A few years ago, many industry pundits were predicting that we could expect fully autonomous cars to be cruising the streets by 2020. As we grew closer to 2020, however, the little rascals started to pull back, pushing the date of widespread deployment further and further into the future.
In August of this year, I was presenting a couple of papers at the Embedded Systems Conference (ESC) in Silicon Valley. As part of this, I got to ride in an autonomous Ford Fusion. On the one hand, it has to be acknowledged that this was an interesting experience. On the other hand, there was someone sitting in the driver’s seat, “Just in case.” Also, the ride was painfully slow.
One of the keynote presentations at the conference was given by the guy who headed Google’s self-driving car project. This project has now been spun off as a subsidiary of Alphabet (Google’s parent company) called Waymo, which stands for a new way forward in mobility.
Sad to relate, I can’t recall the guy’s name, but he gave a really interesting talk, including photos from the first DARPA Grand Challenge in 2004. A lot of these photos showed vehicles upside down in ditches or sinking in ponds. The prospect of a fully autonomous car seemed a long way away back then. Now, only 15 years later, the first limited-scale real-world deployments have come to pass.
For example, my chum Charles Pfeil just sent me a link to an article titled Waymo’s Driverless Car: Ghost-Riding in the Back Seat. As we see in this video — which was posted only yesterday as I pen these words — Andy Hawkins from The Verge got to ride in one of Waymo’s fully driverless vehicles, which are providing passenger trips in the suburbs outside Phoenix, Arizona.
Now, Phoenix is a great area for autonomous cars, as compared to other locations that are inundated with problematic environmental conditions like rain, fog, ice, and snow. Restricting autonomous cars to a relatively small, well-defined area that typically enjoys good weather obviously makes things much simpler for the cars. “Still and all,” as they say in Ireland, this really is a giant leap forward.
Personally, I now think we can expect widespread deployment of fully autonomous cars by about 2030. Do you agree with this assessment, or do you think I’m being overly optimistic?
I agree with a widespread deployment of fully autonomous cars by 2030. There is a general fear (including from my wife) that autonomous cars would be too dangerous. As for me, the situation is already extremely dangerous (5 million accidents per year in the US with >37,000 deaths and 2 million injuries). To me, it is not a high bar to improve on those stats. I would assume “widespread” would include being able to deal with snow, ice, unexpected jaywalkers, and being distracted by texting, eating and UFOs.
I know a lot of people are reluctant to trust their lives to an autonomous car — but when I look at the way people drive here in Huntsville, Alabama, I think anything has to be an improvement. I’m guessing that only 1 in 10 have the faintest clue what their indicators are for, for example. Autonomous cars can’t come too soon for me 🙂
I think there will be a big rollout but I also see the insurance companies being a big holdup. They have to have someone to blame. So I see a big rollout by 2030 then a pullback then practical implementation on a widespread basis between 2050 and 2060
I think that the first few crashes involving autonomous cars will result in insurance claims — but as they prove to be more and more reliable — and as it becomes obvious that human error is the cause of the vast majority of crashes — the pendulum will swing and the drivers who are doing it themselves will end up carrying the brunt of the liability.
Google is developing an A.I. attorney based on IBM’s Watson to defend them in court.
It would be a scary future indeed if Judge and Jury were AIs — almost as scary as loading the courts with a preponderance of hard-left and hard-right wing views. I’ve been thinking about this a lot — based on what we’ve seen over the past few years — if I got to be made president, first of all I’d do what Obama did — surround myself with experts (not necessarily ones who agreed with me) and listed to what they said. I’d also try to work with congress to come up with meaningful restrictions on the president’s powers. I’d also try to balance the courts — especially the supreme court — so that in addition to a few hard left and a few hard right — there were a bunch of “middle of the road” folks that weren’t “banging a drum” for any particular point of view.
Something my son, Robert, and I have discussed is the following: When all street vehicles are autonomous, how will stop-and-go traffic on freeways (when there isn’t any accident or obstruction on the road) be prevented? After all, humans are blamed for this, usually for sight-seeing, or tail-gating, or a distraction. If humans are taken out of the equation, how will it be prevented? The difficulty occurs when the cars need to change lanes for an exit or entering the freeway.
Interesting question. Let’s start with a queue of cars at a red traffic light. When humans are driving, when the light turns green, the first person moves off — then a bit later the next car starts up — then a bit later the next one and so on. If the cars were all autonomous and talking to each other — they could all start up at once (traffic lights will only still be needed when there’s a mix of human and autonomous drivers — when everything is autonomous, there won’t be any need for traffic lights).
In the case of freeways — since all the cars are talking to each other — if the traffic density is increasing, cars further back can be told to slow down a bit (as much as required) to keep things flowing smoothly (if more slowly)
“Now, Phoenix is a great area for autonomous cars, as compared to other locations that are inundated with problematic environmental conditions like rain, fog, ice, and snow.”
