@sheetal, I don't think you are alone asking that question. And the automobile industry needs to come up with better answer than...say, a generic answer like, "safety." Because clearly, there are a lot of saftey issues carmakers themselves are wading through right now for autonomous cars.
Just make sure EETimes lets us know when someone allows these self-driving cars on the roads. Be sure the cars are equipped with flashing warning lights and sirens (in case the long trail of lawyer-driven cars following them doesn't clue us in). When Calif allows Google cars on the public roads, I'll leave CA. Generally, I'd be happy for anyone who has had a drink in the previous 12 hours, teenagers, anyone with a cell phone/tablet, anyone with a headset, and anyone listening to radio, TV/DVD, children or a yakking spouse to be prohibited - electronically if necessary - from driving a car. But I'd still rather let those folks drive than to have our perfectly programmed electronics driving the cars, which would be worse if there still has to be an alert human when the computer says "I give up, you take over!" (not unlike the Tesla warning "the battery may be about to start a fire, please pull over when it is safe and exit the vehicle - customer service has been notified and your salesman will be along soon to pick you up"
[One person] while any locomotive [i.e., powered vehicle] is in motion, shall precede such locomotive on foot by not less than sixty yards, and shall carry a red flag constantly displayed, and shall warn the riders and drivers of horses of the approach of such locomotives, and shall signal the driver thereof when it shall be necessary to stop, and shall assist horses, and carriages drawn by horses...
Or this gem from Pennsylvania (1896):
Legislators unanimously passed a bill through both houses of the state legislature, which would require all motorists piloting their "horseless carriages", upon chance encounters with cattle or livestock to (1) immediately stop the vehicle, (2) "immediately and as rapidly as possible... disassemble the automobile," and (3) "conceal the various components out of sight, behind nearby bushes" until equestrian or livestock is sufficiently pacified.
I too find Google Cars to be quite alarming and I like the idea disassembling them as quickly as possible :-)
"Full autonomy still requires constant supervision. There is no system that can yet match a human driver's ability to respond to the unexpected."
By redefining what "autonomous car" means, I have no doubt that a target date of 2020 can be reached. However, a constantly vigilant human behind the wheel is not the way I would define "autonomous."
Many airports now have truly autonomous shuttle trains. Absolutely no operator in the train - just passengers. To me, that's what an autonmous car needs to achieve, to merit the title. Otherwise, it's driver assistance. So that's perhaps why I don't buy this notion that autonomous cars already exist, driving on public roads alongside normal cars. Not by my definition, not on public roads anyway.
But, redefine what words mean, and one can make anything happen.
As an aside, I think that truly autonomous vehicles can be developed, but with a lot of cooperation from many different disciplines and organizations, AND that such vehicles can be a lot safer than cars driven by your average human.
An 8-bit MCU is smart enough to keep trains running on a fully-enclosed circular track from running into each other when they only come once every five minutes. No steering required; not even a commitment to keeping the trains running as close to each other as possible. Pure, dumb timing. No steering, no decisions as to whether that's a painted white stripe that I'm supposed to stay just this side of or whether it's light glaring off of a ridge in the after-dark rain-wet pavement. No judgment whether that's a foot-deep pot-hole or a patched pot hole or a board in the road. No decision needed whether it's less damaging to hit the pile of clothes or the concrete blocks in the road. In fact, I'll bet if you jump on the tracks at the airport, the train will run right over you. I seem to remember the DFW airport in about 1980 had little trolley cars on rails that shuttled people from terminal to terminal. They were jerky (hadn't worked out the smooth ramp up and down of the motors yet) but they dutifully did their job without apparent human intervention. (might have been tires in troughs rather than iron wheels on rails).
That's a starting point, then. You need to make the paths these cars take, on the roadways, more predictable, more similar to a train on tracks. That doesn't sound impossible, does it? In this era of virtualization? It's one of the jobs the vehicle to inftrastructure comms must fill.
As to potholes, they can be detected by a combination of onboard sensors and/or roadside sensors and beacons, at least as consistently as they can by your average distracted human driver, busy chatting on the phone or worse. Plus, the autonomous car driving alongside other autonomous cars, i.e. with panic-prone humans out of the loop, will be much more capable of swerving to avoid such obstacles, without running into another car or off the road in the process. Thanks in part to vehicle-to-vehicle comms that aren't hampered by poor visibility left and right while they are busy looking ahead, or by screaming kids in the back seat, or by an irrepressible urge to text while driving, or just by plain old slow reflexes.
