Stop lights, sunsets, junctions are tough work for Google's robo-cars After cruising two million miles of public roads, Google's self-driving cars still find traffic lights, four-way junctions and other aspects of everyday life hard work. To be sure, the hardware and software at the heart of the autonomous vehicles is impressive. But it's just not quite good enough yet to be truly let loose on …
The fundamental part of a self driving system, before ANYTHING else, is being able to see - i.e. decent cameras. I don't care what clever tricks you can come up with in terms of radar/lidar/ultrasonics, no camera (that would be fitted to a car anytime soon) can match the human eye's abilities. As humans, if our vision becomes impaired, our license is taken away. Until a camera can match the human eye, self driving is a dangerous pipe dream. In fact, even if there were cameras that could match or beat the human eye's abilities (and were cost effective enough to put in a car - all about the margins remember), self driving cars are an accident waiting to happen - especially the ones planned with no manual controls! What happens when there's a system crash - software never has bugs, right? System hack - auto makers are great at security, right? EMP? Radar jamming? IR floodlight? What about all the other cars with ultrasonics pinging about, telling me they're all going to happily co-exist?
It's progress for the sake of progress, it's already killed once, it won't be the last time and it'll lull the great unwashed into a false sense of security, then they die.
Fuck ABS, fuck traction control, fuck parking sensors. Strip all that shit back and teach people how to drive properly. Gimme a Mk2 Escort with a roll cage and harnesses any day of the week.
</rant> :)
The 1907 Peking to Paris race victor won a bottle of champagne after a trip of 9,300 odd miles. Trains and camels delivered fuel and new tyres to servicing stations, but the car and passengers were on their own between stations. The race has been re-enacted five times, and more recent participants benefitted from truckers to drag them across flooded rivers.
Peking to Paris was organised 20 years after the London to Brighton run -- the run celebrated the lifting of speed restrictions in the UK, but it was used by car sellers to demonstrate the potential of their wares. London to Brighton was a day trip (if the car worked) but getting stuck in Siberia?
If the vendors of autonomous cars are serious, they have to build a car that goes from Peking to Paris on its own. One of the great things about autonomous vehicles is that nobody needs to die as part of the experiment. There will be times when every car needs to cross a river on the back of a weird looking truck. Aside from the time on the truck, the car has to drive itself.
After all, it's only 9,300 miles and Google tells us that their cars have driven millions of miles.
Because I love the act of driving so, I cannot wait for these autonomous cars to become common.
And then, I'm wanting hacks for other cars as well as for the traffic lights. Other cars can go into meek & mild mode, and the lights will always be green! (Untill they're blue and on the roof of the guy behind me, probably!)
This is the most convoluted, expensive, and inefficient way to solve traffic problems. Adding more cars to the roads is what contributes to traffic in the first place, and all these promises of efficiency gains from automation haven't been proven with any of the prototypes so far.
Stop building cars! The only way to make more efficient transit is to reduce the number of variables, and Google is basically saying roads are too unpredictable to for autonomous cars. This is a problem that many countries have addressed by investing in TRAINS, which aren't as susceptible to weather.
Come on, people. The United States highway system was originally built to aid the defense of military bases in the event of an invasion, because Eisenhower was a paranoid mess after WWII. It took until 2001 for the next major attack on U.S. soil, and in that case the traffic jams in downtown NYC only exacerbated the situation.
There is no defensible excuse for failing to expand railways in the United States. And seeing Google asking for government assistance to help roll out the cars is just rubbing salt in the wound.
Fucking stupid.
Personally, I like to think of the highway system as a gift to GM, Ford and Chrysler. After all, they had to convert back from military production to civilian production - after having increased their production capabilities considerably during the war.
As we have seen with the Tesla model S, a rush to market mentality on self-driving cars can be lethal. With two or more fatalities linked to the tesla S when operated in autonomous mode, it's clear that Tesla and engineering friends have failed to build mission critical class AVs with fail safe systems - which should be mandated by all federal governments and documented to be fail safe before these vehicles are allowed on roadways.
Google like Tesla has had problems with AV operation with numerous accidents and mechanical failures because they are going about the entire project bass ackwards. You design a fail safe mission critical system and then you build the car around it instead of trying to retro fit a cobbled together system that isn't adequate nor reliable.
It might need to understand yellow lights. I am reminded of a scene in _Taxi_:
What does a yellow light mean?
Go slower.
What....does....a....yellow....light....mean?
Go slower!
What........does........a........yellow........light........mean?
GO SLOWER!
I'm........going.........as........slow.......as........I........can!
(continues...)
My question has always been: Will the self driving car park in my garage (so I can get out of the car!)?
The accepted practice here in Vancouver for turning left is to get into the intersection as far as possible, then wait for the oncoming traffic to finish coming through, usually partway or completely through the yellow period, then complete your left turn. How does a self-driving car deal with this situation?
Our traffic is now sufficiently heavy throughout the day and evening, that if you don't follow this system, no one would be able to turn left at intersections that lack a discrete left turn signal.
Autonomous-car-friendly infrastructure isn't really out the window, it's just not a short-term solution. But if such cars become popular enough, the technology and infrastructure will start being improved organically.
Not to say it won't take time, though. Google et.al. have to work with what they have for now. Just saying there's a difference between "never" and "indefinitely".
Still won't help with kids running into traffic, of course.
I am sure that Google will use streetview to help automated navigation easier .... compare what you are seeing with what Google has previously analysed to be the edges and features to fix on, and the differences are where you need to apply processing power.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
