"New and Improved" has graced many a package in my day. Presumably it is technically correct. But it is hard to tell because the differences between the "old" version and the "new" version of the product are (with the exception of the packaging) often modest at best. I have done two posts on this subject previously. The first one can be found at http://sigma5.blogspot.com/2011/03/robot-cars.html. The second can be found at http://sigma5.blogspot.com/2015/05/robot-cars-update.html. Frankly not that much has changed in the slightly more than two years since my last post. So my usage of "New and Improved" is in line with common Madison Avenue usage.
One thing that has changed is that we now have an approved acronym: "AV". "AV" stands for Autonomous Vehicle. It is starting to pop up everywhere. And one of the places where it popped up recently is in a pair of articles published in the magazine "Science". "Science" is the top U. S. scientific journal. (The top scientific journal in the world, and the only one ranked higher than "Science", is the British journal "Nature".). In their December 15, 2017 issue they asked "When will we get there?" on the cover and printed two articles purporting to answer the question inside.
Another thing that has changed since the second article is that I have a new car that is equipped with a lot of automation. It warns me if there is a car in my blind spot. It has a backup camera. It also can look to the side when I am backing out of a parking spot and warn me if something is coming. It also has a feature that will maintain a consistent distance between my car and the car in front of me. It can often detect when I am wandering into the next lane. It also warns me if the car in front of me has slowed down or, in the case of a traffic light, if the car in front of me has started moving. It's not a full up AV but it is several important steps closer than my old car was.
So what did "Science" have to say (besides endorsing the use of "AV")? The Society of Automotive Engineers (the group behind the "SAE" numbers on oil cans) and the National Highway Traffic Safety Administration (NHTSA - a Federal Agency) got together and came up with a 6 level ratings system for AVs. A level 0 car is like what your grandparents used to drive. It has no automation. A level 5 vehicle is a completely self driving car, what everybody assumes we are pointing at. (The other levels fall somewhere in between.) In the first article ("Not so Fast") the author characterizes the likely timing of when level 5 vehicles will be available as "Somewhere over the rainbow", in other words, way out in the far distant future. Elsewhere he characterizes it as being "still decades away".
The author is generally pessimistic about how fast things will progress. This is a nice counterbalance to the cockeyed optimism of most reporting on the subject. But I think the author is excessively pessimistic. So how does he justify his position?
Well, first he reports there is "surprisingly little research" to back up most predictions of how fast things will progress. He is completely correct here. Most of what we know is from press releases and not from, say, public demonstrations of capability. Sure, there are any number of stories of some reporter being hauled around in a "self driving car". But most of what is demonstrated is the capability to handle carefully set up test tracks, not realistic simulations of real world situations. The last public "bake off" between groups of AV builders happened nearly a decade ago and involved a tightly controlled environment. Various vendors have racked up an impressive number of "open road" miles since but very little detail is available about how uncontrolled the circumstances were.
We know, for instance, that there have been more than a dozen accidents involving Google test cars. These all happened in urban areas with good sight lines and slow speeds. And in almost all of these cases some inattentive driver rear-ended a Google car that had been stopped at a stop sign or traffic light for several seconds. The author of the Science piece correctly noted that these are relatively easy situations for an AV to cope with.
Then there is the now infamous incident where a Tesla driver was killed while the car was driving itself. But this driver was well known for ignoring the restrictions Tesla placed on when its automation should be used. And the situation was quite unusual. A truck with a trailer connected by a long tow bar was turning. The Tesla went between the truck and the trailer and hit the bar, which it had not detected. So the car made a mistake that drivers can and often do make and this driver forced the car's automation outside the parameters he was specifically directed to stay within.
This incident is included in story after story about AVs (including this one) because it is the one and only example so far of a fatality associated with autonomous vehicle operation. Most fatalities caused by drivers don't make the news precisely because there are so many of them, about 30,000 per year.
What else? He defines an AV as "a car that takes you where you want to go, at any time and under any drivable condition, without any human intervention". At first blush that seems to be a perfectly sensible definition. But then he takes "under any drivable conditions" and "where you want to go" to extremes. He cites "to New York City or the Gobi Desert" as examples of where you might want to go. I do not expect my current car to be able to go all the way to the Gobi Desert and it doesn't matter whether it's me or automation that is in control. I freely concede that an AV that can take someone to a specific point in the Gobi desert and do it without any human intervention is a long ways away.
But I live in Seattle, and I have a brother who lives in New York City. I think it's reasonable to expect an AV to be able to do most of the work necessary to go from here to there. It should be able to pick me up at my doorstep and eventually deposit me at my brother's doorstep. But I would expect to help with the selection of gas stations and motels.
