SCRcat6

If you are hoping that things will get better for road cyclists when cars become autonomous, there is some bad news. Nissan driverless car guilty of "close pass" overtake of UK cyclist | Bicycle Business | BikeBiz

In the UK, a prototype autonomous Nissan vehicle passed a cyclist way too close. Carlos Goshen, CEO of Nissan, has complained that cyclists are particularly difficult for autonomous systems to deal with. Apparently, the systems have difficulty determining if cyclists are pedestrians or vehicles.

This got me to thinking. Someone needs to invent a device that fools autonomous cars into thinking cyclists are the size of a car or truck, so we are passed with proper clearance. Call it Reverse-Stealth Technology (I won't trademark it).

Somebody get on this.

99Klein 

You expected more?

Dan333SP

Pretty sure the engineers at all the major automakers/tech companies that are working on driverless systems are already all over this. There's no chance an autonomous car that occasionally makes mistakes with cyclists would be certified for use on roadways in a place like Denmark or Holland, so they'll have to solve this before they can realistically approach the EU for approval of a given design. This is another reason why widespread adoption of driverless cars is still pretty far from reality.

dabac

It'll be difficult.
The tech used in self-drive cars is basically imaging technology.
Means merely bringing back a stronger signal - which would have been easy - isn't good enough.

I see two options:
1) wait for tech to improve further
2) agree on transponders for bikes, identifying them as bikes

1) will happen automatically.
2) while doable, will open its own can of worms.

Dan333SP

Yea, transponders would be the best solution but the liability issue would be huge for the automakers unless laws are passed requiring cyclists to add transponders to their bikes, which would be shot down in a heartbeat if proposed here.

wphamilton

That rings a bell, someone is already on this idea. I can't quite dredge it up from my traitorous memory, so maybe someone will recall before I do.

In the meantime it would be easier and more reliable to have active transponders on bikes that would alert the autonomous cars to their position and heading. I can see lawmakers and car makers latching onto that idea: if you want to ride on their streets and enhance your safety, buy the locator.

WhyFi

Don't they already need to factor in peds? Is the fact that cyclists my be traveling parallel to them that gives them fits?

Dan333SP

Sure, but if a driverless car hits and kills a cyclist who does not have a transponder because they either couldn't afford it or a battery died or whatever, there would be a whole slew of lawsuits. Just passing a law requiring someone to have a transponder wouldn't protect the automaker from any liability, to say nothing about how hard it would be to get such a law passed in the first place.

Who would pay for transponders? Cyclists? They'd protest being required to purchase something just to accommodate a new car technology that they're deriving no benefit from. Car makers? They wouldn't like that, given how many bikes we're talking about. Imagine trying to get into the Chinese market, but having to pay for bike transponders for every single bicycle in that country. It'd be a complete bar to entry.

IMO it'd be cheaper and smarter to let the technology develop to a point where they can safely pass cyclists 100% of the time barring some completely unforeseeable event.

gsa103

Goshen is largely an idiot. Cyclists aren't difficult for autonomous systems to identify. Any half decent camera system should have zero trouble picking them up. Google has been driving cars around for years with zero incidents. Of course, Nissan seems to be well behind compared to the rest of the autonomous industry, so them having issues isn't a problem.

Now, there is a problem with cyclists behaving erratically. Things like running stop signs, salmoning, half-taking lanes can definitely cause fits for physical and autonomous drivers.

wphamilton

You'd think so, but car drivers are what amounts to a protected class aren't they? A while back, there was a case here where the driver turned into a residential street, at night, hit and killed a pedestrian. From a basic understanding of liability, not to mention human decency, the driver was by definition driving too fast for the conditions. Don't take the corner blind at such speed that you cannot stop or avoid someone in your way. But because it was dark, no street lights, she was wearing dark clothing and in the street, there was no fault. SHE should have known better, not the driver, or at least that was the conclusion of local law.

I think it's typical, and if there is something we can do for our safety, but don't, it will be our own lookout.

prathmann

Yes. Note that in the video the computer screen inside the car clearly shows that the system 'sees' the cyclist and his speed/position in plenty of time. So there was no need for any transponder on the cyclist.

