Researchers warn robot cars can be crashed with tinfoil and paint daubed on cardboard A team of researchers from prominent universities – including SUNY Buffalo, Iowa State, UNC Charlotte, and Purdue – were able to turn an autonomous vehicle (AV) operated on the open sourced Apollo driving platform from Chinese web giant Baidu into a deadly weapon by tricking its multi-sensor fusion system, and suggest the attack …
A piece of cardboard and aluminium foil is very likely to be almost completely destroyed, and the wreckage overlooked in a subsequent collision investigation.
Even if identified as the cause, there won't be purchase records etc, unlike wreckage of a drone that may have traceable serial numbers.
Dropping something made with bits from the local supermarket from a drone is also very cheap, especially as you get to use the drone more than once.
A "kick me" sign is much cheaper than actually kicking someone.
I don't think deniability would be feasible, not after the first couple of times this happens. The robo-car camera will be constantly recording the last few minutes to something that is likely to survive a crash. I'd be mildly surprised if this wasn't the case already.
From the same drone, you can drop a rock through the windshield, or a balloon full of paint, or shine a strong light in the driver's face. That's just off the top of my head, and it's all easy to find stuff that doesn't require crashing the drone itself. None of that is a guaranteed kill, obviously, but neither is getting rear-ended by a robo-car.
All I'm saying is, these studies are interesting and worthwhile, but if the premise is that someone wants to kill me and has the ability to physically mess with my car at speed, I'm in really big trouble already, regardless of who or what is driving. There was a time we had assholes dropping rocks from highway overpasses for kicks, adn people got killed, and that was before drones or robo-cars.
Reports of remote software-only hacks are much more worrying, and they aren't exclusive to robo-cars either.
..Alternatively, the drone can cover up the windshield of a fast driving car, blinding the driver and thus provoke a crash.
....Please fund my next paper...
I need money for 10 cars of various types, 10 drones, 10 cans of spray paint, some tarp and some flash-bang grenades. I also wish to rent a part of the M25.
I mean like the Citroen 2CV.
If you ever see one with a dent perpendicular across the bonnet, then it will have had a very good driver.
A fault in the latching mechanism meant that the bonnect could come loose and open while the car was driving at speed (max. ca. 90 kph* for the "large" engine with 32 hp if I recall correctly).
This would leave the driver with no forward view (keep in mind that the side-windows could also not be fully lowered, just folded down halfways). A good driver would keep the steering centered and apply brakes without crashing.
*: At which point, the un-supported upper side-walls would vibrate at about 2 Hz with an amplitude of about 5 cm. ...Good times :)
It happened to me in a Peugeot 406 (although many times before with a 2CV as well). Seems like the last time oil was checked, the bonnet didn't correctly lock itself on closing.
Fortunately, I wasn't fully engaged on the highway and could gently slow down and stop on the emergency lane.
My three passengers looked quite pale and aghast by the time we reached stand point. I smiled at them and gently asked : "Hey, don't tell me your never drove a 2CV before !"
"While Baidu Apollo platforms were used in the attack, the attack strategy could theoretically be applied to other multi-sensor fusion systems."
I'm curious was there a reason researchers in America chose to do their research on Baidu which is based in China? Was it fear of litigation in the US? Something like this could hit shareholder and investor confidence. As for mitigation I'm not too clued up on the technology but couldn't adjustments be made to the standard camera because it's not like it's invisible to one or is it?
"I'm curious was there a reason researchers in America chose to do their research on Baidu which is based in China? "
If you're suggesting this is about economic nationalism, you might care to follow the link to the paper abstract and note the authors' names. That doesn't disprove any economic nationalism theory but it's interesting to observe they all appear to be of Chinese heritage.
"Have they done the same experiment on human drivers?"
They hardly need to. Observation suggests many human drivers' attention is easily snatched away by flashing lights, attractive members of their preferred gender, weird clothing or hair, etc etc. A few bits of tinfoil and colour, less so (perhaps unless displaying an inflammatory slogan).
Mind you, you'd have thought that the safety debate on autonomous taxis would be easily solved, by comparing the safety stats for Waymo and Baidu against the accident data for meatsack taxis in the same operating areas (subject to the usual caveats on data validity). The data exists for both, you have to wonder why regulators in both the US and China haven't done this work and published the results.
I can crash a car with one snip of a brake cable. Ivy league only. I don't consort with riff raff.
No wonder the UK government are wiping most of the uni sector out if this is the sort of thing they do nowadays. I remember when academics used to build faster computers, create new computer languages, invent entirely new tech and push the boundaries in their fields. Now they are the tenure tracked wing of 4chan.
Fund my paper:
“If I fly a drone with ‘the brightest flashlight in the world’ in front of a human driven car and then shine the light at their eyes - they might crash”
Then you can fly away and leave nothing behind but carnage.
Seriously - there’s a reason that there aren’t more terrible car/airplane/bike crashes, most people are not horrible arseholes.
This post has been deleted by its author
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
