3 ms·

With the latest Tesla crash in the news, seeing the point cloud got me thinking. Since mirrors do not create the specular reflections required for LIDAR to wor

With the latest Tesla crash in the news, seeing the point cloud got me thinking. Since mirrors do not create the specular reflections required for LIDAR to work, does this mean a box truck covered in mirrors would be able to render at least the LIDAR portion of an autonomous vehicle useless? Sounds like an attack vector to me.

web007 9y ago
It would only apply to a mirror that's offset to point your LiDAR to the sky or at some odd angle. If it's a flat mirror, it will just see itself approaching from 2x the distance / speed between the car and mirror.
- flyingcircus3 9y ago
  Do I interpret your comment correctly as saying there is some redundant system that is specifically looking for anomalies in sensor readings? Would a top of the line lidar system be able to "understand" that its current inputs were resulting in erroneous outputs? I am reminded of one of the recent Japanese satellites that was effectively an infant mortality, because of some quirk in its redundant systems. I've forgotten the details, but more or less, there were cascading failures across a primary system, its secondary redundancy, and its tertiary redundancy. So the feedback loops designed by the engineers to be negative, and mitigating, ended up being positive, and therefore aggravating, in the cruel vacuum of space. It came down to some error in an orientation sensor, and somehow, the redundant system actually ended up relying on information from the primary system. The boosters designed to slow down the rotation relied on the notion of the satellite's actual rotation, according to the original sensor. As the story goes "Does the machine ever give the right answer given the wrong inputs, Mr Babbage?" Nearly 200 years later, we may finally becoming around to seeing that the woman that asked the question had a point, and Charles simply dismissed it out of hand.
  - web007 8y ago
    I'm only suggesting that mirrors work both ways. If you stand 1m away from a mirror, you will see "yourself" 2m away as a reflection. (1m you:mirror + 1m mirror:reflection) If you move 1m backward, you will see the reflection now 4m away (2m you:mirror + 2m mirror:reflection). Speed works the same way, if you step toward the mirror at 1m/s your reflection will approach you at 2m/s. The same thing applies to LiDAR, either it sees the mirror and calculates the proper distance, or it sees the things reflected in the mirror at 2x actual distance and/or 2x relative speed.

web007 9y ago

It would only apply to a mirror that's offset to point your LiDAR to the sky or at some odd angle. If it's a flat mirror, it will just see itself approaching from 2x the distance / speed between the car and mirror.

flyingcircus3 9y ago
Do I interpret your comment correctly as saying there is some redundant system that is specifically looking for anomalies in sensor readings? Would a top of the line lidar system be able to "understand" that its current inputs were resulting in erroneous outputs? I am reminded of one of the recent Japanese satellites that was effectively an infant mortality, because of some quirk in its redundant systems. I've forgotten the details, but more or less, there were cascading failures across a primary system, its secondary redundancy, and its tertiary redundancy. So the feedback loops designed by the engineers to be negative, and mitigating, ended up being positive, and therefore aggravating, in the cruel vacuum of space. It came down to some error in an orientation sensor, and somehow, the redundant system actually ended up relying on information from the primary system. The boosters designed to slow down the rotation relied on the notion of the satellite's actual rotation, according to the original sensor. As the story goes "Does the machine ever give the right answer given the wrong inputs, Mr Babbage?" Nearly 200 years later, we may finally becoming around to seeing that the woman that asked the question had a point, and Charles simply dismissed it out of hand.
- web007 8y ago
  I'm only suggesting that mirrors work both ways. If you stand 1m away from a mirror, you will see "yourself" 2m away as a reflection. (1m you:mirror + 1m mirror:reflection) If you move 1m backward, you will see the reflection now 4m away (2m you:mirror + 2m mirror:reflection). Speed works the same way, if you step toward the mirror at 1m/s your reflection will approach you at 2m/s. The same thing applies to LiDAR, either it sees the mirror and calculates the proper distance, or it sees the things reflected in the mirror at 2x actual distance and/or 2x relative speed.