MIT shows ground penetrating radar for autonomous vehicles in snow and ice | Hacker News Reader

MIT shows ground penetrating radar for autonomous vehicles in snow and ice

ll.mit.edu

90 pointsrmason36 comments

I think for those of us in the snow belt this could potentially be great news. I first became aware of ground penetrating radar when I worked with remote sensing in Michigan where clouds often obscured satellite photos of farm fields. The military had the capability to look through the clouds but unfortunately the technology never became inexpensive enough for agricultural use.

People may not realize that when it snows the landscape dramatically changes for the sensors on the Google cars and they can't follow the map and suddenly they're quite literally lost.

What do you do when you can't see the lines on the road? Or when cars are suddenly four abreast on a three lane highway? Or things like the especially treacherous black ice. So I am encouraged that it may be possible for us in the great white North to use our self driving cars year round.

> The military had the capability to look through the clouds but unfortunately the technology never became inexpensive enough for agricultural use.

How do the economics of this work out? It seems strange that anyone can afford to spend tens of millions on a satellite launch, but can't afford to put fancier equipment on the satellite. Maybe the penetrating radar apparatus is much heavier and so more costly to launch?

> Maybe the penetrating radar apparatus is much heavier and so more costly to launch?

Radar is power intensive. As a commercial operator, you this weigh launch cost, recharge time, and resolution. I've seen implementations where an hour of charge bursts to minutes of capture.

> What do you do when you can't see the lines on the road? Or when cars are suddenly four abreast on a three lane highway? Or things like the especially treacherous black ice.

Indeed, these things are also quite tricky for human drivers (Winter is a windfall for body shops!).

This reminds me of a researcher I met, they were proposing to add a vast array of sensors to each wheel of a military vehicle that had at minimum 8 wheels. Every wheel would have rotation sensors, torque sensors, vibration sensors, a camera looking ahead, a camera looking behind, lidar looking ahead, lidar looking behind, and ground penetrating radar.

The idea being that you could know exactly what sort of terrain you were on and what terrain you were going to hit so an active suspension could compensate. The case of the sensors looking behind was interesting, the idea there is that you'd measure how much you deformed the terrain to better understand what you were moving.

Oh and said military vehicle was going to be omnidirectional with each wheel capable of independent rotation, steering, and height change.(for the active suspension)

Some cars with active suspension sort of do this. The front shocks tell the rear ones what is coming and they adjust their properties to compensate.

Mercedes was apparently the first to do this with a camera in their "magic body control" suspension system.

http://youtu.be/ScpgI1w5F6A

Related, hilarious Mercedes chicken ad: http://youtu.be/TzhLIc6c4DU

Mercedes was the first to commercialize it but Bose had an earlier implementation and... the crazy part. It would jump over stuff.

https://www.youtube.com/watch?v=z3gX2HwFf5I

The bose based suspensions ? I wonder how many got sold.

No, even things like the new Mustang GT350. Basically anything with active suspension that a lot of performance cars now have.

Another fine example of the role of federal military funding in high tech research. Makes me chuckle whenever I hear a person employed in Silicon Valley refer to themselves as a "limited-government Libertarian". We're all sucking on the government's teat, buddy.

I doubt there's very many libertarians that would claim the private sector doesn't benefit from government spending at times. Every complex issue has it's nuances.

Generally, the stance of libertarians is that the TCO associated with the military industrial complex FAR outweigh it's benefit.

"At times" grossly understates the role of DARPA and other government support in creating technologies commercialized by Silicon Valley. The entire industry was a direct product of public investment and most key innovations are a culmination of long-term sustained taxpayer-funded research.

Take for example autonomous driving. It's a product of DARPA. See https://en.wikipedia.org/wiki/DARPA_Grand_Challenge

Even Siri on your iPhone is a result of decades of public investment in AI. See https://www.sri.com/work/timeline-innovation/timeline.php?ti...

Need I even mention the Internet itself..

And there are drones killing civilians in the middle east. Though we benefit greatly in the private sector from this relationship, some of us consider the harm it creates much greater.

I had a very interesting talk with Vint Cerf about this a while back btw. It's a sad situation to be in.

Does he have any public comments on the matter? Would appreciate the link.

I'm not sure. We had a good conversation of the good and bad of working with groups like the DOD. He's happy to be with Google and recognizes the benefit of working with the DOD, but not exactly pro military.

I can benefit indirectly from military spending, and still feel that I would happily sacrifice that particular benefit to be free of the costs of military spending.

You would sacrifice pretty much the entire high tech industry? That's bold. Because that is what it would take, given the central role of DARPA and government spending in creating and sustaining it.

Note that I am with you on wanting an alternative to military spending. My point is that SV "Libertarians" have for the most part made their money on the basis of massive taxpayer support.

Whether this pans out or not, it is technology like this that will eventually ensure that autonomous vehicles are far superior to flawed human drivers. The only question is when exactly that will happen. I think everyone now knows that it's within the next 20 years, but does that mean 1 - 2 years, or does it mean 10 - 15 years?

This is very interesting work, as many weather conditions still pose big challenges for autonomous vehicles.

I am curious about how practical it will be to maintain data for all roadways. Though deep sub-surface features may be stable over time, it would seem the actual roadway markings may change requiring careful updating.

Question: What happens if there are many cars all using this system at once? Is this sort of radar subject to interference from nearby signals?

What if the lines never existed in the first place?

At the end of the day, it is still a human that "paints" the line on the road.

So, how would it handle this, under snow?

http://media.nbcwashington.com/images/1232*675/restriping+wi...

I don't think it's looking at paint. It's looking for road edges.

This also seems like a fantastic addition to snow plows. They could even serve as a great testing platform--still human driven, but with tons of real world snowy road data. Fast forward a few years and autonomous snow plows will be really nice--an army of plows working through the night and even during the worst storms.

The only scenario where I can see this failing is when snow covered lane markings have been temporarily redrawn, eg. for nearby construction sites. Then you have to hope for accurately placed road markers. But that would be a similar challenge for humans as well.

Does this mean that if you stand next to the car, you are being X-rayed? :)

> "They rely on optics to 'see' lane markings, road surface maps, and surrounding infrastructure to orient themselves. Optical systems work well in fair weather conditions, but it is challenging, even impossible, for them to work when snow covers the markings and surfaces or precipitation obscures points of reference."

Humans drivers are a pretty glaring counterexample.

Humans are challenged, too.

Just a thought, could autonomous cars be programmed to eliminate the need for road salt?

Seems extraordinarily complicated.

Not really -- it's a straightforward and obvious application of image correlation. Nothing exotic about the hardware, either.

It certainly shouldn't have been patented. The first claim is as broad as a barn, basically giving MIT ownership of an entire problem space.

I didn't really mean the technology itself, more the logistics of actually making it work at scale.

Something like an embedded wire in the centerline of each lane that could be followed by a simple metal detector seems a lot simpler.

That's how GM did automatic driving for Firebird II and III in the 1960s.

Volvo's solution for snow is to drive magnets in the form of nails into the pavement along lane centerlines. They suggest that this can provide guidance for snowplows as well as assist autonomous vehicles.

Lane marking in areas of heavy snow is hard. Some areas put up plastic posts along the roadside. Japan puts up arrows hanging over the road in parts of Hokkaido.

The thing about this is, you don't have to mess with the roadway at all. It relies on the fact that subsurface features beneath the pavement don't tend to change much over time.

Embedding wires or magnets or whatever in every paved street in the world would be MUCH more complicated, not to mention economically disruptive.

Embedding a wire in millions of miles of road seems way more complicated than this.

Reply on news.ycombinator.com