Hacker News new | past | comments | ask | show | jobs | submit login
Drones learn to search forest trails for lost people (phys.org)
92 points by jonbaer on Feb 10, 2016 | hide | past | favorite | 39 comments



Forest trails are just a distraction from the real story here:

> When tested on a new, previously unseen trail, the deep neural network was able to find the correct direction in 85% of cases; in comparison, humans faced with the same task guessed correctly 82% of the time.

They made a computer better at identifying trails than a human. That they can manage that and quickly enough to feed a flight control loop with directions is mighty impressive.


A couple of people hating on this. Despite the obvious fact that you can in fact get lost on a trail, the progress of autonomous rescue drones doesn't just end at following trails.

Once they're able to detect trails and follow them looking for people, the next step would be detecting signs of where they might have wandered off trail, or (especially in snowy conditions) detecting and following where they have started their own trail.


You can also be injured, left behind, faint, stop to help someone, etc.

A drone that can follow a trail that changes quite a bit over the year and is trained to recognize obvious trail signals (jacket on tree, reflectors) or human voices offtrail as it passes ("help! over here!") would be excellent. A drone pass at closing time might become a standard precaution and CYA thing for parks, plus they can give trail notes like where it's poorly marked, blockage and collapses, and so on.


Ugh, I hope not! Nothing like ruining the joy of alpenglow with the high-pitched buzz of the nanny state.


To be fair, the wilderness being littered with dead bodies would probably ruin it more.


It isn't, though!

The only wilderness place I know that is 'littered with dead bodies' is Mt. Everest. And that is from hypoxia and exposure, not getting lost.

EDIT: Hmm, and maybe Aokigahara Suicide Forest in Japan. Also not from getting lost.


Amazing work! Where I lived in Australia tourists often would get lost and quickly dehydrate, be at risk of hypothermia overnight or immobilize themselves by breaking a leg. Phone coverage is not good in the country so you couldn't call for help. Depending on your location, vegetation is difficult to see through for search-and-rescue spotters in the air. Once you get off trail it's not as simple as backtracking and so have to wait for help to arrive (so long as you let someone know you were hiking ahead of time).


Makes me wonder if, when searching for someone lost in scrub, it'd be feasible to launch a temporary, roaming mobile phone tower/repeater that might give the hiker temporary phone access to get help?



A drone trained to seek out humans is one of my worst fears.


It was inevitable, even from a purely benevolent perspective.

Personally, I am much more worried about heart-disease, dementia, and depression.


And when new treatment arrives, how any individual will be able to afford any two.


Most developed nations have free universal healthcare paid for by taxation. Perhaps the countries that don't will implement it if the alternative is watching citizens die because they can't afford treatment.


See also http://uavchallenge.org/, plus Andrew Tridgell's talk from LCA 2014, 2015 and 2016 :D


I wonder when we'll have a "Gumball Rally" for drones: an unaided, coast-to-coast race. :)


Autonomous hunting drone becomes reality just a switch away.


It should work the opposite way. You have a pack of tiny drones with you that you can release and they go alert authorities.


If by "tiny drones" you mean electromagnetic waves, you can just bring a sat phone and call your mum and ask her to come pick you up.


Or maybe just pack a tiny GPS device with extra batteries for emergencies instead of a tiny drone.


s/lost people|missing persons/tenorists/g

Welcome the suicide drones.


First they come for the tenors, then they come for the baritones…


We can weaponise practically any technology breakthrough. That's a really bad reason to stop making new things. We need to develop new, useful things and work hard on making the world a place where everyone can live in peace so they're not used as weapons.


To name a few counter-examples: solar panels, wind turbines, MRI, artificial hearts, defibrillator, antibiotics, vaccines, refrigerator, lightbulb, the list goes on. I wouldn't say people should stop making new things, but some things are obviously going to be weaponized and that shouldn't be taken lightly or worse ignored.


It's the other way round. The money for this kind of tech mostly comes from the defence sector. I'm cool with that, but let's not kid ourselves: the reasons these things exist is that they are weapons.


What about batteries? I think most quadcopter drones like the one pictured only last for a few minutes before needing to be recharged. They won't be able to cover much ground in that time.


