Slashdot Mirror
Ford Tests Its Self-Driving Car In Total Darkness Using LiDAR Tech (fortune.com)
An anonymous reader writes: Using a combination of radar, cameras, and light-sensitive radar called LiDAR, one of Ford's self-driving cars has successfully navigated a winding road at night and without headlights. LiDAR works by emitting short pulses of laser light -- 2.8 million laser pulses a second -- so that the vehicle's software can create a real-time, high-definition 3D image of what's around it to determine the best driving path. Ford's self-driving cars come equipped with high-definition 3D maps, which include information about road markings, signs, geography, landmarks, and topography. If a vehicle isn't able to see the ground due to inclement conditions, it will detect above-ground landmarks to locate itself on the map. Ford's self-driving cars equipped with the LiDAR radar system are particularly noteworthy because they can operate without the usual cameras that depend on sunshine and street lamps.
Driving to bars will be a snap when combined with my gaydar.
So one of the problems with modern depth sensors is handling sunlight. Blackbody radiation can cause all kinds of noise in the signal that is bounced back at the camera from a laser. This makes it harder, not easier, for a lot of depth perception tech to work in good lighting conditions.
Driving at night should, for the most part, be easier for many systems than driving during the day.
What happens when all the other cars around it are also emitting the same pattern?
Great idea! Lets put people in a box on wheels that they can't control that has no headlights so not only do they have no control over where the vehicle is going or what it's really doing, they can't even see what's going on. Why bother spending money on windows or a windshield for these, even? Just make it a metal box that you close yourself up in. For extra safety, you get strapped down in a horizontal position, including your arms, so you don't hurt yourself until you reach your destination. Earplugs and a gag in your mouth so your screams don't bother the other 'passengers'. Make it coffin-shaped for extra convenience when the goddamned thing fucks up and kills you in a firey crash. Fuck you people and your goddamned death machines I WANT TO DRIVE MYSELF!
... A World On Fire
Why is this such a breakthrough? Google self driving cars have been using LIDAR for years: http://www.extremetech.com/ext...
Just because I can hook a shark from a boat, I do no offer to wrestle it in the water.
Sure but the lights aren't just there so the driver can see.
Often times it's important for other people to see you.
Minimum threshold fixed. Thanks!
the revenuers won't have a clue
LIDAR is an active sensor, meaning it provides it's own light to measure the surroundings. Therefore the car does not drive in complete darkness. Actually more impressive is when an active sensor works even in noisy environments. In this case it would be full sunlight with strong reflections coming off the road.
Do not look into lidar with remaining good eye...
My eyes reflect the stars and a smile lights up my face.
If rain/snow interferes with visible light it will also interfere with Lidar.
Knight Rider was doing this in the 80's.
Not necessarily. Lidar can use multiple beams and multiple wavelengths to work around this.
http://velodynelidar.com/faq.h...
"Velodyne's LiDAR sensors work well in snow, sleet, and rain. The multiple beam approach of Velodyne's LiDAR sensors with laser beams with millions of laser beams at different angles enables to find "holes" in-between the snowflakes to "see" the environment. An inferior LiDAR with only one or a few laser beams would not work as well as one with 16, 32 or 64 laser beams."
I'm presuming that deer (and other animals, like Moose) will be a considerable risk with this technology.
If the cars aren't emitting any visible light and are probably close to silent (assuming that when this technology comes about many cars will be electric) won't they be close to invisible to wildlife?
Wondering from the Great White (seriously, April 11 and we have snow here in Toronto) North.
Mimetics Inc. Twitter
Does this mean my handy "avoid-speeding-tickets" detector is going to beep every time a Ford goes by?
I'm already annoyed at the K Band alert going off every time I near a new vehicle that has driving assist technologies enabled.
I hope they had the good sense to encode a unique identifier into the LIDAR beam so that each car can tell its own LIDAR reflection from another, otherwise you will see a very different kind of "traffic jam" in the near future. Each car will be jamming the signals of every other car, and nobody will move an inch, not knowing where anything is. I searched the "news story" and found absolutely no indication that they even planned that far ahead. But don't worry, they didn't miss all the important Stock Market information. Now I know exactly what not to buy.
