a 4 star General it becomes alot easier to start mass killings

Link to the original creators https://www.youtube.com/watch?v=9CO6M2HsoIA

That's the first thing I thought of when I saw it. The Viet Cong are lucky this kind of thing didn't exist in 1968.

Periodic plug for Daniel Suarez https://daniel-suarez.com/ , I'm certain I recall a "death drone swarm" in one of the books but can't recall which one. It happens on a snowy mountain with a cabin, if that will ring bells for anybody else.

His writing is fairly-near-future (10-30 years) and uncomfortable in much the same way that the show Silicon Valley can be uncomfortable (that is, it's so close to the truth / likely near-future that it hurts).

I swear I don't know the guy, but this will be my... sixth recommendation of his books on HN. I just love me some near-future peri-apocalyptic fiction.

Ah, remembered the title, it was https://en.wikipedia.org/wiki/Kill_Decision

one of the rules of modern warfare is "don't use a drone when a bomb would do". I'm certainly not worried about collateral damage to trees. if my target is in a forest, just blow the whole thing up

But where in a large forest is your enemy? These things can search and destroy, especially if they get quieter. Imagine letting these things lose in a cave system. Explore, seek and destroy with no human intervention.

Video seems missing/broken for me.

YouTube link: https://www.youtube.com/watch?v=IujQzSQtaTU&ab_channel=Scien...

That horrible haunting sound

Images of war come to mind

Though this and AI may allow us to map out forest flora (fauna will avoid them)

Sounds too ominous for those of us who grew up with Happy Tree Friends.

>Though this and AI may allow us to map out forest flora (fauna will avoid them)

Or map out and chase poachers

Do they sound like angry bees, or do angry bees sound like them in some sort of twisted evolutionary pre-echo of our eventual extinction?

uBlock Origin blocked it for me.

Hm. In the simulation they stay together (Congratulations! that's really hard). But if they were supposed to stay in formation e.g. to search those woods, they left immense gaps. They didn't stay in formation at all.

there were only like 4 drones.. what kind of formation were you expecting? A quadrilateral, yes they created a quadrilateral shape lol

There were more than four drones in the simulation.

a simulation for such tangible milestones is hardly worthy of a research paper. Deploy 'em drones and see if the algorithm works out!

I would assume "in formation" would be more advanced than "kinda jumbled near each other" given the degree of coordination required. The sample video of real-life drones shows them crossing each other's paths (drone on R goes past both on the L) which doesn't seem terribly useful. The fact that the simulation had the drones move in an hourglass shape seems to show the same problem, they're not maintaining any kind of order for the flight, only spread out at the start and end.

I don't see how this is more advanced than each drone getting a start and end point and attempting to maintain a line. Is staying close to each other that significant of a challenge that I'm missing something?

Edit: The article is trash, the actual project is just navigating multiple drones in a highly-obstructed environment, nothing about maintaining formation. That's the impressive part, and this seems pretty damn good.

Let’s assume that over the next ten years quadrocopters get:

* More range and speed

* More obstacle avoidance autonomy

* More heavily armed

* Much, much cheaper

What’s the effective countermeasure against an enemy sending ten thousand of them at you?

Ground based defenses would have a certain advantage due to having access to more energy. That means more sensors, faster processing power and greater effective range.

https://www.thedrive.com/the-war-zone/9345/the-u-s-armys-las...

I think it's high time for schools to switch from First Robotics Competition (FRC) teams to First Robotics Countermeasure Competition teams.

Maybe you mean First Law of Robotics? (is joke, in case not obvious)

Directed energy is about the only thing that's going to scale as a tactical countermeasure. Pulsed laser/EMF in the joules per shot range would seemingly work with most smaller vehicles.

https://www.youtube.com/watch?v=JnoZGzLqKgE

Counter-drones, obviously. One who has the most drones (most money) will win.

Masers or lasers.

Or, simply, belt-fed automatic shotguns.

Self targeting guns, lasers, maybe even high pressure water cannons.

That sound. The flight of the slaughterbots.

