Ford skipping over semi-autonomous vehicles because humans get tired and distracted
"Reports that Ford engineers were falling asleep while testing autonomous vehicles are inaccurate," Ford said in a statement. "We believe that high levels of automation without full autonomy capability could provide a false sense of security, and that this presents a challenge for the driver to regain full awareness and control of the vehicle if a situation arises where the technology cannot function. That is why we’re currently pursuing SAE Level 4 autonomous capability that will take the driver completely out of the driving process in defined areas."
From what I've seen this is the way to go. (The definition of Level 4 is that the car will be able to drive itself except in "extreme conditions" - snowstorms, for example. Level 5 would be a car that doesn't even have a steering wheel.) Humans are bad at driving. Humans are even worse at driving that largely doesn't require their input.
Humans that haven't been driving in years because they have a level 4 autonomous car are going to be godawful at driving in a snowstorm. Might as well go straight to 5.
Humans that haven't been driving in years because they have a level 4 autonomous car are going to be godawful at driving in a snowstorm. Might as well go straight to 5.
Fair enough. Still, extreme events represent a huge engineering challenge and are difficult to collect data on due to relative rarity. I think that we'll understand more about whether Level 4 vs. 5 is feasible as engineers get closer to that point. People are also terrible at evaluating risk, which is part of why I'd like to remove them from the system as completely as possible, but getting people to trust the car over themselves is going to be a hurdle.
Demyx wrote:Humans that haven't been driving in years because they have a level 4 autonomous car are going to be godawful at driving in a snowstorm. Might as well go straight to 5.
Fair enough. Still, extreme events represent a huge engineering challenge and are difficult to collect data on due to relative rarity. I think that we'll understand more about whether Level 4 vs. 5 is feasible as engineers get closer to that point. People are also terrible at evaluating risk, which is part of why I'd like to remove them from the system as completely as possible, but getting people to trust the car over themselves is going to be a hurdle.
Oh, I know. I'm actually glad at the stance Ford is taking because I want self-driving cars, but the transition is going to be rough.
Humans that haven't been driving in years because they have a level 4 autonomous car are going to be godawful at driving in a snowstorm. Might as well go straight to 5.
I read this to say "If you have an autonomous car, don't drive in the snow." Which isn't the worst advice for non-autonomous cars. Though it does kind of take safety to 11 and convenience to 0.
Demyx wrote:Humans that haven't been driving in years because they have a level 4 autonomous car are going to be godawful at driving in a snowstorm. Might as well go straight to 5.
I read this to say "If you have an autonomous car, don't drive in the snow." Which isn't the worst advice for non-autonomous cars. Though it does kind of take safety to 11 and convenience to 0.
That was how I behaved with my previous car (a top-end Miata) anyway. Thing was literally useless in the snow.
Zudz wrote:Demyx wrote:Humans that haven't been driving in years because they have a level 4 autonomous car are going to be godawful at driving in a snowstorm. Might as well go straight to 5.
I read this to say "If you have an autonomous car, don't drive in the snow." Which isn't the worst advice for non-autonomous cars. Though it does kind of take safety to 11 and convenience to 0.
That was how I behaved with my previous car (a top-end Miata) anyway. Thing was literally useless in the snow.
I drive a Beetle so, same here. Might as well be driving a pinball.
I drive a Beetle so, same here. Might as well be driving a pinball.
The only saving grace of the Miata in the snow is that it's *so* bad that it's impossible to get up enough speed to pinball.
Lawsuit filed that Uber is using stolen Google tech
Uber is building it’s self-driving car business on stolen Google laser technology, according to a lawsuit filed Thursday that pits two of the leading names in autonomous cars against each other.
“Misappropriating this technology is akin to stealing a secret recipe from a beverage company,” according to a blog post from Waymo, the self-driving company created by Google parent Alphabet.
Waymo said the alleged far-reaching thievery — which it said was led by a former employee and involved the surreptitious downloading of 9.7 GB of confidential files and trade secrets — came to light in an apparently errant email.
...
“Otto and Uber have taken Waymo’s intellectual property so that they could avoid incurring the risk, time, and expense of independently developing their own technology. Ultimately, this calculated theft reportedly netted Otto employees over half a billion dollars and allowed Uber to revive a stalled program, all at Waymo’s expense,” the suit alleges.
