Slashdot Mirror
Software Update Adds Autonomous Driving To Tesla's Bag of Tricks (nytimes.com)
An anonymous reader writes with the news that Tesla owners today found their cars had been upgraded with the company's new autopilot feature: "That means the next time you see a Model S cruising next to you on the interstate, look closely: It may be driving itself."
Adds the submitter: Well, I guess some of you will be celebrating this; but this submitters' fear, is that if this technology becomes pervasive, the skill of operating a vehicle will be lost, as is any skill that isn't practiced regularly. It is unlikely that 'self-driving cars' will reach a point where they can handle 100% of all driving circumstances without human intervention, emergency circumstances being the first and foremost example of what an automated system could not adequately handle unaided; what will we do then, when injuries that could have been avoided or when lives are lost because people aren't competent to operate a vehicle any longer?
It costs $2,500 to unlock this new software feature.
Better known as 318230.
One might argue that many drivers on the roads today already aren't particularly proficient at controlling their vehicles. While it might be that some persons skills would grow worse with disuse, I think there are a goodly number of individuals out there who would be safer 'drivers' if they weren't in direct control over their cars themselves. And I don't mean just those who have poor eyesight or slow reflexes.
... "I read part of it all the way through." -- Movie Mogul Sam Goldwyn (and some slashdot readers)
People barely have any skill at that *now*.
We already have incompetent people killing others in mundane situations due to carelessness and incompetence
I told them when GM introduced its new fangled hydramatic transmission, it is going degrade the driver's skill, soon no one would know how to declutch and shift. And I was proven right. I was just bragging about my prediction coming true the other day and my grandpa chimed in. "Son, the slippery slope goes way back. I never liked them self starter anyways ... Nothing like cranking up the old tin lizzy with a cranking rod to fully wake up in the morning" he went.
sed -e 's/Chuck Norris/Rajnikant/g' joke > fact
Computers:
* See and process information from all directions at once
* React in a millisecond to changing conditions
* Never get bored, tired, or distracted
* Don't drive recklessly for thrills
The notion that humans will actually react better than an automated system in an emergency seems backwards to me. I expect a computer to react much more competently and predictably, if for no other reason than the computer can analyze and react a thousand times faster. It's humans that are *causing* most of the emergencies in the first place by needlessly driving into each other at high speeds.
Irony: Agile development has too much intertia to be abandoned now.
Sure humans might lose some of their proficiency at controlling a vehicle but the self-driving car would make those skills less necessary.
Which would have fewer fatal accidents: automated vehicles with a human with poor skills or a standard car with normal everyday drivers?
I am betting the automated car wins. Sure the automated car may have some accidents that the human might avoid but I'm betting the total goes down.
Now I don't want an automated vehicle but that is because I really enjoy driving but the accident thing is IMO a red herring.
but but but but what if Scotty beams down 500 orphans with their arms linked in a circle around your car while you're going 200mph?! What will your car do then?!?!
I am getting tired of all of the "Which should an autonomous car hit" questions when the answer is "Neither because if the car is functioning correctly, the car sensors should have picked up the little old lady as soon as she stepped into the street, and the busload of school kids when it came around the corner a block away, and will have decided the path to take to avoid every single obstacle within a few milliseconds or come to an ABS-assisted stop." I think people have joked so much about the light pole just jumping out in front of you that they are actually beginning to believe that can actually happen. Sure, someone might throw themselves off an overpass immediately in front of you and they're gonna die, but a human would have hit them too.
There are serious objections to autonomous driving (sensor reliability being the top one) but people are fixated on whatever moral alignment the car will have (sign me up for Lawful Evil).
If I have been able to see further than others, it is because I bought a pair of binoculars.
People arguing that some people will get hurt because a human no longer knows how to control the vehicle in an emergency are like anti-vaxxers saying their one child *might* react negatively to the vaccine. Both groups are ignoring the 99.9% of cases where people will NO LONGER BE DYING from STUPID SHIT.
