Google’s can be stopped with a t-shirt that has a stop sign on it. I suppose adversarial research is only just starting for all this and many holes will be blown through this tech in the coming years. Especially through the single modality systems like Tesla’s.
That would be great on my bike jersey. Not a stop sign, but a 30 kmh speed limit sign, as all the new cars sold in the EU starting with 6 of July 2024 come with ISA, mandated speed controls.
> The boffins suggest that the attack could be carried out with drones, which serve to "hide" a secondary vehicle from the victim AV by projecting or carrying the adversarial object.
It's an interesting topic to explore, but for practicality I feel as though at the point where you're flying a drone (carrying an object specifically designed for the car's sensor system) between the autonomous vehicle and a vehicle you want to hide in hopes of causing a crash, you've already passed up countless simpler and less conspicuous opportunities to cause problems. Reminded a bit of: https://xkcd.com/1958/
We live in a world where people will call an armed swat team to someone's house because they don't like their twitch stream. The easier it is to troll from afar, the more people will do it.
I don't think the level of involvement and conspicuousness required for this attack makes it any easier to cause crashes (even from afar) than it already was with human drivers, like by messing with the vehicle's brakes while it's parked, or using the same drone to drop a metal object through the windshield from high up.
In other words, robo vehicles are less able to discern that someone is messing with them. The attack they are discussing would be akin to the drone carrying an image of a piece of empty road. A human would recognize the scene didn't fit together, the robot does not.