People have flown Phantom 3s longer than 3 miles (one way). That's a lot of ground to cover out in the wilderness.

https://www.youtube.com/watch?v=oTr2vtWFYMU


A very cool project, and drones could certainly move faster than people. However practically speaking, lost people are frequently not on the trail.


In my experience (7 years of NSW State Emergency Service)

(a) People are very often on the trail. An initial report of "lost" typically means "didn't report in on time". Many times that means that they're just slow, or injured, or walking in the wrong direction on a poorly marked trail.

They get found, no one makes a big deal of it, it never gets on the news, so it doesn't register in our consciousnesses, but it's fairly frequent.

(b) The first step in a co-ordinated search effort is to search the paths. If the search area is large (e.g. we know where they parked their car, but not which path they took) then you can need to arrange a fair number of searchers to cover all the paths. That's slow. Not just to do the searching, but also to get a team mobilised and coordinated. If drones can do that job (even just a first pass) it saves a lot of time and frees up a lot of searchers.

(c) There's "trails" and then there's trails. In Australian bush, a "trail" might not actually be created by humans. It could easily be a kangaroo path, and walkers sometimes stumble on to them and end up in places the searchers didn't expect. The drones might be better at searching those paths.

(d) It's not uncommon for someone who gets lost to eventually stumble upon a trail and not know where it is, where it leads, or which way to go. Just because they're on a trail now, doesn't mean it's the one they started on.

(e) For people who want to be found, there's likely to be a fairly high success rate in drones that move along all discoverable trails, emitting out emergency messages and listening for responses. Most human search teams aren't doing much more than that.


I guess my perspective is skewed. I am on a mountain sar team. The sheriff handles most of the "never gets on the news" cases, and calls us when it is going to the news.


Lost often does include people on trails. Lost simple means they did not return, which might include they died or where injured on the trail, are lost of a trail, change they plans and didn't let someone know, etc. Beyond that, this would be very useful for not getting lost; think Google maps for trails.


Move faster for a short time. They could certainly fly down trails looking for people via image recognition, but its hard to compare than to a person walking a trail for 8/10/16 hours looking for all sorts of signs, even smells.

Many lost humans are also nowhere near trails. Old people, people with mental issues, and young children often actively hide from searchers. Flying robots might not get the results one would expect.

Dogs. Dogs find people.


This project is made for an alpine area. With steep as- and descends a drone would be much faster than a human and a dog. And a thermal camera would be a good replacement for a dog. However, most people get lost while weather conditions with bad sight and a drone wouldn't help much here.


Sorry, didn't mean to downvote; hit the down arrow accidentally.


Interesting, useful and scary at the same time.


I wonder if some kind of "anti-drone invisibility cloak" could be a thing. Or some kind of tiny turreted laser light that can detect a camera and point a beam into its lens. Maybe an app to detect signatures of drones approaching by their radio signals or rotor sounds.


>>> ...a small quadrocopter to autonomously recognize and follow forest trails. [...] to accelerate the search for people lost in the wild.

If there are trails, then this is boarderline "the wild." Lost in parks maybe, but wilderness by definition is off-trail. I'd rather see the drone that can search through farmland. Kids do get lost in cornfields.


It's true that some really lost people in wilderness find themselves off-trail, maybe fallen into steep terrain or on the wrong side of a ridge after coming down hill.

If people are intentionally in trail-free 'true' wilderness, such as snowy high country, they are also much more likely to carry PLBs, leave flags for wayfinding, know how to use a map and compass, etc.

However, a huge amount of search-and-rescue expense is spent looking for less-prepared people who turned the wrong way at a fork, or who broke their leg 10 miles from the trailhead, etc. This technology still seems useful to me.


Do you realise that this is a nasty comment, attempting to make it look like you're more of an outdoor person than whoever wrote this article? Or is it subconscious?


Active RFID beacon that can be worn in a hikers pack/clothes... that allow for MANY drones to spiral locate over a large area....

basically the drones will spiral out in some pattern better defined by not me which will allow them to look for the beacon and then find the person.

The spiral pattern defined by "how long has the person been missing on foot" and "how far could they have reasonably traversed uninjured in this landscape




Consider applying for YC's W25 batch! Applications are open till Nov 12.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: