That's a problem with simple, cheap Doppler radars. Pulse radars are much less jammable and countermeasures against jamming are well understood. A bit of random jitter in the outgoing pulse timing prevents interference from identical systems, as well as synchronization attacks. Military radars have had this since the 1960s.
It does add cost.
LIDAR has the same problem and the same countermeasures. Pulsed LIDAR units should have a few microseconds of random jitter in the pulse timing, too. You can still get a collision, but not multiple consistent collisions in a row, so you know it's noise.
Spoofing radar signals isn't exactly new, and the ability to do it on the ground (so to speak) shouldn't be surprising. The military has been doing it for decades.
Also, it doesn't take nearly the sophistication of the system demo'ed here -- a chaff canister released into traffic would, I imagine, play twenty kinds of havoc on any autonomous driving system that relied on radar.
> Spoofing radar signals isn't exactly new, and the ability to do it on the ground (so to speak) shouldn't be surprising. The military has been doing it for decades.
People who are interested in finding out more can look up Digital Radio Frequency Memory (DRFM) jammers.
I don't mean to sound pessimistic, but the link's assertion that any of these techniques are new suggests that either allaboutcircuits is not familiar with radar/electronic attack or that Duke University and/or the automative radar folks are not up to speed with techniques used in radar/electronic attack in the defense/aerospace industries. Maybe it's the latter, as the arXiv preprint states "...show the novel ability to effectively ‘add’ (i.e., false positive attacks), ‘remove’ (i.e., false negative attacks), or ‘move’ (i.e., translation attacks) object detections...".
The arXiv preprint doesn't list any of the usual suspects for radar or electronic attack sources that I would expect to see in its references. There are a lot of automotive radar sources and, interestingly enough, some LiDAR and LiDAR adversarial attacks instead.
A large cloud of glitter would do the same for optical systems, e.g. human drivers. Shining intense light at drivers is also known to cause havoc and incidents. Now imagine a night road lit by a dancing hall light show; many human drivers would find themselves discombobulated.
Even ignoring the possibility of intentional jamming, I wonder if the increase in radar-equipped cars will start causing significant interference in other cars' operations. Every once in a while I get a phantom "brake!" signal when there is nothing in front of me.
I think what's important here is not this first low-level stab at manipulating car radar systems. It makes a beach head on what adversaries can possibly do and demonstrates that manufacturers need to try harder.
What's funny is that so much effort is going into these radar systems for non-autonomous vehicular systems that can be resolved by teaching people how to correctly set their side mirrors to cover their blind spots. You eliminate your blind spots by moving your side mirrors out enough that you can't see the sides of your car without tilting your head a little. You don't need to see your own car in your side mirrors.
After I got my most recent vehicle I discovered it has a 24GHz rear-facing radar in order to alert me to a possible car in my blind spots. It misfires all the time and alerts me to guard rails and phantoms that don't exist. I don't need it, because I now keep my side mirrors set correctly I know where cars are beside me before the radar knows. I turn it off but now I'm curious if that actually deactivates the radar emitter or just stops alerting me to hits.
> I don't need it, because I now keep my side mirrors set correctly I know where cars are beside me before the radar knows.
It's not there to replace you looking. It's there as an extra information source in case you make a mistake when looking. Maybe you're lucky with the your car, but even with curved ends of mirrors in mine, there's still a blind spot where I would not see a bike on the passenger's side and the little extra notification is great for that. It blinks for guard rails too, but that doesn't bother me - there's extremely rarely a reason to check the mirror on the side close to the rail.
I got a Volkswagen with front assist and adaptive cruise control. I use it mainly for long runs with little traffic, that way my foot doesn’t get cramped for staying at the same position.
When crossing under some bridges, sometimes it decelerates and warns me of a collision, even if there is no one there. I have to override it by stepping on the gas pedal until I’m through.
I’ve also felt it hiccup on a test strip where they were testing radars, I was guessing it was due to them using similar frequencies, but I’m not sure on that one.
Worker attrition is way too high nowadays to keep deep knowledge in niche fields I believe. Old school seniors have retired and are being replaced by nobody.
You can notice this in a lot of hardware products with "noob" mistakes that the prior product generation did not have etc. Software is probably the same but the product is too opaque to see into so ot just feels crappy.
Realistically, it's been possible on and off for quite some time. Chrysler's UConnect system received quite a bit of flak some years back over just how easy it was to take over full control of a vehicle[0]; not just locking/unlocking doors, but full control over the climate system and wipers, all the way up to starting and stopping the engine or changing gear in the transmission.
It goes the other way too. My city has installed a large number of radar detection cross walk lights. You can tell when a car with radar goes down a street because they progressively trigger each cross walk light despite there being no people. There are certain times a year when the sun is in just the right position in the morning to cause them to stay permanently on for hours as well.
