3 ms·

I think it would be hard to argue that we are a long way from fully autonomous vehicles, since millions of miles have already been driven by autonomous vehicles

by Erik816 10y ago

I think it would be hard to argue that we are a long way from fully autonomous vehicles, since millions of miles have already been driven by autonomous vehicles. What remains is mostly regulatory hurdles, psychological hurdles, and fine tuning.

stephancoral 10y ago
> What remains is mostly regulatory hurdles, psychological hurdles, and fine tuning. And driving in non-inclement weather. And driving in places that aren't meticulously mapped by Google. And being able to refuel themselves.... I'm not denying the accomplishments of current autonomous vehicles, but there is still a ways to go technically.
qyv 10y ago
Don't underestimate the difficulty of those hurdles. Driverless cars are a technology that will be directly responsible for human deaths. We as a society don't have a legal or ethical framework to deal with that sort of thing yet. Driverless cars could very well be the first deployment of autonomous robots within the general public, that is significant.
exDM69 10y ago
> What remains is mostly regulatory hurdles, psychological hurdles, and fine tuning. I think this is too optimistic. Self-driving cars only work because they are in the minority. Can you imagine a city full of current state-of-the-art self driving cars and it not turning into a traffic chaos? In particular, I think that self-driving cars cannot work fully autonomously without communicating with other cars or the road infrastructure. And I am not aware of any effort to create a cross-vendor car-to-car communication protocol. There are trivial cases where pure autonomous following of traffic rules will lead to deadlock (or other issues). Take these two simple examples: 1) 4 cars arrive at the same time in a four-way intersection without signals or yields 2) on a three lane road, two cars from the outer lanes try to switch to the middle lane at the same time. These are by no means difficult problems (I'm sure there are more difficult cases too), but they must be solved and so far they're not being worked on.
- 05 10y ago
  You think you're smarter than a team of Googlers and that you can just think of something they haven't thought of in 10 years of development? 1) They have a random the 4way stop delay which means one of the cars will be the first to start moving again. 2) They would just both abort, then retry at a later time. It's very unlikely that the situation could repeat again on a retry unless you think perfect symmetry can be achieved in physical world:)
- walrus1066 10y ago
  Add to that, car to human-driven-car, and car to human (police officer, traffic coordinator) communication. Sometimes the communication needs to be complex (I.e. Verbal). I've had a few instances when I've had to negotiate to unblock a traffic deadlock, mostly when a two way lane is constricted to one vehicle only due to parked cars, or a HGV which is too big to drive past. In London this happens quite frequently. I really cannot think how an autonomous car, with no human inside, could handle these scenarios.
TorKlingberg 10y ago
Driving millions of miles is easy if you are driving circles around a test track. Which is not quite what current self-driving cars are doing, but they aren't driving on arbitrary roads in all conditions either. I am exited about autonomous cars, but I'm not convinced they are as ready as various companies want us to believe.
marssaxman 10y ago
The number of miles is not what matters, but the number of trips in which a human driver can be reliably replaced by a robot. It doesn't matter how far the vehicle can go; what we care about is the degree of confidence we can have that the vehicle will reach its destination without incident. Designing for the unexpected edge cases is harder than designing for the normal cases. In accordance with Cargill's ninety-ninety rule, we can expect that the "fine tuning" you mention will be turn out to be a very big job. As an extreme demonstration, consider the fact that you could rack up a million "autonomous" miles in a matter of months, using vehicles capable of nothing more than "lane keeping assist" and "adaptive cruise control" - just set 'em loose on a NASCAR loop and run them around in circles until this artificial example's point has been proven.

by Erik816 10y ago

stephancoral 10y ago

> What remains is mostly regulatory hurdles, psychological hurdles, and fine tuning. And driving in non-inclement weather. And driving in places that aren't meticulously mapped by Google. And being able to refuel themselves.... I'm not denying the accomplishments of current autonomous vehicles, but there is still a ways to go technically.

qyv 10y ago

Don't underestimate the difficulty of those hurdles. Driverless cars are a technology that will be directly responsible for human deaths. We as a society don't have a legal or ethical framework to deal with that sort of thing yet. Driverless cars could very well be the first deployment of autonomous robots within the general public, that is significant.

exDM69 10y ago

> What remains is mostly regulatory hurdles, psychological hurdles, and fine tuning. I think this is too optimistic. Self-driving cars only work because they are in the minority. Can you imagine a city full of current state-of-the-art self driving cars and it not turning into a traffic chaos? In particular, I think that self-driving cars cannot work fully autonomously without communicating with other cars or the road infrastructure. And I am not aware of any effort to create a cross-vendor car-to-car communication protocol. There are trivial cases where pure autonomous following of traffic rules will lead to deadlock (or other issues). Take these two simple examples: 1) 4 cars arrive at the same time in a four-way intersection without signals or yields 2) on a three lane road, two cars from the outer lanes try to switch to the middle lane at the same time. These are by no means difficult problems (I'm sure there are more difficult cases too), but they must be solved and so far they're not being worked on.

05 10y ago
You think you're smarter than a team of Googlers and that you can just think of something they haven't thought of in 10 years of development? 1) They have a random the 4way stop delay which means one of the cars will be the first to start moving again. 2) They would just both abort, then retry at a later time. It's very unlikely that the situation could repeat again on a retry unless you think perfect symmetry can be achieved in physical world:)
walrus1066 10y ago
Add to that, car to human-driven-car, and car to human (police officer, traffic coordinator) communication. Sometimes the communication needs to be complex (I.e. Verbal). I've had a few instances when I've had to negotiate to unblock a traffic deadlock, mostly when a two way lane is constricted to one vehicle only due to parked cars, or a HGV which is too big to drive past. In London this happens quite frequently. I really cannot think how an autonomous car, with no human inside, could handle these scenarios.

TorKlingberg 10y ago

Driving millions of miles is easy if you are driving circles around a test track. Which is not quite what current self-driving cars are doing, but they aren't driving on arbitrary roads in all conditions either. I am exited about autonomous cars, but I'm not convinced they are as ready as various companies want us to believe.

marssaxman 10y ago

The number of miles is not what matters, but the number of trips in which a human driver can be reliably replaced by a robot. It doesn't matter how far the vehicle can go; what we care about is the degree of confidence we can have that the vehicle will reach its destination without incident. Designing for the unexpected edge cases is harder than designing for the normal cases. In accordance with Cargill's ninety-ninety rule, we can expect that the "fine tuning" you mention will be turn out to be a very big job. As an extreme demonstration, consider the fact that you could rack up a million "autonomous" miles in a matter of months, using vehicles capable of nothing more than "lane keeping assist" and "adaptive cruise control" - just set 'em loose on a NASCAR loop and run them around in circles until this artificial example's point has been proven.