Jack Ganssle, Editor of The Embedded Muse Jack Ganssle's Blog
RSS Feed This is Jack's outlet for thoughts about designing and programming embedded systems. It's a complement to my bi-weekly newsletter The Embedded Muse. Contact me at jack@ganssle.com. I'm an old-timer engineer who still finds the field endlessly fascinating (bio).

For novel ideas about building embedded systems (both hardware and firmware), join the 40,000+ engineers who subscribe to The Embedded Muse, a free biweekly newsletter. The Muse has no hype and no vendor PR. Click here to subscribe.

LIDAR in Cars - Really?

February 11, 2019

Pretty much all of the candidate autonomous cars driving around sport a LIDAR sensor mounted like an old-fashioned "cherry top" on a police vehicle.

A recent article in Electronic Design (by, of course, a LIDAR vendor) gives 11 myths about LIDAR in automotive applications, The default assumption in that article, and elsewhere, is that LIDAR is a critical component for autonomous cars.

Why?

It seems to me that LIDAR has some devastating flaws in this application. Like how do you put a car top carrier over the LIDAR without compromising its field of view? Where will that surfboard go?

The cited article claims that as its use becomes more widespread costs will fall to "a couple of hundred dollars." That's an awful lot of money for a single component in a mass-produced device like an automobile. Realistically, if LIDAR is indeed so critical to the safe operation of a car then redundancy will be needed. When one sensor assembly starts to get anywhere near the thousand-dollar mark it will be uneconomical in a $20-$40k product. Toss in redundant computers, RADARs, other sensors and actuators and one wonders if there will be a budget for tires and the engine.

We do know of a sensor that, after over a century of being integrated into cars, works pretty darn well: eyeballs. Most of us have two, which gives decent depth perception. I'd wager that absent stupid driving behavior (e.g., excessive speed, tailgating, and the like) those two optical sensors are just about perfect for automotive applications.

Sure, there are situations where visibility goes to near zero, but LIDAR would be just as blind. The article I mentioned says "A camera becomes rather useless when there isn't enough ambient lighting; and when it's dark, a camera can only see as far as the headlights, which can be quite dangerous." Yet the headlights can see far enough ahead to stop the car safely; just how far down the road must one peer to be safe?

I think cameras, perhaps coupled with RADAR to measure closing rates (today's 77 GHz RADARs cost little), are the ideal sensors for autonomous cars. They're cheap: Digi-Key lists about 100 that go for under five bucks. In automotive quantities that would probably fall to tens of cents. They're so cheap a car could be infested with them, surveying pretty much every angle. One could cluster sub-par cameras each of which is optimized for different conditions: One for bright lights, another for roads so dark they're horror-film fodder.

Though our eyes are situated only inches apart they provide lots of depth perception. Why not position cameras a meter or more apart in a car, yielding an incredible amount of depth information?

The problem with cameras is they generate a flood of data. But that's not intractable. Consider mobile phones: the SoC has hardware designed to analyze the visual field to extract faces and other features.

Long ago when a 32 KB SRAM was expensive a semi vendor told me to design boards with JEDEC outlines for bigger parts. Semiconductor prices fall and planning around today's prices is sometimes unwise. Similarly, the cost to process visual data has been tumbling, and will certainly continue to fall.

For some insight into how the microprocessor world evolved, see this.

Will cameras do well in all driving conditions? Of course not. Black ice or zero visibility will always be dangerous. Perhaps future smart cars will need a moron mode which determines the only safe option is to stay in the driveway or pull over.

(I've long been enthusiastic about autonomous cars, but lately am having some second thoughts. How does a car understand that it makes sense to obey the cop waving one through an intersection where the light is red? What if there are several police officers handling traffic from multiple directions? Or how to merge from a 12-lane tollbooth into three lanes where there are no painted guide lines? Though I'm sure these sorts of problems will be solved, they are very complex.)

Feel free to email me with comments.

Back to Jack's blog index page.

If you'd like to post a comment without logging in, click in the "Name" box under "Or sign up with Disqus" and click on "I'd rather post as a guest."

Recent blog postings: