flangehead

Question: Under what situations should I be on alert for failures by semi-autonomous vehicles?

The setup: I had to drive my daughter home along the same road I use to commute to work (4 lane, 2' shoulder, 45 mph posted, rush hour). As we left a red light I noticed the driver in a late model high-end SUV behind me was looking in her lap. I asked my daughter to turn around and watch. She said the driver did not look up until we reached the next traffic signal, which is about 3/4 mile away.

I had thought distracted driving was 5-10 seconds of inattention. Gulp.

I have not been in the market for, or paid any attention to, new cars since 2009. I brought up the situation with a youthful colleague at work and he said he has a Lexus that essentially has auto-pilot. He then went on to describe situations where his vehicle has misbehaved. Apparently one guide the system uses is painted lines and if they go away or curve "wrong", it has caused his vehicle to brake suddenly. I asked him if it would detect a bicycle and he said he assumed it would, but was not sure.

So I have to believe that the driver I observed was on some kind of auto-pilot and was therefore free to text or read or who knows what without watching the road.

Car and Driver calls this "semi-autonomous" technology and must be what is enabling drivers to ignore the road for minutes at a time. Now that I know this behavior is out there, I want to know more about the failure modes of this stuff. When should the hairs on the back of my neck rise up? Will they automatically give me 3 feet of clearance? Will they detect a bicycle ahead and slow down, and what situations (rain? fog? glare?) degrade that capability? I presume the answers vary by brand and model; you would hope Volvo would be pretty good (unless Uber gets ahold of it) but does the Volkswagen system only work on the test track then switch off in real life?

This is a new threat to my existence that I’m trying to get my head around. Any experiences or guidelines out there?

350htrr

Semi-Autonomous vehicles are a fail, IMO... People would rely on them way more (too much) and would expect more from them then should be expected, thus a fail...

ThermionicScott 

I agree. Vehicles either need to be fully driven by a person or a computer. Either one can do an adequate job on their own, but sharing control doesn't work.

Darth Lefty 

Now that you two have agreed can we get on with talking about tactics?

It would be useful to know what they sense and how, and if there’s any hardware that would help. I don’t know if they are a classic scanning radar that finds the range to things nearby and works out targeting solutions or if they are doing image processing. I’m imagining something like a corner reflector. The ones in my spokes only have a little layer of chrome and might not be making much of an RCS

Darth Lefty 

Ok, after some reading, Level 2 is where we are and what we must worry about. This is “hands off” but not good enough to call an autopilot. It looks like the scariest level because it relies on a driver but also encourages that driver’s inattention. Sensors are lidar range finding, and stereo digital video with machine vision.

RoadKill

Anything less than level 4/5 I feel should not have ever been sold to the public. People get too comfortable with it once they feel that it works reasonably well and will not pay attention, then we end up with Uber drivers in Tempe not paying attention and running over people who are crossing in unexpected places.

At least for the most part I feel that even a semi-autonomous vehicle isn't going to run me down as long as I am behaving in an expected manner. It's not going to dive in to the bike lane or jump up the curb or probably won't even run me down from behind if I'm taking the full lane. My worry is mostly related to roads where markings aren't so obvious as to where the car should be.

InOmaha

We drove for a couple hundred miles in light snow a couple weeks ago. When we ended the drive there was an indication in the car that the parking sensors were failing. It was a combination of salt and slush covering the front of the car.


Semi-Autonomous Vehicles may require more frequent car washes to keep the sensors clean and functional.

RubeRad

don't be like this cyclist/pedestrian, crossing the street at night, away from intersections and streetlights, right in front of a moving car.

WARNING: video might be too graphic for some

https://www.theguardian.com/technolo...man-in-arizona

I don't think a driver paying attention could have avoided that one, but maybe human visual perception IRL would have been better than the way lighting is rendered in that video.

RubeRad

Humans can do an adequate job if they are focused, but often aren't (whether semi-autonomous or texting or applying lipstick or whatever). Computers will be able to do an adequate job (significantly better than the average human driver), but I don't think they're there yet. But we're making progress and we'll be there pretty soon.

I wouldn't worry too much about it in the meantime. A few cyclists may get hit (and even a few may die), but (a) every one will be maniacally publicized, causing instant and intensive R&D to fix the error, and (b) in the long run (and even the short run) way fewer cyclists/pedestrians/drivers will die from faulty automated driving, than currently die from faulty human driving.

HardyWeinberg

+1

robertorolfo

This incident has been thoroughly debated in the Advocacy section, but suffice to say that the lighting in that video is way darker than what other videos, and people (someone from this forum drives that road on occasion) say is realistic. In this case it's a pretty decent bet that a human would have avoided this pedestrian, or at least attempted some evasive action.


Anyway, in the long run it's probably a good thing. Humans have shown themselves to be horrible drivers, but in the meantime they need to handle the transition very well, and they should never reach the point where they outlaw human operation of the vehicle.