My car can't drive itself. But I can plug the address of a gas station, a motel, even that of my brother, into the car's navigation system and it can provide me with detailed instructions for getting there. If I plug successive waypoints in as I go along the route it can get me there in stages. And my car can handle the entire route from my front door to my brother's front door in one shot if I want to drive day and night straight through. Let's say I could take the current capabilities of my car and add an additional capability that would permit me to press an "AV" button. When pressed it would make the car then do the driving instead of me. If such a feature could be added then I would say that that the resulting vehicle would count as an AV.
So I think the author goes too far. But he does point out a valid concern. He points out that "[c]ompanies have gotten very good at crafting statements that will be presented in the most positive light". Too true. And the author points out that an AV needs to handle common suboptimal situations like rainy weather or crowded roads. From there he quotes an expert who says "he wouldn't be surprised if it's 2075 before we get there". I think this is a valid concern but I think both the author and the expert are excessively pessimistic.
First of all, in many cases crowded roads are easier to deal with than less congested ones. Things play out slowly and there are only a few ways for things to go wrong. And if things go wrong the usual solution is to stop the vehicle, something that can be quickly because the vehicle is going so slowly. The same thing is true of rain. There are two issues. Firstly, how slippery is the road. Computers have access to sensor data that people can't even comprehend. So they can continuously monitor traction conditions and act accordingly. Most people are not very good at this.
Secondly, what's the visibility situation? I live in a part of the country where it rains a lot. For the most part visibility remains good. It does tend to get dark out but cameras and other equipment used in AVs are actually better at coping with this than people are. So again in many common situations AVs will cope with the situation better than people do. But that's not always the case.
If you are on a congested road where everybody is in "stop and go" mode then an AV can easily do a fine job of following the car in front of it. In an actual test in my car I found my car did a much better job than I could in these circumstances. But what about really congested situations when you need to move over a lane to get to an exit? What if there is no gap? We've all become more or less adept at bulling our way into the next lane and have accepted that this sometimes may result in honking horns or rude gestures. But what would an AV do in the same situation?
Most of the time when I am driving in rain visibility is diminished but remains adequate. But sometimes the sky opens up and visibility drops to effectively nothing. You can run into a similar situation if there is a bus or big truck in the next lane throwing spray everywhere. Can an AV handle these situations?
How about snow? Snow can make roads slippery. But it can also blot out lane markers. Or it can come down so fast that visibility drops to almost nothing. Or you can get snow over ice, an unfortunately all to common situation where I live. This can give the appearance of tricky but manageable conditions while masking extremely dangerous conditions.
I had not given sufficient consideration to the fact that an AV must be able to handle not only easy situations (dry road, good visibility) but poor situations (rain, congestion). And there is a continuum here. Moderately poor situations are common enough that the AV must be able to handle them. And as the degree of difficulty rises so does the rarity. At some point a tradeoff will be made. An AV will be designated "good enough" if it can handle poor situations up to a certain degree of difficulty. Where will the line be drawn? That is yet to be determined. But AVs will not be required to handle "anything, anywhere, any time".
The first AVs to be put into circulation will only be able to handle fairly easy situations. They are likely to be restricted to urban areas. I think it is likely that they will be handle both day and night time. But they will probably not be able to handle moderately bad weather. They will not be able to handle truly awful situations. But my city tends to shut down when assaulted with only modest amounts of snow. So the fact that some or all AVs might be pulled out of service when it gets truly awful is less of a change than most people think it is.
Another location that is likely to see AVs early is the Interstate. There is a lot of interest in autonomous long haul trucks. There is a lot of money available in this market. It is common for a big rig to put 100,000 or 200,000 miles on in a year. And a lot of that mileage is racked up on the Interstate at night. And there are complex rules governing how long a driver can drive between bouts of down time. If a driver can put his truck on autopilot for several hours straight and it doesn't count as driving time then the truck can be kept on the road doing productive work for more hours per year. That's worth a lot of money to trucking companies. And that's why there are several different groups working on AV trucks.
Right now there is a lot of talk about "how safe is safe enough?" And the author goes into this. The conventional wisdom is that AVs have to be a lot safer than regular cars. Certainly the amount of coverage devoted to AV accidents and not devoted to regular accidents would indicate that this is true. But there is a big fear/novelty thing going on. The very fact that there hasn't been a second "AV" fatality guarantees that whenever it happens it will get a lot of coverage. So far the nonfatal accidents that AVs have racked up have generated little coverage. That's because they tend to be very low speed fender benders. The lack of blood means these incidents also lack newsworthiness.