I can't speak for everyone but I have these type features in my car (adaptive cruise, lane assist, proximity warnings, blind spot detection, etc) and I can say without a doubt for me it surely hasn't made me a better driver. I get into my other car without all of that and find myself making noob driving mistakes (not checking blind spots, not keeping consistent speed, much longer parallel parking, etc). It's actually kind of unnerving at how fast I came to rely on the car to do these tasks for me.
I haven't seen any studies so maybe I'm just a goof but I consciously try not to rely on those things because I don't want to forget how to actually drive.
I live in a very rural area where houses and intersections were at least 1000 feet off from my GPS. My house showed up over a mile from the road and the the road was labeled with the wrong name. Then one day about 4 or 5 years ago everything was accurate. Super accurate as a matter of fact. I looked at the little man icon on the map, clicked it and noticed that a Google Street View car, or whatever they're called, drove by a couple months before that (based on what was shown in the picture) and the maps were updated. When the maps were wrong, street view was not available in my area. I would also assume Google, and everyone else, takes GPS readings from your phone and updates maps where they haven't sent their own equipment. The problem you envisioned will be corrected long before the first snowfall and an autonomous vehicle meet.
I get that _you_ may feel safer if something else does things for you but lets be realistic about the numbers and risk. Fear mongering is not how you go about advocating change, but that is what you are attempting to do. The appeal to emotion is way too obvious.
To start, we are moving the numbers to more recent 2013, in which you had a .0088% chance of a fatal car crash.
By comparison, you had a .17% chance of dying do to heart disease, a .02% chance of dying from diabetes. You had a higher chance of death by suicide and influenza than you did from a car wreck. (math done using a sample size of 350,000,000 and numbers from the CDC and here (easier to find than numbers hidden in the bowels of the CDC PDF).
The point is there are lots of risks in life. Breathing in a lung full of air could cause you to catch influenza, or pneumonia. You are way more likely to DIE from those things than by driving a car, even with shitty drivers on the road. Eating poorly, not exercising, and ingesting the wrong substances (carcinogens) are exponentially more deadly than cars.
If you want to push self driving cars I'm fine with that. You can buy one and do as you wish. Current technology does not make them that much better than humans. Come to Mountain View and drive around near one. They can't differentiate between a speed limit sign and a "during school speed limit" sign so we end up having big backups on some main roads because of those cars. They don't accelerate any faster than my grandma, and don't break any better or worse than a person either.
One day I'm sure they will be great, but that day is not today. I would still rather have the option of manual versus no control of the car. Think about tyranny and extortion for a minute, and that can be corporate as well as government.
-The wise argue that there are few absolutes, the fool argues that there are no probabilities.
So what if driving skills are lost. How many people can genuinely start a fire without a match, lighter or some other ready to go ignition device? How about those people who can actually remember the composition of gunpowder, and if they can know a way to actually get those ingredients? Ok, now about how to skin an animal, how to hunt, how to build shelter?
If driving a car goes the way of riding horses then skills are lost to the general public and only retained by those with a particular interest in them. And you know what? Nothing of value was lost.
Such as when the center engine of a DC-10 goes "bang", cutting off the hydraulics and the only control is from adjusting the throttles of the remaining two good engines?
And for atrophied skills, consider Air France 330 (IIRC) from 2009, which was flown in a controlled stall into the Atlantic Ocean because the co-pilot forgot that recovery from a stall requires the nose to be pushed down. Original cause was autopilot decoupling when the pitot tube got iced up.
Main problem with computer control is trusting that the people writing the software properly anticipated all of the situations that could be encountered. The quality of most code leaves me with a bad feeling about this. An example, an Airbus on a demonstration flight crashed because the software countermanded the pilot's attempt to pull out of a dive, the software was trying to prevent excessive g-loads but the programmer didn't consider that hitting the ground would be worse than bending the airframe.
A Shadeless room is a brighter room.
