Toyota chucks $50m at AI car tech Car manufacturer Toyota has begun working with researchers at Stanford and MIT in the US, having agreed to pony up $50m over the next five years to invest in robotics tech. That cash will be used to build joint research centres at the universities, the company said in a statement on Friday. Toyota has also roped in US Defense …
"... and potentially even helping us develop a vehicle incapable of getting into a collision."
A worthy goal, but only if they sort out the security of the systems involved. It's no good having the means to prevent shunts if at the same time the system is wide open to any ne'er do well who can use any flaws to take over control of the car and ram it into a wall or whatever takes their fancy.
"Safety first" doesn't just mean making sure that any "smart" system does as it is designed to do, it should also mean that it cannot do things it was not meant to do, and so far the track record of "connected" cars is not reassuring in that respect.
I'd feel safer if Toyota spent that money on fixing its existing code and development system. See "Toyota's killer firmware: Bad design and its consequences" at http://www.edn.com/design/automotive/4423428/2/Toyota-s-killer-firmware--Bad-design-and-its-consequences.
"Toyota's killer firmware: Bad design and its consequences"
I'm not arguing that Toyota's firmware wasn't defective (though the expert advisors in the Camry case did not prove that a bit flip occurred, they merely showed that unintended acceleration would have occurred had that taken place. I'm allowed to be unconvinced that the court would have found the same way - that this was what actually happened - had it been Ford or GM. At the time the state of the US autgo industry was an extremely politically sensitive subject).
However, the main killer on the roads continues not to be firmware but, by a very large margin, wetware. So improving the firmware to reduce the foulups of the wetware is likely to be quite effective. Since the design of car decision support systems is quite separate from fixing bugs, there is no reason at all why both should not proceed in parallel.
" and potentially even helping us develop a vehicle incapable of getting into a collision."
If there is a stationary obstacle within your braking distance and no way of steering around it then you will hit it. It doesn't matter how many MIT boffins worked on the software, if a vehicle comes out of a side street into your path immediately in front of you a collision is inevitable.
The majority of 'accidents' are not the slightest bit accidental at all, driver error is the problem (incorrect assessment of conditions leading to incorrect action taken).
If an asteroid hits your car it is an accident, anything else is.human/machine error (even a plane hitting your car).
"if a vehicle comes out of a side street into your path immediately in front of you a collision is inevitable." --- Smooth Newt
Unless the vehicle hits your door, this is almost always a failure of anticipation. You didn't see the sidestreet or you did but you didn't see the approaching vehicle was unlikely to stop, etc. If you *couldn't see* either of those things, you were almost certainly doing some combination of going too fast and paying too little attention.
"you were almost certainly doing some combination of going too fast and paying too little attention."
As someone who has been in an accident exactly as described (riding a motorcycle no less) I can assure you that's utter bollocks. The chunky bolts holding the handlebars got bent as I tried to hang on and the front frame strut buckled, yet the very reason I'm now able to challenge you is that I wasn't going fast at all precisely due to my paranoid nature - the side street was joining in at an acute angle and cars were parked on the side right up to the junction, so none of us saw the other right up to the crash. Yes, he should have been much more cautious but there never will be an autonomous car that could have avoided that crash without slowing down to an impractical crawl at every damn junction...
DropBear, I'm very glad you survived this collision. I wouldn't call it an accident because your description clearly shows the other motorist at fault. The fact that you hit him rather than him hitting you broadside is down to a bit of a knife-edge case though - for nearly all of the scenarios where a vehicle emerges from this sidestreet into your path, you would have been able to stop.
This is why I was at pains to use the term almost always. There may never be an autonomous car that could have avoided that motorist but there will certainly be one - in fact there probably already is one - that can avoid very many more of these situations than us humans can -- even just the reduction of a 200ms eye-to-brake latency translates to nearly 3m at 30mph, and that's before we factor in the ability of a computer to exert nearly maximum breaking nearly instantly.
I think you are both missing the point of autonomous vehicles. If both vehicles were autonomous then both would have communicated with each other and negotiated a mutual arrangement for handling the potential incident safely. The DARPA Challenge was an early attempt to address such issues, although there was no inter-vehicle communications to the best of my knowledge, but it's coming, and the sooner the better because I'm getting tired of dealing with the unpredictable behavior at the 4-way stop in my town. Sometimes I don't even realize what I'm going to do next, I'm ashamed to admit.
This post has been deleted by its author
Toyota should invest that in learning how to install an exhaust pipe.
They had some design competition going on, but so far the solutions are not convicing.
I'm amazed that most people seem to believe the following two things at the same time...
1) Self-driving cars will pick you up, drive you to work, and then (rather than being parked) will continue to drive around empty, on their way to do something else. Key point, driving around empty, not parked. Net increase in traffic, possible as much as doubled.
2) At the same time, self-driving cars will somehow reduce traffic congestion.
It reminds me of Manhattan, NYC. The streets were jammed with yellow taxi cabs. Almost all of which were simultaneously 'busy' (not available) and 'empty' (except for the driver of course). The inefficiency of having vehicles wandering around 'empty' does not help reduce traffic congestion.
The only rebuttal is some naive belief that 'magic happens here' where they have faith that the coder drones will figure something out.
'Self-Driving Cars' is going to be the next great A.I. disappointment. They're going to be much less capable and more easily confused than most are thinking.
Strong A.I. is very much like flying cars. It's been "coming soon" for 50+ years
The naivety on this subject is infinite.
very few people actually believe those two things. First, reducing congestion is pretty far down on the list of proposed benefits. My number one is not being run down by some idiot fondling their phone when they pretending to be driving. The car sharing idea is an unrelated fantasy. Sure, A.I. cars might make it a little more effective, but if it were such an attractive idea we'd all be using taxis already.
Second, this has nothing to do with "strong" AI. What is making self-driving cars possible is the Google approach of throwing massive computational resources and some clever domain specific algorithms at the problem and observing what works.
Finally, I'd take an AI car that sits there confused at a show-off cyclist (or whatever) over the Natural Idiot who plows on confident that their "above-average" skills will get them out of anything their stupidity can get them into.
Agreed. And speaking of phones, I just saw the future in a very upscale neighborhood recently. As a gated community it had a preponderance of golf carts plying its streets, and when school is out, as in weekends and after hours, most of the drivers are kids, some, half the legal driving age. From a fixed point I see at least a half dozen kid carts an hour pass, always with multiple kids aboard, sometimes as many as six, and usually blaring music, although that is beside the point. Which is... last week I saw a cart full of kids piloted by one about 12 years old, with no hands on the wheel because he was staring down at the phone in his hands. As this is a huge rich community (with 3 private golf courses, etc.) I am anticipating the sad day when multi-millionaire sues multi-millionaire over one of their kids killing or maiming another's.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
