GM Volt Forum banner

Eh, might want to like...pay attention even when Autopilot is active

11K views 60 replies 31 participants last post by  jupitermoon 
#1 · (Edited)
See less See more
1
#2 ·
It's beta software

Also, AP doesn't relieve the driver of the need to watch what is going on. It is a driver-assist program, not a level 5 autonomous driving program. I have seen a human do precisely that same maneuver in a construction zone. They were either confused by the poorly marked lanes, or simply weren't paying attention.

AI doesn't have to be perfect, just better than human drivers. This is a pretty low bar actually, so I welcome any safety measure that lowers the chance that some yahoo busy texting will be prevented from plowing into me.
 
#3 ·
Ouch. I'm torn on stuff like this... on one hand, autonomy/assist can theoretically make things safer. But on the other, certain levels of autonomy/assist (at least below full autonomy) seem to lead to some people being careless. And even full autonomy will have to deal with unanticipated scenarios, and people will undoubtedly get hurt because of that. Even if the overall numbers show a safety improvement, I'd be furious if my car crashed and injured me or a pedestrian.
 
#4 ·
At least the warning flashers appear to come on automatically.

On my 2017 Volt with ACC I find it unnerving to begin to feel my Volt begin to accelerate because the distance to the car immediately in front of me is increasing but I can see a car on my right that I know will try to cut in front of me.

Maybe when all of our cars are "talking" to each other issues like the one I describe can be resolved between the autonomous cars.

I hear that like me autonomous cars have difficulty at a four-way stop!
 
#6 ·
Despite Tesla's marketing efforts, neither AP1 or AP2 are anything like autonomous driving systems. It's driver assistance only. True autonomous cars, properly tested and certified, won't do things like this. There'll be enough redundancy with radar or lidar sensors to avoid a concrete barrier.
 
#7 ·
+1 This.

Yeah AP sucks. While it's marketed (hyped?) as being a lot more, it's basically a gussied up version of ACC and LKA. The problem is that you need the driver to be involved, but there isn't a good way to let the driver know when they need to be involved. Something unexpected happens and you're in the ditch, so to speak. Things happen so fast there is no good way to handle the transition from auto to human pilot.

My sense is that at the moment you're safer with ACC and LKA since with these aids you have to be engaged. I was excited about super cruise but I wonder if that may not be ready for prime time either. Given that ACC and LKA takes a lot of the driving burden off your shoulders, I'm thinking that at the moment less may be more.
 
#8 · (Edited)
"The car is AP1 (first generation Autopilot) and I've never had any problems until today. Autopilot was on and didn't give me a warning," the author wrote. "It misread the road and hit the barrier."

Taking a closer look at the footage, the construction site did not include clear road markings to guide traffic onto the median. The original line markers as still visible, potentially confusing the Autopilot system.
Part of the issue is calling it "Autopilot", allowing some to think it has more power than it does. "Lane keep assist" seems more accurate. Emphasis on assist.

But it's good to see owners willing to be pioneers for the rest of us, providing us with "don't do this" examples. I wonder what the driver was doing at the time? Reading? Texting? Playing Solitaire on the phone?
 
#11 ·
It looks like AutoPilot kept it in its lane during the hit and kept it from taking out the SUV in the right lane. It would have been more impressive for AP to avoid the obstruction all together, but I was still impressed with the "after the hit" control.
 
#12 ·
Could be many things I suppose, but the car looks to be "smoking" after bumping the wall. I would totally get that from a combustible engine, but I have to wonder what exactly is going on here? I understand it is pure speculation, but it does seem to be coming from both sides of the car as there seems to be two distinct puffs rolling out the rear.
 
#13 ·
The tire, and the heat exchanger can both create smoke on a collision as the car is moving, and briefly afterwards. Some smoke might have come from the airbag if the air is vented outside.
 
#14 ·
Very impressed with the control the car had given the driver was not paying attention. That said, part of the problem is the road was not marked for safety (even human). You can read posts and details from other sites on this accident. It was poorly done construction in Dallas. Still, driver error as car was assisting only.
 
#15 ·
Look forward at least 1/4 mile. Know what lane is the best lane. Know what the cars around you are doing. If that transition was a broken car instead, the outcome would have been much worse. The pickup in front would have avoided it, and the car would hit it at full speed since both the driver and sensors ignored the lead vehicle and right side lane marking entirely.

This is what I see as the major issue with autonomous systems. They currently do not anticipate potential threats, only active threats. If all you do when you drive is react to active threats, you will eventually crash, because the number of active threats is related to the number of potential threats. The more potential threats you address, the fewer active threats you will encounter.

In this case, the potential threats were tailgating, speed, incorrect lane choice, bright colored warning signs. Reacting to any of those risk enhancements would have probably stopped an active situation from occurring. Reacting to all of them would have reduced to the risk to almost 0.
 
#17 ·
What's cool is since Tesla controls all the service, if they need more traffic in the repair business, they simply send out an update to the AP to enhance service and parts income. :D
 
#19 ·
Here’s what I see in that video, and I’m trying to be as neutral as I can.