Ay, there’s the rub…
My new car has some of the sensors that this self-driving technology might use. The lane departure sensing leaves something to be desired. Some of the roads that I use have a patch of a long strip of tar down the middle of the road. In a particular light, the sensors see this as a line of the road and try to shift the cars position. With a particular light, state of the painted lines, rain or snow and the lane departure system just shuts down.
The self park feature is also sub-optimal. It chickens out (claiming not enough room) when a human (or at least this human) can manage the manoeuvre with no distress.
That’s not including false positives when it comes to collision detection and even worse, some false negatives- it missed a pedestrian walking in front of me. I didn’t hit the guy since I was hardly moving. When I complained to tech support, they mumbled about the different conditions where it would sense, but I remain unconvinced.
Given the current approaches, I don’t think self driving cars will ever become pervasive. There needs to be a sea change in how the car senses location and object presence and in fact there would have to be an alternate system working in parallel. It is possible that they may develop in 10 years, but I am not holding my breath.
The thing is that they will use sensor fusion — mix GPS with data from accelerometers, gyroscopes, magnetometers etc. I also saw something recently where a company is talking about mapping out all the roads using ground-penetrating radar (things rarely change underground, but each part has a unique digital signature like a fingerprint) — the cars would also use ground-penetrating radar as yest another form of input.
I think the autonomous driving systems have a difficult learning curve ahead. Things like a piece of paper blowing in the wind that would immediately be dismissed by a human driver might be interpreted by AI as a heavy object and could result in panic braking. As mentioned in the video you linked in the article, a human seeing a moving van with the rear door open and someone moving around inside would know that the van is not about to start moving, but the AI might not arrive at the same conclusion.
Difficult — Yes; Impossible — No
How exactly does an autonomous car negotiate with humans as to who goes first at a stop sign or at a crosswalk? Sensor fusion is an admission that machines still are much worse than humans at understanding what they are looking at. How does an autonomous car know the rules of the road as actually implemented by the human users of the road, as distinguished from the formal written rules? Autonomous cars have no experience having conversations with humans, so are severely autistic. Let’s see an autonomous car drive across Boston during a weekday.
Just attach a big snow-plow on the front and let the car negotiate with that LOL
I think it would be interesting to discuss the things that autonomous cars will do better than humans. How about 3D 360 degree awareness. Early knowledge of emergency vehicles and being able to react in advance. China would love that because on freeways, whenever a high-level politician is on the road, all cars need to pull-over to let them through. Also, what bad human behavior will be eliminated. How about drunk or medicated driving, driving recklessly because you are late for an appointment (maybe by then we will have exceptional holograms and you could attend the meeting from your car), no more grandmas and grandpas driving who don’t know where they are (Florida), no more young people who think it is just fine to drive 90 MPH on the freeways, zigging and zagging through the other cars.
It’s also interesting to discuss how other things might be impacted if all cars / trucks / etc. are fully autonomous. Most of the things we have in place today are there to make things easier for people. For example, we wouldn’t need traffic lights anymore — the cars could be talking to each other and each one speeding-up/slowing-down so they could be zipping through interleaved in all dimensions. What about lane markings — why do we need them — in many cases two lanes could accommodate 3 cars side-by-side if they were under computer control. In my home town of Sheffield, England, there’s one area with 5 lanes — two always accommodate cars going from west to east — two for cars going east to west — and the middle one handles one direction in the morning and the other direction in the afternoon and evening. If all the cars were autonomous, the direction of the lanes would be irrelevant — on one of today’s highways with 2 lanes in either direction — they could change between 3:1, 2:2, and 1: 3 on the fly to accommodate current requirements. Of course, all this only works if NO cars are driven by people.
Railroads already have automatic lane guidance, central control, ability to change lane directions as needed, running over 100 freight cars coupled together, all on right-of-way layed-out over a century ago.
On the one hand, the railroad is awesome — on the other hand, I find it hard to believe that the USA let the system fade away the way it has — its the cheapest and most efficient way to move goods and people around — why aren’t we using it to its full advantage?
The advances possible by communication between cars is all well and good if the cars are the only things on or near the roadway.
In residential areas, you have to take children into account. If you see a ball bounce into the street, it’s wise to assume that a child might dart out after it.
In city centers, you might have areas of high pedestrian traffic where cars have to take their activities into account. Traffic lights prove very useful to allow pedestrians to cross the streets safely.
And then there are people like me who prefer to ride bicycles and motorcycles on nice days.
So there are always going to be interactions between cars and pedestrians, bicyclists, and motorcyclists. I’ve seen too many predictions of autonomous cars that fail to take other users of the road into account. On the other hand, there are certainly times when I’d like to have an autonomous car so I can sit and relax as the miles go by.
I agree — if you see a bouncing ball (or a frisbee) you can assume a kid is not far behind — autonomous cars should be able to infer this also. I also agree that while people are riding bicycles or walking around, we need to account for car-human interactions.