Self-Driving Cars: How Much Interference Between Detection Technologies?

Factual Questions
1

This has been mentioned a couple times in relation to the fatal Uber AV accident in Tempe, AZ.

Exactly what potential exists for interference between the detection technologies used on one AV and those used on the other AVs sharing the road? It seems to me that the active detection technologies (traditionally these are laser, microwave, and ultrasonic) have a great deal of potential to affect the performance of other AVs using the same type of equipment. This would not be true of technologies like intelligent video or infrared.

I can easily envision twelve or more cars facing each other at an intersection and all twelve are aiming their active sensing systems at the same small area and at each other. I doubt that twelve identical radar guns would work properly in these circumstances.

I know that military radar systems can switch frequencies or operate in a pseudorandom manner, but is this the case for AVs?

(Yes, I doubt that AVs will use ultrasonic detection, but it is traditionally a technology used for motion detection at short range.)

2

There is massive potential for self-driving cars to communicate/to share information. So the massive pile-up would be a thing of the past. And the twelve self-driving cars at the intersection would be telling each other exactly where they are.

3

No autonomous piloted road vehicle that I am aware of uses radar specifically for the reasons you mention. Some combination of machine vision, LIDAR mapping, and Assisted GPS is used to determine relative position and build an internal representation of objects and potential hazards.

Although a “massive potential” for intercar communication exists there is currently no protocol or standard to support it, and no one seems to be working toward it currently. And it wouldn’t work with unaided human drivers, so to really achieve its potential would require eliminating or act least a massive reduction in unaided human-piloted vehicles.

Stranger

Stranger

4

I appreciate the quick responses, but…

It is my understanding that LIDAR and some additional technologies are used specifically for detecting objects in the road. That’s why I mentioned Tempe. GPS does nothing to help in this regard.

Is it the expectation that all AVs will communicate about the technology they are using (e.g., LIDAR) and coordinate the frequency or timing of the emitters? That seems horrendously complex. Again, twelve AVs sitting at an intersection would ALL presumably be using their LIDAR (or other technology) to ensure that crosswalks are clear.

I’m not trying to say that this can’t be resolved, but (1) is it a real problem, and (2) what is the current thinking/strategy?

5

I don’t think there is any real strategy regarding intercar communication at this point. Autonomous vehicles will have to deal with human-piloted cars and other road hazards so the technology has to be able to work independent of any communications technology regardless. LIDAR is not subject to the kind of significant interference problems as radar is, although it may be severely limited by rain or mist, and of course if the sensors are covered by mud or snow.

One potential problem with future intercar coordination is security and authenication; allowing for this capability obviously offers up the potential for spoofing and other deliberate interference, so some protocol would need to be established to assure signals are valid and secure.

Stranger

6

According to this Techcrunch article, the Uber cars use a combination of lidar, radar, visual spectrum cameras, and GPS. I know little about radar - is this standard radar that could have the interference problems being discussed, or are there different kinds?

7

My understanding is that LIDAR is directional in that it only accepts signals that were received along the exact same path the laser was sent out from. So the only danger would be another laser being sent along the exact path and that would only block the LIDAR for the duration that the laser is aligned so as long as one of the vehicles was moving it would not be a problem.

8

Radar isn’t unique to autonomous cars. Many cars already use radar for collision avoidance and autonomous cruise control. If interference were a problem, I think it would have become a problem already.

9

I don’t want to appear argumentative, but according to this:

https://www.extremetech.com/extreme/189486-how-googles-self-driving-cars-detect-and-avoid-obstacles

“Google’s driverless car tech uses an array of detection technologies including sonar devices, stereo cameras, lasers, and radar.”

Granted, this article is 3-4 years old and deals only with Google AVs, but three of the technologies mentioned are active (relying on energy introduced into the environment for operation). LIDAR relies on laser scanning, but I’ve never seen any information on whether or not having a dozen of these units operating in close proximity to each other and in a reflective environment is an issue. This was the point of my OP.

10

I’m glad people already corrected the misstatement that autonomous cars don’t use radar. It seems to be part of the suite of sensors for most autonomous vehicles and it is a key factor in today’s radar-assisted cruise control and automatic braking.

I think it depends on how you define “real strategy.” The NHTSA put out a Notice of Proposed Rulemaking on Vehicle to Vehicle communication and they want to promote standards. Auto industry suppliers are working on their own efforts but they lack the ability to force the industry to harmonize by themselves. Automakers, like Audi, are adopting limited amounts of infrastructure-to-vehicle communication, which allows traffic lights to tell cars when the lights will be changing. This is a trivial application of the communication protocols necessary for V2V communication but every great accomplishment starts with a small step.

The NHTSA recognizes the need to work on security for V2V comunication.

PDFs:

11

To the OP:

First of all, yes, robustness to interference is absolutely a design criterion for these systems. Interference is absolutely a problem, but not as big a problem as you might imagine.

You refer to an intersection at which 12 cars are “aiming their sensors at the same small and at each other.” You may well already know this, but active sensors aren’t “aimed.” They’re scanned. That’s why you often see spinning radar emitters at airports and atop ships–they’re constantly emitting and then sensing what comes back.

So even if all 12 cars’ lidar sensors are scanning simultaneously, no one car’s beam will be pointed at another car’s sensor for more than an instant. And of course, they’re not using LIDAR “to scan the crosswalk for pedestrians.” They’re continuously canning in all directions at all times.