You will find it in a small fraction of the aircraft flying today, at a small fraction of the airports. It requires special certification for the aircraft and the pilot and the airport. It is not a commodity item that Joe Pilot can have Frank Mechanic install in his C182 and then pass control of the airplane over to Joe Junior the eight year old prodigy.
All true, but not likely for the reasons you might think...
Everything in aviation is just stupid expensive, for several reasons... The first is that it is an amazingly small market, so there is no volume to absorb R&D costs... The second is that it is completely and totally regulated by the FAA which is a very conservative organization. Perhaps rightly so in many ways, but having witnessed it firsthand, I can say that it has no incentive to change.
Using such systems does not make getting the pilot license easier, either. A large part of the training for pilots of those aircraft is not "how to use the autopilot", it is "how to disable the autopilot when it fails".
Actually, most aircraft that are used for learning to fly, don't have autopilots. The newer 172s do, sometimes, but the vast majority of training planes have no autopilot.
As for not learning "how to use the autopilot", you'd be correct, but that is a massive mistake. There is some very old thinking that says that using an autopilot makes you a worse pilot. Nonsense, the autopilot can fly better than you can, what you need the human for is decision making and problem solving, something the computer isn't as good at. If you're busy moving the controls, you're not paying as much attention to the big picture.
I have thousands of hours of dual instruction given, I spent 3 years as the chief flight instructor of a FAA part 141 flight school, I've signed off dozens of initial CFI applications. Pilot training is really outdated.
As I recall, the C182 I fly that has a G1000 glass cockpit has at least nine different ways of disabling the autopilot, and at least two of those methods are part of every pre-flight check before every flight just to make sure they still work.
The autopilot in that system is amazingly basic and it should be easy to disable. There is just no money to make a good one, it would raise the price of the plane too much due to the low volume of sales of light aircraft. They already are crazy expensive.
This problem won't exist with cars. In one month, more passenger vehicles are sold than all the light aircraft in the history of aviation. Cessna has only ever made just over 50,000 C172 in the history of the company, all their aircraft combined barely breaks 100K, and that is over 50 years and they are one of the largest light aircraft builders in the world.
Build 30 million of something in a year and suddenly spending $10 billion to develop it properly becomes reasonable.
No, I'm sorry, the aviation autopilot is a much simpler device and is not the proof-of-concept for AV cars.
Try a Global Hawk:
https://en.wikipedia.org/wiki/...
"Global Hawk is intended to operate autonomously and "untethered" using a satellite data link (either Ku or Ultra high frequency) for sending data from the aircraft to the MCE."
It is not a "Remote control airplane", it has to be able to take off and fly anywhere in the world without direct control from a pilot. It is expensive and there have been crashes, but it also was tossed into operations while still in prototype stage due to 9/11.
How about if the autonomous car just stopped itself as quickly as possible in the case of an incident that it cannot handle?
What could possibly go wrong?
systemd is Roko's Basilisk.
I think you'll be able to find cases where humans react better and cases where computers work better.
This is true, for now... but consider what happens over time: every time a computer does something sufficiently poorly (i.e. badly enough to cause an accident), there will be a full black-box recording and log of the conditions and operations that led up to the accident. The car company's programmers will go over the situation with a fine-toothed comb to understand what happened, and update the software to handle that situation better in the future.
Rinse and repeat for a decade or two, and the number of scenarios where the car is still worse than a human will start to become quite small.
(Meanwhile, human beings will continue to drive at more or less their present skill level, since they don't learn much from each others' mistakes)
I don't care if it's 90,000 hectares. That lake was not my doing.
Actually, I think bad weather is one of the places where automated cars will make a very positive impact on safety. We already have a limited form of this technology with things like anti-lock brakes, traction control, stability control, etc. If you've spent your entire life driving cars with these safety features, they probably feel normal to you. But, as the article suggests, once you come to rely on these features, you lose your ability to handle the vehicle safely in the absence of them. Have you ever pumped your brakes to prevent skidding? Do you know how to steer out of oversteer? Can your brain detect these conditions and react to them before you are in a dangerous situation? For most people, the answer is "probably not".