Did you know that fully analogue human-piloted cars are vulnerable? Researchers have recently described the "MadBrick attack": a malicious attacker can remotely disable a human-piloted car or cause it to veer off road by dropping a brick through the windshield from an overpass. Clearly, car manufacturers need to promptly address this class of vulnerabilities.
This has been addressed already, though not by manufacturers. If you live (or drive through) an area where brick attacks are known to happen, you just put protective films on the windshield. It usually still ends up damaged but you can continue to drive just fine (I know because I have).
There are some really ingenious ways to make a self driving vehicle that won't collide with other vehicles. But they all end up just becoming a train/trolley.
Does the trick with tape on the ground causing Teslas to steer into walls still work or did they patch that for all the ignorant alpha testers out there?
Animats|2 years ago
LIDAR has the same problem and the same countermeasures. Pulsed LIDAR units should have a few microseconds of random jitter in the pulse timing, too. You can still get a collision, but not multiple consistent collisions in a row, so you know it's noise.
GlenTheMachine|2 years ago
Also, it doesn't take nearly the sophistication of the system demo'ed here -- a chaff canister released into traffic would, I imagine, play twenty kinds of havoc on any autonomous driving system that relied on radar.
derstander|2 years ago
People who are interested in finding out more can look up Digital Radio Frequency Memory (DRFM) jammers.
https://en.wikipedia.org/wiki/Digital_radio_frequency_memory
I don't mean to sound pessimistic, but the link's assertion that any of these techniques are new suggests that either allaboutcircuits is not familiar with radar/electronic attack or that Duke University and/or the automative radar folks are not up to speed with techniques used in radar/electronic attack in the defense/aerospace industries. Maybe it's the latter, as the arXiv preprint states "...show the novel ability to effectively ‘add’ (i.e., false positive attacks), ‘remove’ (i.e., false negative attacks), or ‘move’ (i.e., translation attacks) object detections...".
These are open data, too. See https://en.wikipedia.org/wiki/Radar_jamming_and_deception.
The arXiv preprint doesn't list any of the usual suspects for radar or electronic attack sources that I would expect to see in its references. There are a lot of automotive radar sources and, interestingly enough, some LiDAR and LiDAR adversarial attacks instead.
nine_k|2 years ago
lend000|2 years ago
dosman33|2 years ago
dosman33|2 years ago
https://www.caranddriver.com/features/a15131074/how-to-adjus...
After I got my most recent vehicle I discovered it has a 24GHz rear-facing radar in order to alert me to a possible car in my blind spots. It misfires all the time and alerts me to guard rails and phantoms that don't exist. I don't need it, because I now keep my side mirrors set correctly I know where cars are beside me before the radar knows. I turn it off but now I'm curious if that actually deactivates the radar emitter or just stops alerting me to hits.
viraptor|2 years ago
It's not there to replace you looking. It's there as an extra information source in case you make a mistake when looking. Maybe you're lucky with the your car, but even with curved ends of mirrors in mine, there's still a blind spot where I would not see a bike on the passenger's side and the little extra notification is great for that. It blinks for guard rails too, but that doesn't bother me - there's extremely rarely a reason to check the mirror on the side close to the rail.
kolinko|2 years ago
Each time I visit US I'm frustrated by the american cars having flat mirrors. The blind spots are insane. I guess it's due to some weird regulation?
In EU mirrors have their last 2-4cm curved, and there's virtually no blind spot, and you don't need to adjust mirrors too precisely even.
chinathrow|2 years ago
trklausss|2 years ago
When crossing under some bridges, sometimes it decelerates and warns me of a collision, even if there is no one there. I have to override it by stepping on the gas pedal until I’m through.
I’ve also felt it hiccup on a test strip where they were testing radars, I was guessing it was due to them using similar frequencies, but I’m not sure on that one.
iknowstuff|2 years ago
numlock86|2 years ago
Amazing. We've gone full circle now. I wonder if this is a problem of generational differences in knowledge transfer ... or rather the lack of it.
rightbyte|2 years ago
You can notice this in a lot of hardware products with "noob" mistakes that the prior product generation did not have etc. Software is probably the same but the product is too opaque to see into so ot just feels crappy.
Metacelsus|2 years ago
blibble|2 years ago
how long until smart 12 year olds can remotely hijack major car brands on a freeway with a laptop and a small python script?
FractalParadigm|2 years ago
[0]: https://jalopnik.com/chryslers-uconnect-vulnerable-to-remote...
superkuh|2 years ago
tetromino_|2 years ago
fipar|2 years ago
snapplebobapple|2 years ago
csours|2 years ago
foxyv|2 years ago
Steering? Rails.
Anti-collision? Signals.
Fuel/Charging? Overhead power lines.
ck2|2 years ago