And for LIDAR in particular, this is a solved problem. When a soldier points a laser designator at a target, he’s essentially marking it with a very fancy laser pointer. A plane above can then release a laser-guided bomb that “looks” for the spot created by the soldier’s laser designator. When it sees the bright laser spot, the bomb guides itself to it, at which point destruction ensues.

But it wouldn’t take much to confuse a simple laser-guided bomb…all you need to do is create a bunch of laser spots all over the battlefield, and then the bomb can’t decide which one is being projected by the soldier with the designator. Some people consider disco balls offensive, but not as many understand that they can also be defensive. :wink:

Anyway, a laser designator typically doesn’t shine a steady beam, but rather one that pulsates in a pre-determined sequence–one the bomb can validate as authentic. The bomb can then ignore any spot that doesn’t pulsate with the right “code.”

In other words, you can encode information in your laser beam. So a car’s LIDAR system might encode what’s called a Globally Unique IDentifier (GUID) so it can always tell what signals are reflections of its own emissions. And with a GUID, the emphasis is on “globally unique,” not “identification.” An AV using LIDAR isn’t broadcasting any information that anyone could use to trace that car back to a registered owner or something like that. That’s what license plates are for.

Sonar’s range is so short (in air) that there’s no need to encode an identifier. Radar is more of a challenge, but there are a number of other ways to minimize interference, including encryption.

So many cars can use LIDAR simultaneously without interfering with each other, so there’s no need for inter-car communcation to solve that problem.

Autonomous vehicles tend to be the most robust when they’re integrating information from many different kinds of sensors. Some sensors overlap in what/where they can “see,” so if five or six different sensors using two or three different sensing technologies all agree that you’re about to hit a stopped car, the computer can begin braking with a high degree of confidence that there’s actually a car there (as opposed to a weird reflection or a shadow).

When AVs start sharing sensor data with each other, they’ll be able to confirm that their “map” of the space nearby is accurate. Essentially, this is one person nudging another and asking, “do you see what I see?”

12

I should have said that radar is not used for forming a detailed model of the exterior world. It is used, of course, for close range collision detection but cannot form images of sufficient fidelity to recognize specific forms or objects, hence why LIDAR technology is so important, though it too has its limitations. LIDAR beams can be modulated so that each car will recognize return from its specific beam, so interference isn’t a particular issue. The biggest problem with LIDAR isn’t reliability or resolution, but the costs of the system, which have been steadily decreasing with each evolution of the technology.

Stranger

13

What I’d like to be would be a government-imposed requirement for transponders on all vehicles on the road. What exactly the transponder would transmit would depend on the age of the vehicle, and the technologies implemented in it. At the very least, every single car would be required to send a signal that can be used to pinpoint its location, and which would state the car’s maximum performance parameters: Maximum acceleration in speed-up, braking, and turning, minimum turning radius, and so on. If the car also has powered controls like power steering, it’d have to also transmit the input the driver is giving to them. If the car has cruise control, and the cruise control is set, it’d transmit the set speed. If it has an integrated navigation system (OnStar or the like), then it’d transmit the route that it’s currently directing the driver along. And of course, if the car has any autonomous driving features, it’d transmit what it’s doing with those, too, some amount ahead of time whenever possible. And of course, the same regulation which established the requirement would also set standards for how the signals are transmitted (which would be decided in collaboration with industry).

This would, I think, be an easily-borne burden, but it would give autonomous vehicles a lot more information about what other vehicles will do, are likely to do, or at least might do.

14

The problem with this is that it would be straightforward to hack it so that a car’s transponder lies. But not at all straightforward to design an autonomy suite to always gracefully handle “the other car lied”.

15

It’s also straightforward to hack a human-driven car so that it runs down people in a crosswalk instead of stopping for them. All you have to do is, when you’re approaching an intersection, hit the gas instead of the brake. The proper response to that is to charge the person who did it with whatever crime is appropriate.

See XKCD.

16

Thank you, EdelweissPirate (and everyone else). That’s a nice explanation.

I was just envisioning twelve (or more) AVs simultaneously scanning an intersection with whatever technology they use and that it seems likely that there would be some lasers (or radar or whatever) energy emitted directly at the other cars from time to time.

I would be curious if any group that is developing AV technology has tried simulating this by placing twelve (or more, since some AVs use as many as four LIDAR units each) in a similar configuration and letting them “go at it” for a few hundred trials. With as many intersections as there are and with the repetition of stoplight sequences hundreds of times a day at each one, even a very low frequency of interference problems would presumably show up on a regular basis.

17

I agree this is true for now, but it may end up being more ambiguous than this, as some people are attacking AV AI based on a “sensory fusion” approach that will be, essentially, neural networks run on all the inputs (including radar). I’m not a fan (it’s pretty much a black box), but I can see the intuitive appeal.

But back to the OP’s question, if all the cars in an LA traffic jam are chirping out radar at the same time, or sweeping their lidars, is there any meaningful chance of interference? Reading the thread the consensus seems to be “not”.

18

Oh, absolutely there will be interference. But things like randomized chirp patterns reduce the chance that any single pulse or lidar point was interfered with.

But there absolutely is noise in the data. That radar histogram will have peaks in it that are reflected energy from the radars of other cars. There will be a few points in the lidar point cloud that are dead wrong. This is true whether or not the car is by itself.

The system then has to filter, removing and smoothing these extra peaks. They do affect the consensus mean but not very much.