Sonar and Radar do not need Light either.
And Laser not needing light to work. Is that like magic?
Lidar exists as an acronym of Light Detection And Ranging, and was originally created as a portmanteau of "light" and "radar".
Imagine walking or driving by 1000 car a day pulsing laser and radar all over the place, that should be a real concern yet I don't read any mention.
No need for the quirky caps. Lidar is "light radar". It isn't an acronym. ("Sonar" and "radar" aren't anymore, either.)
and that's probably true for a 'stationary' Lidar gun being used for speed enforcement. And which can be wiped off frequently.
The problem is a car driving through snow/ice that builds up in front of the laser/sensors.
I know I've driven in ice storms where I stopped to 'de-ice' my car and I quite literally had a 2-3 inch high pile of ice all around the front of my car when I was done. That's a LOT of ice for said lasers to get through.
People in cars cause accidents....accidents in cars cause people
Driving at night should, for the most part, be easier for many systems than driving during the day.
Depends on the technology used.
The technology compete (or more often in practice: supplementing) with visible-light camera and image recognition.
Either using simple perspective recognition in case of single cams (like on Tesla, etc.) or using stereo correlation with a pair of cams (like recent Mercedes, etc.)
Such cameras work better with good lighting conditions, in complete dark they won't work.
Also, don't forget the main technology "competing" with such tools: the human driver.
We human see badly in darkness, we see better in sun light.
If the car uses technology that can see better in darkness it can react to things that the human driver might have missed.
By combining several technologies (e.g: lidar, camera, short range sonar, long range radar on Volvos) the car extends the range of situation where it can cleary "see" obstacles.
And avoid them (for the current type of collision avoiding technologies already in cars on the streets for the past few years) or drive around them (for the upcoming self-driving cars that are slowly starting to appear).
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
That's why car with collision avoidance systems currently on the streets (e.g.: Volvo)
combine the input from several sensors.
Not only do they use LiDAR (like TFA) and camera (also limited by visible light).
But also short-range sonars ("Parking sensors") that are less disturbed by snow (but are easier to disturb by air turbulence/compressed air).
and long-range radars (the thing mainly used by adaptive cruise-control).
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Actually head-lights have the opposite effect to wild-life: they increase danger.
When they get shone on by the head-lights, wild animals get startled and most often their natural reaction is to freeze
(hoping that the potential predator won't notice them ? i think that's a plausible explanation for the behaviour)
Standard procedure (as taught in driving course here around) is to hit the horn, so the noise will scare them into fleeing.
(It works, in my personal experience)
No headlight could in theory avoid the deers freezing.
Also the lidar can clearly recognize de obstacle and hit the breaks on time (though I have no first hand experience with that situation).
But in practice you'll probably have the headlights on (to be seen by other drivers, and also because the other technology used simultaneously by most system - camera - does need good lighting condition)
Still, the collision avoidance system is likely to hit the brakes in time.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Nope: your detector typically detect radio waves (as used by some doppler based speed traps).
It can't detect laser (in advance) because it's highly directional. The only situation where it could detect a laser is once the laser is already pointed at your car, at which point it is already too late, you're already being measured and eventually fined.
So no detector will try detecting lasers and thus no detector will get set of by a Ford.
Same with the other "avoid.speeding tickets" gadgets which are based on GPS technology
(along tables of speed limitation for roads,
and (hopefully kept up-to-date) tables of known radar positions).
Those will alert you of laser-based or plate-reading camera-based speed-traps, even when it can't sense them.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Thus lasers don't need light, they are already a light source.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Pfeh. News for geeks indeed.
In related news, Ford tests Model T in total darkness using "headlights" tech.
Heating + spray + wipers
I imagine people in your climate are viewed as edge-cases, despite the millions of you. An electrically heated windshield (and other surfaces), however, could probably prevent ice buildup. What I'm curious about is how clean the sensors have to be kept. Will one strategically placed bird shit or bug splat cause my car to do something unpredictable?
There are multiple sensors and the system is able to handle some degradation. The engineering required to make this work is well understood and has been in use for military and aerospace application for a long time. Properly implemented, these systems are not that delicate.