Honestly, guys, you'd think driving a car was the most difficult task the human race has ever had to master. Yet, just about anyone can do it. Doesn't anyone wonder why? If computers can beat the most expert chess players, and run the most precise machines, what makes anyone think they can't do something as repetitive as driving a car? Especially if the other cars on the road are also autonomous, and not haphazardly driven by humans.
Bert said: Honestly, guys, you'd think driving a car was the most difficult task the human race has ever had to master. Yet, just about anyone can do it.
When I stop hearing about collisions on each morning's traffic report, I'll agree with you. When I see an "out of business" sign at Cook's Collision, I'll agree with you. When my car insurance premiums become pocket change, I'll agree with you.
As Samuel Johnson said about a different topic: "[It] is like a dog's walking on his hind legs. It is not done well; but you are surprised to find it done at all."
"When I stop hearing about collisions on each morning's traffic report, I'll agree with you. When I see an "out of business" sign at Cook's Collision, I'll agree with you. When my car insurance premiums become pocket change, I'll agree with you."
I guess you're saying, even though just about anyone can drive, they do a poor job of it.
I can agree with that. I've no doubt computers could do a better job.
Accidents are usually caused by a driver failing to anticipate something on the road. Theoretically a software-hardware combination should be able to read the road and anticipate just about anything. But if Toyota cannot even get a fly-by-wire accellerator right, I'll trust human drivers rather than software any time.
"Accidents are usually caused by a driver failing to anticipate something on the road. Theoretically a software-hardware combination should be able to read the road and anticipate just about anything."
"But if Toyota cannot even get a fly-by-wire accellerator right, I'll trust human drivers rather than software any time."
Human designers, however, were capable of finding the flaw, and the fix seems really straightforward. Design flaws can be fixed. Haphazard human behavior, on the other hand, will always be with us. That's the difference.
I actually agree with you almost 100%, Bert. I just don't see self-driving cars being ready to use anytime within the next 5 years, and when they are I think they will have to have dedicated roads, because by themselves they will be safe enough, but add in the said haphazard human behaviour and no software will be able to anticipate everything that might happen. There are certainly interesting times ahead.
I wonder what happens if you tailgate a selfie-car in a human-driven car? Defensive driving classes tell you to handle tailgaters by slowing down so that (1) you have more stopping distance ahead of you so you can slow down more safely, and (2) to encourage the tailgater to pass you, so it makes sense for the selfie-car to be programmed accordingly. But what happens if the tail-gater has anger-management issues or is DUI and decides run you and your selfie-car off the road or start shooting? (I'm writing from the USA, where driving practices vary from sensible to Wild West.) I guess the advantage of riding your selfie-car is that if you're asleep or engrossed in Candy Crush you won't know what hit you.
@Betajet a very good scenario to justify autonomous cars having the road to themselves, without human drivers. And a very good illustration of my long-held view that our technological advancement has far outstripped our social advancement!
Frank Tu wrote: I seem to remember the Dallas-Fort Worth airport in about 1980 had little trolley cars on rails that shuttled people from terminal to terminal. They were jerky (hadn't worked out the smooth ramp up and down of the motors yet) but they dutifully did their job without apparent human intervention.
I remember those. They actually worked quite well. I had an hour or so to burn between flights so I rode around for a while to add another rail experience to my collection.
One thing I noticed is that the doors closed automatically like an elevator without caring if there were people waiting to get on. There was a "door open" button, but it was really hard to find so nobody could press it in time. I thought at the time that the DFW shuttle was basically a horizontal elevator, and it would actually improve things to have an elevator-like control panel at each door so that riders would have a familiar user interface. Then they would be able to find the "door open" button, and the shuttle wouldn't have to stop if there was nobody to pick up or drop off.