The vehicle strikes the “Jersey wall” at the lane shift. I do not know unequivocally if the car was self-driving, manually piloted, or some combination. I do know that those lanes were not legally/correctly marked for the circumstances and it was raining. So I’m not going to make a judgment about the –cause- of the collision. I am, however, going to make some observations about what happens AFTER the collision.

Upon strike, the vehicle is physically thrown to the right, it’s emergency flashers instantly activate and the vehicle recovers, without over-correcting, and begins a completely controlled and gradual deceleration.

This is absolutely astounding for a number of reasons.

If you pay close attention you can see the uninterrupted skid mark, off center from the rear wheel. This tells us that, at minimum, the driver’s front wheel was locked up by the collision and was just sliding. That means this car maintained control, did not over-correct, fishtail, or strike other vehicles with one of its steering/control wheels out of commission and dragging like an anchor. I can tell you for a fact that 90% of human drivers would have done one, or a combination, of the following things…
  • Spun out
  • Crossed over in front of the Jeep (who probably would have then struck the Tesla).
  • Over corrected back into the left barrier.
  • Wobbled all over the place.
  • Jammed on the brakes abruptly, and been rear-ended by the car with the dash cam.

In addition, full telemetry from the car, and “machine learning” models mean that the minute Tesla’s programmers and engineers figure out why this happened, ALL Tesla cars will know about it and such a thing will probably never happen again.

So perhaps the Tesla is responsible for the collision in the first place, I don't know, but I do know that from the moment of impact on the car handled things better than any human could have.
 
#22 ·
So perhaps the Tesla is responsible for the collision in the first place, I don't know, but I do know that from the moment of impact on the car handled things better than any human could have.
Totally agree, most drivers would've smoked that wall and spun out across the lanes to the right. The Tesla stayed glued to it's lane. I get upset when crews build a temporary wall across lane markings like that, it happens all the time were I live. Sometimes they paint temporary orange lines over top but a painted lane marking should never end in a concrete wall (auto pilot or not).

Can you image the rude awakening the drive got!
 
#20 · (Edited)
For those hoping Autopilot 2.0 will be an improvement....who knows, maybe it eventually will, but nowhere near AP 1.0 right now.

Frankly, I'm shocked Tesla even released "local Autosteer" to the public in its current state. That video makes a drunk driver look good!
In case you're wondering whether darkness was an issue, the same guy did the same run during the daylight, and it wasn't any better.

 
#57 ·
For those hoping Autopilot 2.0 will be an improvement....who knows, maybe it eventually will, but nowhere near AP 1.0 right now.

Frankly, I'm shocked Tesla even released "local Autosteer" to the public in its current state. That video makes a drunk driver look good!
In case you're wondering whether darkness was an issue, the same guy did the same run during the daylight, and it wasn't any better.
AP2 is pretty much on par with AP1 now.

I'm at over 29K in my AP1 Tesla. I would guess that 90+% of that is using AutoPilot as a driver assist feature. I even use in my local 4-6 lane 45 mph roads. My hands are always on the wheel and I'm paying attention. It does a great job of staying at +5 over the posted signs, staying 2.5 secs behind cars, stopping/going while behind cars at stoplights, and seeing TWO cars ahead with it's latest radar improvements. It happens regularly when there is a mini-van, SUV, etc in front of my that I can't see past but the Tesla display shows the car ahead of it. I've had my car brake even when the car directly in front of me did not but the car in front of it did. I like the safety, comfort, and convenience of AP as a driver assist feature.

30Mar2017 -- 8.1 | 17.11.3 http://insideevs.com/latest-tesla-autopilot-2-0-update-tested-32-minute-video/

DeRage has a reputation for providing long, detailed videos for our enjoyment. This one makes you feel like you are actually behind the wheel of a Tesla, utilizing Tesla Autopilot 2.0.

The camera is mounted in a driver point-of-view location, and provides a crystal clear view of exactly what is happening on the road, on the instrument cluster, and on the touch screen. DeRage provides a continuous commentary regarding exactly what he is doing, and how the car is responding. Fortunately, unlike some of the other videos that we have seen, the audio is crystal clear.

<snip>

DeRage’s overall impression is that the Tesla Autopilot 2.0 with the newest update is very stable. He explained that now, drivers are just waiting on some of the automated convenience feature, like automatic windshield wipers.
 
#21 ·
Calling this an AP crash is premature. How do we know the car was in AP mode?

I was following a guy a few months ago that hit the wall in a normal marked right-hand curve. Knocked the front wheel/tire completely off the car and the car traveled across traffic to the right berm. I had to avoid the tire that was still rolling up the road against the wall. I'm sure it was not a Tesla.

My car (the ELR) was in 'human' mode. ACC sometimes doesn't handle curves well since it loses the lock on the lead car when it sees the wall. I'm sure this is programming and that Tesla's is similar. Having the car brake because of a curved wall would cause more problems. (False positive reaction.)