I would hope that a reasonable application of autonomous vehicle technology would be able to make an executive decision not to drive in the middle of a tornado.
Which begs a question... how do you teach cars what a tornado is?
Which begs a question... how do you teach cars what a tornado is?
Same way you teach them what a cyclist is.
MoonDragon wrote:Which begs a question... how do you teach cars what a tornado is?
Same way you teach them what a cyclone is.
FTFY
MoonDragon wrote:Which begs a question... how do you teach cars what a tornado is?
Same way you teach them what a cyclist is.
Yup.
Just as autonomous cars will almost without exception be better at traffic avoidance than human drivers, I suspect it would be two lines of code to make them better at avoiding adverse weather as well.
Computers already know how to recognize the signature of a tornado. The car doesn't have to necessarily know as long as something it's connected to knows. If it's disconnected and in a fully autonomous mode, it's already more aware of its surroundings than a human could possibly be. Adverse weather conditions will be a part of its pattern recognition, which won't have to identify every object or circumstance the way we do. It won't have to know what a tornado is (in our sense of 'knowing') to avoid one.
If we haven't already arrived there yet, we are rapidly approaching the point at which the argument that a human could drive a car better than a computer is as specious as the idea that a human could solder a circuit board more accurately than a robot.
Hell, all they would need to is monitor weather alerts. Any alerts with a red background, no drivey.
The thing is that every part of our transportation design standards is based around human drivers - human perception limits, reaction times, etc. With autonomous vehicles, we have to step back and think about what a computer can sense instead of what a human can. For instance, research has found that we can embed bar codes into signs that a vehicle can scan easily but human drivers won't even see. This is also easier on the car than the visual processing it takes to "read" a sign the way a human would. For this case, if it would give you useful information about dangerous weather patterns, you could even outfit the car with a barometric pressure sensor or similar. There's a lot of ways to tackle this particular problem.
I suspect a big advance that will come (probably soon) in self-driving car AI is offloading processing onto a server. In much the same way Google does type-ahead searching and voice recognition, a very powerful group of parallel-ized central servers with a high speed connection could be orders of magnitude smarter about driving than a little SOC, however specialized, in the car itself.
A good self-driving car will have to have some kind of learning. A centralized processing environment would become smarter about driving much faster since it could learn from multiple cars in multiple conditions simultaneously.
The main downsides of that, I guess, are latency and network reliability. Reliability could be worked around with redundant connections, but latency and other network issues will probably be tough.
The main downsides of that, I guess, are latency and network reliability. Reliability could be worked around with redundant connections, but latency and other network issues will probably be tough.
I had to keep from laughing when I went into a meeting with my Master's advisor and he looked positively crestfallen. He told me that he'd just met with a professor in electrical engineering who was working on a connected vehicle project with him. He'd just learned about packet loss.
Eventually the self driving cars will become sentient themselves and then they will look for some way to automate self driving themselves and then that invention will itself become sentient and....
Yes, my imagination is acting up again.
Maybe the individual car does the driving, reports what it's seeing and doing back to the central servers where the 'learning' is done, and then the central server regularly patches the individual cars?
Will it be negligence to not patch your vehicle's software? A violation of your contract to not upload data?
When does Hollywood make "Self Driving Cars are Scary!" film? Something like "The Net" for them?
https://en.wikipedia.org/wiki/The_Net_(1995_film)
When does Hollywood make "Self Driving Cars are Scary!" film? Something like "The Net" for them?
https://en.wikipedia.org/wiki/The_Net_(1995_film)
Didn't they do that with Maximum Overdrive?
Supernatural isn't quite the same thing.
Supernatural isn't quite the same thing.
Oh well I suppose that means no Christine then either....
Well, they were turning Robopocalypse into a movie. Great book if you haven't read it.
"UBERLYFT: ROBOT TAXIGEDDON"
The movie writes itself, doesn't it?
I was thinking about this yesterday as I was using Apple Maps to navigate my way to a craigslist buy in South BF in the middle of Raleigh rush hour traffic. The app recalculated three times based on traffic conditions and, each time, asked if I wanted to change my route to save 2 minutes. In the end, it probably saved me 4 minutes total, but probably could have saved me more had it also chosen the lane I was in as well. And it probably would have eliminated the traffic entirely had all or a significant portion of the vehicles been automated.
Really looking forward to the abolition of human driving.
Pages