Assumptions like the car will be surrounded by other self-driven cars or on a road with linked sensors is ludicrous - only realistic in airport trains. Look around yourself when you drive and think "how many of my 50 closest (distance-wise) driving companions here are also driving this high-end BMW?" Those are the guys your wizzy car will be communicating with - if they continue to pay for the service, and that road is equipped, and you're not driving to the mountains or the beach (where the roads are remote and curvy and poorly tended). How many are 18-wheelers? Or trucks dragging trailers with lawn mowers? How many are old cars or just "modest" cars? I can't even buy a battery for my car today without checking carefully as to the model and make because there's no standardization for size and capacity. You think these magical common standards in autonomous cars will be universal? Today's cars don't even have wiper switches, headlight switches, or radio buttons in the same place. Every time I get into a rent car, I cuss trying to find these things. You're dreaming in some Eutopian world there, man. What is physically possible and what is practical in the real world - those are very different.
And referring to autopilots on airplanes and a handful of Google or DARPA cars are also foolish. Let's see, airplanes are surrounded by miles of closely managed open space (almost literally) for 99.99% of their lives. Landing and take-off are at a handful (OK, a thousand) well-known/-mapped/-controlled airports. Each aircraft essentially has the sky to itself. And yes, an airplane is actually easy to fly on the first level. No autopilot took over the plane when it sucked up birds and Scully landed gently on the Hudson/Potomac river. Automobiles are easy to drive, but they are subject to complete randomness in terms of route, road surfaces, surrounding vehicles, and driver attitudes. If you want to shake my seat when someone in my blind spot is getting too close, fine, that's worth about $5 to me. But loading my car up with $5000 of electronics so it can drive itself on 5 roads in the area is a no-value show-stopper. Give me the left lane of the highway with a gutter that will catch my wheel, and some 20' fiberglass poles with magnets at the end to connect/separate me from the guy in front/behind and we'll all tool along close-cropped at 100 mph in each other's slipstream, and maybe I'll buy into it. But have your insurance man and a lawyer ready too because I don't know who's at the front of the line.
Anyone who can see over the windshield and reach the pedals can drive a car. Teenagers, drunks, old ladies, orange cones, and the idiotic "guard rail damage ahead" signs have proven to us that driving a car safely, at speed on busy public roads requires intelligence, attention, and skill. Often you are essentially tiptoeing along the edge of a cliff: one small mistake and you've got a long painful fall ahead of you. These guys have to get out of the theoretical world and look at the practical world.
How much easier would it be to use a couple of computers to match up people going from Point A to Point B at Time C and get them into one car, eliminating 60% of traffic? But that wouldn't sell more cars, would it?
"Assumptions like the car will be surrounded by other self-driven cars or on a road with linked sensors is ludicrous - only realistic in airport trains."
What is truly ludicrous is to think that true autonomous vehicles, driving in true autonomous mode, would ever be able to drive along with those erratic human-driven cars. Think about it. The vast majority of what is unpredictable when driving a car is the numbskull driving next to you. Set aside stretches of road that are only open to autonous vehicles, e.g. such as the HOV lanes sometimes segregated from the normal lanes, and you can take that unpredictable human element out of the equation.
That's all it takes.
If those "twisty mountain roads" you mention are also properly provided with the needed autonomous vehicle infrastructure, I have no doubt that autonomous vehicles would be safer on those roads too, than cars driven by humans.
If human drivers remain in the mix, I'm afraid the best you can do with any hope of some level of safety is to assist the human driver. Someone alert will have to be behind the wheel at all times. Not my definition of autonomous.
@Frank, Bert...all good points, but I tend to agree with Bert that autonomous vehicles will be a lot easier to implement and safer if you remove human drivers from the road. Possibly a mixed-mode vehicle that uses autonomous mode on freeways to get people the big distances fast and safely (and no human drivers permitted) and then restores control to the human driver on local roads. And Frank, your idea about car-sharing - how often has this been done successfully? It needs a mix of carrot (faster lanes for cars with 3 or more passengers, and a good system for matching up passengers with the same origin and destination, and financial incentives) and stick (higher taxes and tolls for single driver cars). No-one yet seems to have got all the ingredients right.....
I have to agree with Frank Tu on this one. Very few consumers will be willing to pay for expensive electronics that permit their autonomous car to drive only on a small number of roads. And what roads are we talking about anyway -- new ones, or existing ones from which human-controlled cars will be banned? Either way, that's a tough sell politically & economically.