Unless he has one of these
This is not a detector, this is a jammer.
Whenever a laser beam is shone from afar on the car
(from a long distance, like when the cop start pointing the gun toward a car of which they want to measure the speed)
at that distance the beam is diffused (even if only a bit), if it was visible light wavelenght, you the driver would see the laser gun glimering whenever it is pointed in the general direction of the vehicle.
This laser light is detected by the detector, which triggers an alarm (so the driver knows to react) and starts shining light from various emitter on the car.
At that distance, the slight diffusion of light is enough that the emitter confuse the laser gun, slow down its measuring, and give some time for the driver to react.
Which I can only imagine would cause interesting readings on the LiDAR-equipped vehicle.
Nope.
The first and foremost reason:
all vehicle use a combination of multiple sensors.
At the range concerned by jammer, cars don't use the lidar at all.
They use mostly a conventional radar, and some also use visible light camera with optical recognition.
None of which would be severly affected by the jammer (maybe the car will be seeming blinking in the distance, depending on the IR filters used by the camera).
The LiDAR in car is used for much shorter ranges. It's generally pointing slightly downward. ... any object that is near the car that could collide with it so the car can autonomously break in reaction. LiDAR (and camera, and optionally sonars) are the main input for the forward collision avoidance systems.
It's not used to "see" cars hundreds of meters away (that's what the radar and/or camera are for, they are the main input for the adaptive cruise control).
It's used to see object with more precision in the immediate vicinity of the car.
It (combined with input from optical recognition from the camera) helps recognising pedestrian, cylcists, *nearby* car,
At that distance, the effect isn't that much of a single light source shone in the general direction of an object from a distance (as is the case with a laser speed trap), but a cloud of lots of small points covering the object (used to recognize shape and size, in addition to distance and speed).
The jammer will probably not see it as a big light exciting its detectors, put occasional small points crossing 1 or 2 sensors at a time. How it will react to this is an unknown to me, but I'll surely it will be more optimized for long range (to predict laser speed traps) and might (correctly) assume this to be useless noise.
I don't own such a jammer (and not that much interested in that technology) so I can't test it.
Also at this distance, the emmiters will be probably seen as individual points of light. That will be probably seen as a couple of bogus point in the point cloud and *should probably* be ignored.
Anyway, to avoid all the LiDAR cars (and other LiDAR based measuring instruments) jamming each-other, LiDAR usually use specific patterns (can't seems to find in the litterature if they are random, or if there's an ID encoded in it) so each LiDAR can successfully recognize it's own dots and ignore other LiDAR's.
The LiDAR *should probably* refuse to take into acount the jammer's emitter as valid dots due to not being part of its specific pattern.
The various LiDAR equipped cars' I've driven have never been jammed by any other light source be it a specific laser jammer or other cars' LiDAR.
Granted, it's possible that I've been just lucky (radar jammers are considered illegal and banned in most european jurisdictions).
But LiDAR-equipped cars are slowly satrting to get more frequent here around and I've never experienced or even heard about a car getting confused.
Last but not least, as I've said in the beginning, cars tend to integrate the input of numerous sensors a
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
There is a reason that military and aerospace equipment is extremely expensive. Would a self driving car be viable if the chassis cost $10k and the sensor suite cost $1000K?
It seems like one good use of this technology would be to integrate it with some sort of HUD overlay on the windshield so that a driver can see things in the dark. I would love to have a car that put red outlines around things like deer in the road or even the lines on the road.
Early prototype and concept cars using Forward Collision Warning Systems mused with the idea of red outlines/augmented reality. /. but I can manage to find the proper link).
(There was even some such prototype mentioned here on
With the idea, e.g.: to have a screen on the dashboard displaying a "night view" of the road ahead with silhouettes outlined.
In the end, I suspected that preliminary research has found it to be too much distracting.
Car currently on the road seem to have gone for much simpler and primitive displays.
e.g: On Volvos, they use simply a line of LEDs behind the dashboard that get reflect in the windshield. You just see it as light bar hovering in front of the car.