And this brings up a point that only occurred to me after reading these stories AVs currently operate in an extremely conservative manner. What do I mean? Have you ever drifted through a stop sign? Have you ever "accidently" exceeded the speed limit. Have you ever cut one or another additional corner (i.e. barging when you needed - or just wanted - to change lanes)? We all do. It's a routine part of how most of us drive. We conclude we can safely cheat a bit so we do. But AVs all ALWAYS scrupulously follow all traffic regulations. And their first response to anything hinky is to slow down. People are already starting to notice that AVs are extremely annoying to share the road with because they operate in such an extremely conservative manner.
They currently only annoy a few people because AVs are only on public roads in a few places. But if significant numbers of AVs hit the road in a bunch of different places this could quickly become a serious problem. I live in one of the most congested cities in the country. Something that, at least in the short run, is likely to increase not decrease, congestion will quickly become extremely unpopular. From a technical point of view this problem is easily solved. AV programming can be changed to behave differently (i.e. drive more aggressively with a less scrupulous adherence to traffic regulations). But the political, social, and cultural forces that would inevitably be brought to bear to reel things back in would make this impossible.
Now let me move on for a moment to the second article entitled "A mater of trust". It deals with people riding in AVs. Surveys show that a large percentage of the population is afraid or at least concerned. We have all seen "technology goes berserk" movies and TV shows without number. We have been conditioned to be afraid of killer robots and technology spiraling out of control. And an AV can easily kill someone. So there is something to the concern. But the results of the research reported in the article can be succinctly summarized. In a number of different situations people went from being concerned to being bored in about 15 minutes. A few minutes of actual experience with a simulated AV is all it takes to totally eliminate the concern.
There are some easily implemented recommendations. A display inside the car that lets the passengers know what is happening or is about to happen (i.e. "you are about to reach your destination") seems helpful. Similarly, indicators like flashing lights or the like on the outside of the vehicle seem helpful. They would signal say a pedestrians that wished to cross in front of an AV that the AV sees them and will wait for them. But only a modest amount of this sort of thing is all it takes to make people comfortable with AVs.
I think these results from the second article feed into the concern raised by the first article about safety. People currently have no personal experience with EVs. So they tend to err on the side of caution. But I think experience will quickly ameliorate this need for caution. We already see some effects. Even a non-AV car like the one I own logs a ton of data. This makes the determination of whose fault an accident was much easier and more determinative. We saw this with the fatal crash. Tons of data about what the driver was and was not doing was available. We have also seen it with the Google accidents. Tons of date made it crystal clear that the Google car has behaved appropriately and the "loose screw behind the wheel" in the other car had not.
I expect car makers to initially self-insure because everybody expects that the "victims" in an AV - non-AV crash will ask for millions. But these victims are likely to be confronted by tons of evidence from the AV that the AV was not at fault. Once things settle down, if a car company has to shell out a big settlement every once in a while but most of the paying ends up being done by the non-AV side of the conflict things will change quickly. And I think it will take more than fifteen minutes for this to all shake out. But I don't think it will take long.
This brings me to my final subject. Will AVs change behavior? Specifically, with people with access to AVs pile on the mileage? Preliminary indications at that the answer is yes. In one experiment people were given access to a simulated AV. In this case it was just a car with chauffeur. People in the test racked up considerably increased mileage. It was just so damned handy.
But this test only lasted a week. So there was definitely a novelty factor going on. Who wouldn't want to show off their chauffeured car? It is unclear what the long term effect would have been. But I have to confess that this and several other studies indicated that access to AVs increases average miles driven per week. I can wave my hands and propound arguments that the long term result would be different. But at this time I have zero evidence to put behind such an argument. The best I can come up with is "its too soon to tell" and that's weak.
Everybody has been saying "EVs in 2020" for years now. 2020 is getting closer and closer. Also, the number of players keeps increasing. And today's newspaper includes a one paragraph story reporting that General Motors has just filed paperwork with the U.S. Department of Transportation requesting permission to put cars "without steering wheels or pedals" on the road next year. Certainly the official story from a number of these players is that they still think they can hit the 2020 target.
So I will "revise and extend" my old prediction. I predict that we will see AVs on the road by 2020, at the latest by 2021. They might only be available in "selected areas". But these will be fully autonomous cars (or trucks) on regular public streets. And I predict that we will see AVs widely available by 2030. If I keep my current car for 15 years (I kept my last one for 16) then I will be looking for a new car in 2030. I have told numerous people that I believe my current car will be the last conventional car I own. By then I expect to either shift to an AV or go carless. I would use an Uber-like service instead. Welcome to the future.
No comments:
Post a Comment