A friend of mine works at Skydio, which is an autonomous drone company. You give the higher-level instruction like where you want the drone to go to, or get it to follow you, etc and the AI navigation system figures out how to get there while avoiding obstacles and keeping the subject focused in the camera: https://www.skydio.com/

Following an ATV through a forest: https://youtu.be/mXeUHD-eVMs?t=329

I've been watching Skydio for a while. They're doing amazing work. I tried to build a very simple autonomous drone in my last semester of school using a DJI Mavic Pro. Was pretty difficult, though that probably had something to do with the DJI SDK not really being setup to do what we wanted. Still, really makes me appreciate the quality of work that Skydio is doing.

They've got beastly hardware in those drones. I'm not sure if it's public info, so I won't get specific, but it's the kind of thing you'd find in high end gaming builds. It's surprising to me that they have the power budget for that given how much power it takes to fly a quadrocopter in the first place.

I was part of a student drone group last year. Previously they have been using various embedded systems like intel nuc's and nvidia tx2's. However this year they decided that it would be much better to just strap on a good old i9 instead, makes development so much easier when you don't have to spend a bunch of time optimizing stuff.

I love the apple music as drones fly in formation through heavy foliage, their probable only future application being the hunt of humans.

I wouldn't call that dense forest.

I wouldn't call that flying in formation either :))

I agree. Annoying editorializing distracts from something that is actually pretty cool.

Yup! It's literally some landscaping along a sidewalk.

Anybody here remember "Boids"? https://en.wikipedia.org/wiki/Boids

Now reaching a physical universe near you!

Sure. I tried that algorithm once, and knew Craig Reynolds. That's what they're using here. It works OK when you don't have turning circle problems and aren't trying to fit through tight spots not much larger than your vehicle. Not useful for ordinary wheeled vehicles, but fine for quadrotor drones.

The achievement here is not the planning. It's the mapping, from vision only, in a small package.

What's the state of the art with regard to silent flying drones? Is it possible?

As scary as these things look, a silent drone fleet would be truly terrifying. Imagine hiking along and turning around to that.

There are some people looking into active noise cancellation[0], but I have no idea whether that's viable.

A (more) silent drone will be huge for online shopping/meal delivery. Amazon has ambition to eventually deliver most packages to most places in matter of (tens of) minutes.

[0] https://scholarworks.gsu.edu/cs_theses/98/

> There are some people looking into active noise cancellation[0], but I have no idea whether that's viable.

Birds do it all the time ...

What do birds do?

Gliders are effectively silent until propulsion is activated.

This brings to mind a great video which I've posted here a few times: https://www.youtube.com/watch?v=9CO6M2HsoIA

What happens when technology progresses so much that anyone can make a swarm of killer drones? Will politicians no longer be able to go out in public? What about a distant future, when enriching uranium can be done in your garage for cheap...

“The team, which was also behind the earlier strategy, adapted it for swarms by getting drones to broadcast their trajectories over a wireless network. That allowed the other drones to choose routes that avoided collisions while staying in formation.“

I wish my car could do this.

Maybe Ajit Pai’s FCC replacement could get V2V/DSRC back on track?

> Maybe Ajit Pai’s FCC replacement could get V2V/DSRC back on track?

It's a bit late for that. The FCC reallocated the spectrum last month.

https://docs.fcc.gov/public/attachments/DOC-368228A1.pdf

The forensic and rescue applications of this are encouraging; Finding a body or lost person in a forest currently takes an insane amount of humans and effort. If a drone-swarm could produce video or imagery that can be reviewed by a crowd of desk-jockeys, it seems our chances of success just went up a good bit.

That’s not “heavy forest”, there is 2m between trunks and little foliage - a DJI drone can handle this.

Looks great, can you use this code with commercial drones or do you need some extra hardware?

Heavy forest? Dense? That is a park, and not dense at all.

The trees seem to be have white paint applied to them.

Trees are sometimes painted to keep beetles or similar pests off them - it's unlikely they painted them to help the drones.

Some types of trees can actually get sunburned.[0] White paint acts as a sunscreen.

[0]https://www.gardenzeus.com/protecting-your-lemon-trees-sunbu...

reminds me of that forest scene in return of the jedi but much slowwwwer

Incredible..

And they shared the source code.

....and they are Chinese!

(We detached this subthread from https://news.ycombinator.com/item?id=25467009)

I got a peek at the code.

if (aboutToCrash) { dont(); }

Reply on news.ycombinator.com