If autonomous cars are ever to reach the marketplace, it seems clear to me that they must share the same roads with error-prone human drivers -- and be designed accordingly to deal with those unpredictable humans.
Maybe the first application would be to monitor a new driver and provide feedack to them on how well they are doing. Create good driving habits at the very beginning. Monitor older drivers too? Maybe you get a discount on your car insurance if you let the monitor your driving habits. Lots of possibilities...
Are these cars going to co-exist with the non-intelligent cars (cars of today)? I assume that might not be easy isn't it? Because not all of the present cars would be capable of communicating to the automatic cars. What is the plan for that?
@sanjib. Good question. However, as any ADAS technology promoters would tell us, this is about a new ADAS-equipped self-driving car being able to detect objects and pedestrians around the car. In other words, your non-self-driving cars will be, in theory, "watched" better by those self-driving cars.
@Davide. I think you are being overly optimistic to consider the current research platforms such as Google Car to be approaching anything like AI. An AI would learn, adapt and change its responses based on experience. The current flock of attempts are nowhere near this level of power.
In a fully automatic (Level 4) vehicle by current definition, the driver is not needed, so all the sensors to establish the state of the human are unnecessary. Consider that one of the projected hot applications of the automated car is taxi cabs; why would the human in a cab be expected to take control if the automation reached its limit?? In a Level 3 vehicle you would potentially need to pass control back to the human, and here it really just becomes ADAS, with no potential for AI involved at all.
"Full autonomy still requires constant supervision." ...if this is true we are dead in the water for Level 4.
"There is no system that can yet match a human driver's ability to respond to the unexpected." ...so not true within the limits of the current automation. Watch Audi do Pikes Peak at the limit of the tires!
" The last thing we want to do is leave a confused car in control. " ...we are talking computer control here aren't we. When was the last time you had a confused PC? An AI or any advanced logic system either has a solution or not; it's an automaton. Next you'll have an angry car?
I hold grave doubts for the island of automation (individual automated cars) solutions to achieve our societal goals and would dearly like the logic to be in the infrastructure so we only need to deliver ADAS/Level3 automation in the vehicle. One set of logic governing all the (autonomous) cars (one ring to bind them all if you like a magic metaphor).
We can save immense amounts of investment by providing coverage in the infrastructure for our highest occupancy freeways and roads. Even the Google car today is not able to cope with some of the merging and ramp complexities, and infrastructure sensors can be designed to have an excellent view of these situations. It makes no sense IMO to design automation with the severely limited horizon of the individual cars sensors when you can get a view of hundreds of cars from the freeway infrastructure.
Let's have a sensible discussion about computing power in automated vehicles and not make it science fantasy.
"When was the last time you had a confused PC?" I'd have to say, this morning.
Pretty much the rest of what you're saying, JC, I see as being right on target.
Right now, we spend huge amounts of money engineering the road systems based on the limitations of humans and current cars. It's perfectly logical to start engineering road systems for autonomous vehicles.
Unfortunately (IMO) the SDC agenda is being driven by a lot of vested interest groups and individuals who stand to garner money, market position, influence etc by creating the islands of automation solutions beyond Level 3/ADAS.
While many support V2V communications in their autonomous plans, the fact is that the communications latency and size of the vehicle mesh quickly becomes unmangeable. If using V2V only then the control horizon is limited; and V2I followed by I2V has some problems to solve position and speed problems quickly, which is why IMO the lgoic should center in the infrastructure. Infrastructure based control would allow broadcasts messages from a central controller (at least central within a much larger horizon of vehicles). Some might raise the possibility of loss of infrastructure control (some massive failure), but this now becomes a simple "orderly stop" response for all vehicles at an ADAS level.
In large part the current situation can be considered a failing of the govenrment bodies failing to take responsibility and provide the correct longterm solution strategies and direction. The lack of drive from the authorities will I fear make implementation of the correct strategies take decades longer than it might otherwise take.
NASA's Orion Flight Software Production Systems Manager Darrel G. Raines joins Planet Analog Editor Steve Taranovich and Embedded.com Editor Max Maxfield to talk about embedded flight software used in Orion Spacecraft, part of NASA's Mars mission. Live radio show and live chat. Get your questions ready.
Brought to you by