* Never get bored, tired, or distracted
I'd add to that list:
* Never drive preoccupied or in emotional imbalance
* Never drive intoxicated or on drugs
Let's face it, we don't leave the rest of our lives behind when we get behind the wheel. If things are troubling or exciting at home or at work or in your love life or with your friends or relatives the mind is churning on it. And while I don't know many who will blatantly drive drunk, I think quite a few have pushed it with hangovers and such. It certainly doesn't take much to drive better than humans at their worst...
Live today, because you never know what tomorrow brings
You forgot about the part where a sensor fails, the AV is the cause of a multiple fatality accident (e.g., plows head on to a school bus), the entire country hates on the automaker, and the families of the victims all sue the automaker.
I can see the fnords!
How well do they handle signals by flagmen, police officers and so on? As far as I know, no autonomous system to date has the ability to see and correctly interpret traffic control flags or hand signals. (for that matter, how would one program a car so as to recognize a cop or construction workers hand signals but treat bicyclists hand signals differently and ignore non significant gestures by pedestrians, other drivers etc?)
Right now, as far as I know, they will correctly avoid barrels or pylons, but only by treating them as static objects to be navigated around, stopping if it can't figure out a safe path between or around them. There is no special rule set that tells it "objects of these shapes and colour combinations indicate a construction zone or accident site, switch to rule set B (for slower speeds, more weight given to moving objects in the sensor periphery etc)" Back when I was on the road crew, close calls by confused or distracted drivers was a daily occurrence. Sure, the computer is never distracted (one hopes!, the computer equivalent I guess would be wrongly weighting one set of inputs over another) but it would be easier to confuse it, especially when there are multiple workers in safety vests pointing and signalling to each other within the same view arc as the flagman or cop.
A related issue would be properly navigating the thicket of pylons or traffic "barrels", correctly following the temporary lane(s) and not mistakenly taking an opening in the pylon line right into the work site. This particular problem could be at least partly dealt with by more standardization on work site markings, minimum and maximum distances between pylons tightened up. On the car end, the software would have to allow for correct navigation between said pylons when the usual road markings are absent, indeed, even the usual pavement is missing.
As it stands now, construction and accident sites I think are places where the autonomous vehicle just gives up and signals the driver to assume control. Thing is, one of the hoped for benefits of autonomous vehicles is the ability to have a non-driver, sick, sleeping or drunk driver to safely get from A to B. And I'm sure the transport industry is looking forward to when they can have only a single driver or perhaps even no driver at all, allowing the truck to go non-stop. None of that is going to work very well if the vehicles can't handle a construction site.
I need a wheelchair van for my son. Help me get the word out. https://www.gofundme.com/wheelchair-van-for-jj
Can you give details of this Airbus demonstration flight crash? I think you've got a garbled version of Air France flight 296. The plane did refuse the pilot's command to raise the nose, but that was to prevent a stall. At the time this pilot/plane conflict occurred, the pilot had flown the plane into a state where a crash was inevitable.
Inspired by the United Airlines flight 232 (the DC-10 crash you cite), software has been written to control planes by differential thrust, and to do so better than people can. As I recall, it was deemed to expensive to put it into service. (I'm aware of two incidents since then when such software might have been used, and it probably would not have changed the outcome for either. Japan Airlines 123 likely was unlandable even with computer due to vertical stabilizer damage, and the DHL A300 damaged by a SAM at Baghdad in 2003 was landed successfully manually.)
I agree with you that "Main problem with computer control is trusting that the people writing the software properly anticipated all of the situations that could be encountered." However they don't need perfection - so long as they anticipate enough that computer-caused catastrophes are less common than human-caused catastrophes they prevent, it is a safety win.
https://en.wikipedia.org/wiki/...
https://en.wikipedia.org/wiki/...
https://en.wikipedia.org/wiki/...
Quattuor res in hoc mundo sanctae sunt: libri, liberi, libertas et liberalitas.