What is needed is just a change in the computer code specifying a different minimum passing distance. Likely a pretty easy fix which may well have been implemented already after the publicity associated with this video. That illustrates a major difference between conventional and autonomous vehicles. When a mistake is made by the latter the incident can be analyzed and software and other system changes made so that type of mistake won't happen again. OTOH, human drivers will continue to make the same type of driving mistakes.

FlashBazbo

If Tesla's system still can't pick up a semi- tractor trailer rig, I think we're a long way from a system that can safely and reliably pick up cyclists.

But on the reverse stealth front . . . why not assign the job to Arofly? Surely they could engineer something that would mount on a valve stem and broadcast a suitable signal! It could be powered by centrifugal force. And they could sell it for $129!

Dave Cutter

Get autonomous? Cars already have available options that are semi-autonomous. And research is creating autonomous motorcycles. I doubt the citizenry will allow NON-autonomous vehicles on public roadways... once accidents and death rates fall dramatically because of autonomous cars.

When almost every road death is a pedestrian or a cyclists..... I expect cyclists to be banned to the bike paths.

Silvercivic27

Luckily for me, I am about the size of a car...

Shimagnolo

Google cars had no problem detecting a cyclist back in 2015: A Cyclist's Track Stand Befuddled One of Google's Self-Driving Cars

indyfabz

I feel like I stumbled into the A&S forum.

Dan333SP

They're currently under bombardment from Chinese forum-bots in addition to self-driving cars, so we're welcoming them here temporarily.

dabac

It feels like I'm missing something from the original issue.
Both cyclists and pedestrians are "unprotected" - meaning no vehicle around them - road users.
Shouldn't they be assigned equal margin at passing?

Having a car pass at elbow's reach at speed is equally scary whether I'm riding, inline skating, running or walking.

So what's the problem if a cyclist is mistaken for a pedestrian?

CliffordK

It would be trivial to add an active radio signal (no ID?) to daylight bike taillights.

With a good campaign, one might be able to get perhaps 50% of the riders to use them. But, that would be better than nothing.

They could also be given to cars, tractors, etc. Especially useful for blind corners.

From the article above:
Sounds like a programming fault, not a recognition fault. It sees the bicycle and does nothing.
A little ambiguous. Cars on a parallel course can come fairly close (while in different lanes). Perhaps bikes need MORE space as one's course is less predictable and the consequences of a "bump" are far more severe.

For a car, perhaps the rule should be to give as much space as safe and practical up to a full lane. I don't mind tight passes with oncoming vehicles blocking moving over as long as the car driver takes their foot off of the gas pedal.

nycphotography

In V1.

V2 (or eventually) cars/trucks can be mandated (DOT/NTSB) to respect a "beacon" that could be as simple as a light flashing with a specific pattern. Like a blinkie. So blinkies all have run the same pattern and you have to use it day and night, but then cars don't buzz you.

Seems like a pretty big win to me.

This is not a particularly difficult one to figure out.

SCRcat6

That would be great...as long as there is a bike path on every road! I guess pedestrians would be on their own.

rm -rf

Sailboats use radar retroreflectors to make themselves more visible to overtaking boat's radar. But these probably need to be at least a few inches wide, and maybe more, perhaps too large for a bike.

But most self driving cars use lidar (infrared, I assume) which transmits and receives a beam of light. So a normal bike reflector would make the bike very visible to the car computer. (bike reflectors are also corner cube retroreflectors, just an array of tiny ones.)

So this is passive, no batteries, and inexpensive.

wphamilton

Ah that was it! That I couldn't recall - there is a product out now (or at least was available a while back) that is basically a cube reflector, I think it's lidar but it could be radar or both, that makes the bike (or other object) appear considerably larger to those detectors.

nycphotography

I'm thinking maybe I should stick a couple on the back of my minivan (far left and far right).

Heck, maybe a couple on each side too.

700

The "problem" with a car coming upon a bicycle is that drivers often have to break the law crossing a solid yellow line to pass such a slow moving vehicle - especially with the relatively new 3 feet laws that are getting enacted. (Pedestrians are usually not in the traffic lane.)
Do you program the car to wait until it's legal to pass, and hang back for maybe minutes while traffic piles up behind you? Or do you program to cross into the wrong lane? What choice would you authorize?