On other occasions, a long sweeping curve with other traffic causes ACC to lock on to a car in the adjacent lane. Since I drive in the 'fast' lane, the car being detected is slower and causes ACC to brake. Easily over-ridden, but, not if you're texting or something equally stupid.
 
#28 ·
Calling this an AP crash is premature. How do we know the car was in AP mode?
The driver said it was so I'm assuming it was. He also said it was the first time he had had a problem with AP, which is the issue. There are just some things that AP can't handle, which is why you shouldn't be doing the "look ma, no hands no feet" routine. If you notice, before the crash the car drifts to the left part of the lane, likely to give the driver on the right more room. Bad decision. Could a human driver have hit the wall? Sure. It's not as if there aren't plenty of rubber marks proving they could. And I've seen drivers who can't keep their vehicles between the lines. But that begs the question of how "auto" auto pilot is.

Agree that ACC in the ELR does not handle roads with curves. I turn it off. My point is that all these technologies are basically "assist" technologies. They're great but you can get into trouble if you think they're more than what they are.
 
#25 ·
Or maybe future Tesla will come equipped with a feature that slaps you silly like Batman did to Robin every time your eyes are detected looking away from the road for more than a few seconds.

 
#27 ·
Crazy jersey barrier placement. They should change lanes, and then give a runoff area before placing the jersey barriers, in case someone misses the lane change. Some flashing yellow lights above the barrier would help.

The car corrected nicely, and didn't ping-pong to the right, as a driver could easily have wiped out that SUV to the right of it.

The smoke I would guess is from the left front wheel being jammed and dragging on the ground.

Lastly, the driver should have paid more attention given that a temp lane change usually involves a lower speed limit. Once the speed slowed, he should have paid close attention. Also, the driver must not have known about the lane change, so likely their first time on that stretch of road. I think it makes common sense to pay attention to a road the first time you drive it, at least monitor the system.
 
#29 ·
I still think we should invest in converting HOV and similar dedicated lanes (toll, etc) to support self driving cars. Get in the lane which is well marked both on ground and sides with specific information any such car can understand with live communication from antennae along the road. Charge for them like express lane tolls that many cities have installed.
 
#31 ·
I think Duch's reply describes what I saw from this crash.

It does seem and I think was confirmed that AP was on (since AP certainly seemed to be on for it to have recovered from the crash so well).
When I first saw the video, I was surprised with the sudden appearance of the wall. It certainly popped out of nowhere and while the video didn't show anything much prior to that, there didn't seem to be much signage about this. I'm pretty sure a few people had a *WTF* moment if they were following another vehicle that suddenly changed direction at that speed while ground markings kept going strait. AP was probably not expecting to have the lane's lines end in a concrete wall.

I certainly expect tesla to learn from that and hopefully the road construction staff too but tesla is the one that must expect other situations like that. AP was probably seeing this as the front car changing lane, but then wasn't expecting the sudden appearance of a wall.

I do think however that the reaction after the crash was exceptionnal for the reasons Duch described. That crash could have been much worse.
.02$
 
#34 ·
...When I first saw the video, I was surprised with the sudden appearance of the wall. It certainly popped out of nowhere and while the video didn't show anything much prior to that, there didn't seem to be much signage about this. ...
If you watch the longer video at the link, you'll see there's a small orange sign off to the right indicating a lane change coming up. Presumably, there's a bigger sign earlier indicating construction, a speed limit change, and a lane change, but it must be at least a mile before the incident.
 
#32 ·
Smoke Answer:

Maybe some tire, but definitely heat exchanger. Watch the windshield of the chase car. Dry at first, then it's 'raining'. Water mist is coming out after the accident.
 
#35 ·
Google the stat "how many die in car accidents everyday in the united states", most sites report over 100/day...
 
#38 ·
Yup, one of the largest reasons is distracted driving...2008-2012 all the time endless ads on TV, billboards, radio, etc, "don't text and drive"...In the late 2000s the public opinion of texting and driving was WORSE than a DUI...Fast forward to today and texting and driving (or doing anything on your phone) is a little worse than speeding...

Get into an accident with an AV and at least they can learn from it...
 
#37 ·
I'm definitely in the "I told you so" camp. As expected, AP had lulled the driver into a false sense of security and complacency. I maintain that the only rational way to use AP is with both hands on the wheel, foot at the ready, and both eyes on the road. In which case, why bother?

As a career EE in communications and computers, it mystifies me how unrealistic and yet pervasive the whole AP meme has become. Chalk it up to corporate marketing. Engineers are not such great believers. The infinite variability in context that will confront AP, coupled with what is at stake, i.e., the extreme costs and losses in an accident, make this problem almost unsolvable in our lifetime. Just try to program a computer to anticipate all the "unknown unknowns." And reducing the universe of unknowns would require transforming nearly our entire roadway system into something akin to a network of railroad tracks. Where's that money going to come from? -- they can't even keep up with potholes as it is.

I continue to believe that a true AP capability needs another 15 years of development. And much greater reliability in automotive electronics than has ever been achieved.
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top