It gets progressively brighter orange as object get too much near your car.
If the object is dangerously near the car, the light gets suddenly bright red, about right before the moment where the car will start autonomously hitting the brakes if no other input from the driver.
Seems to work better as, instead of having to recognize complex shapes (that in the end don't give any useful information. You don't mind if it's a deer or if it's a cyclist. you mind if it's in front of your and if your car is on a collision course with it) it's simple light.
And driver have already taken the habit to react quickly to a red light suddenly shining somewhere in front of the car: that's what your peripheral vision perceives when the car in front of you brakes. You have an almost pavlovian reflex to hit the brakes when you see it. Works as well if it's actually a deer and the phantom light comes from your own FCWS/FCAS.
but many of these enhanced features like super accurate mapping of the roads
This has been already debunked by google engineer on TED talks:
self driving cars DO NOT have a supper accurate map of the roads.
For this to work, you need an insanely high level of accuracy in your mapping of roads.
Like down to the precise location of each orange cone on a construction site.
And it needs to be extremely up-to-date. It doesn't matter if a couple of hours ago the lane was free, the important part is that right NOW the lane is closed and your self-driving car needs to avoid it.
At that point, just having an SSD in the car with insanely precise map isn't enough. You'll need to have some online connection, so the car gets continuous update.
And to make this updates you need to gather as much information as you can.
So you actually cover your self-driving car in sensors, so the car can continuously update the map and alert the other cars of any new construction cone or pothole on the road.
At that point, if the car has the necessary sensors to be able to update the maps in real time about any pothole or construction cone, you might as well completely skip the whole online and map part, and directly use the sensors to directly react to the environment around the car.
Self-driving cars (and for that matters, current cars with forward collision warning/avoidance systems, and adaptive cruise control) don't actually count on super accurate mapping of the roads.
They function in a manner very close to their human driver: they use GPS and maps to get the general idea of where they should be going (e.g.: according to the map, in about 2km, I need to continue toward City B. The exit is on the two right-most lanes of the highway), and used their perception of their surrounding for the actual minute details (e.g.: there are cones on the left of me. The lane is closed, I can't go there. Also the vehicle in front isn't moving any more, so I should hit the brakes before rear-ending it).
So in short, a self-drive car's GPS isn't much more different that the mundane one in your car. None of them has a "super accurate mapping of the roads".
But the self-driving car has a huge array of sensors helping to super accurately *see* its environment on the roads.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Cost is an issue but technology capable of acceptable reliability can be done for a few $K now and that price should come down with scale.
You talk about "military and aerospace" and then "acceptable reliability" . Those term are generally mutually exclusive. Today's "acceptable reliability" is not being able to work in rain/snow. There is a big leap in cost/complexity to get over those conditions.
I can finally get my sharks with friggin' Lidar beams!
This comment is my opinion and does not represent an official position of Donald Trump or others I do not work for
IT will still "see" a 100 times better than a meat sack with only 2 often fairly degraded optical sensors.
If information wants to be free, why does my internet connection cost so much?
But that meat sack can interpret what it sees while the sensors have yet to crack that.
WTF? What is it about /. and magic brains. Our brains are not magic, they are not in tune with the force, they cannot see the future. Conciseness is not fucking magic. Computers already do better most of the time. In a very short time they will be better ALL OF THE TIME. Already they are better than any human in snow and rain. Humans are just to fucking arrogant to admit they can't see shit and stop driving. Causing accidents and killing each other.
If information wants to be free, why does my internet connection cost so much?
For example, a computer has difficulty differentiating a brown paper bag from a rock in the road. The former can bet driven over. The latter not so much. Even the Google car requires every traffic control signal to be located and tagged so the vehicle can find the relevant one. If by "magic" you mean "not well understood by science" then the human brain is magic. We have very little understanding of how it works and less on how to emulate it in computers.
Under some conditions computers do perform better but when situations become more complex most humans perform better.
Already they are better than any human in snow and rain.
Not when they can not see the white lane marker or curb as they have trouble knowing where the edge of the road is. Remember recently that the Tesla driver assist was taking every single highway exit. That is because it is following the lane marker and not the road.