I disagree with this idea self-driving cars aren't coming. Maybe later than expected, but they are.
I don't see the "pogo stick" thing as an unsolvable issue. You don't need to recognize every single weird thing that could show up on a road, you just need to have a way to deal with an unexpected obstacle showing up, and defaulting to gracefully braking.
If somebody decides to exploit this by randomly wandering into the road, they'll do it in front of multiple cameras per car, which can be easily used to fine such people until it stops happening with any frequency.
For some reason the self-driving car discussion gets bogged down in things that don't really need realistic consideration. Eg, the Trolley Problem. There's no need for a car to make some sort of AI driven decision whether "child" or "woman" is the best thing to run over, when it can just detect a generic obstacle and brake.
The big problem right now is more that Tesla is trying to do it on the cheap and trying to do the job with too few sensors.
Solving self driving is not a problem of using the correct sensors. Is a problem of understanding the semantics of the situation ahead of you.
- Does this ball running into the road, means a child might be running behind it?
- Does the pedestrian on the sidewalk with headphones and looking at the other side, not seeing my vehicle and the red light, is going to cross anyway?
- Since there is crashed truck on the road ahead, does this police officer making me signs to temporarily cross to the other side of the road, wants me to drive ahead ? Do I see its Ok anyway, because his colleague on the other side is holding the traffic coming this way?
- Can I go off road for 2 minutes because the hole in front of me is too big and the off road track looks good?
The discussion of how sensors are better than humans, and the systems don't get tired compared to humans misses all these. You can't solve a problem you started to work on under false premises or considerations of the problem space.
> Does this ball running into the road, means a child might be running behind it?
There's no ball. There's merely a non-car object on the road, which means you smoothly stop and alert the driver.
> Does the pedestrian on the sidewalk with headphones and looking at the other side, not seeing my vehicle and the red light, is going to cross anyway?
If an object intrudes on the road, and could collide with the car, brake as per above.
> Since there is crashed truck on the road ahead, does this police officer making me signs to temporarily cross to the other side of the road, wants me to drive ahead ? Do I see its Ok anyway, because his colleague on the other side is holding the traffic coming this way?
There's an obstruction on the road. Brake, alert the driver, have them figure it out.
> Can I go off road for 2 minutes because the hole in front of me is too big and the off road track looks good?
No. Self-driving cars only automatically drive on a proper road.
> The discussion of how sensors are better than humans, and the systems don't get tired compared to humans misses all these.
My view is that this kind of thing just tends to over-complicate the issue. The initial goal shouldn't be an AI version of a human driver, a system that recognizes and knows about balls, children and police officers.
It should be the car equivalent of an elevator -- a dumb machine with safeties that automatically stops if anything isn't right and waits for a human to figure it out. An elevator doesn't know what's standing in between the doors. Something is, therefore the door doesn't close until it goes away.
> It should be the car equivalent of an elevator -- a dumb machine with safeties that automatically stops if anything isn't right and waits for a human to figure it out. An elevator doesn't know what's standing in between the doors. Something is, therefore the door doesn't close until it goes away.
Here is your issue: you are describing an automated train. A remote operator can pilot it if anything odd is happening. This model doesn't scale well for individual cars, especially on open roads where everything can happen. For this you need a driver inside the car. A self-driving car requiring a driver is not that "self-driving".
> A self-driving car requiring a driver is not that "self-driving".
That seems more like a philosophical issue than engineering at that point. All technology requires some amount of monitoring, maintenance and troubleshooting. You don't need to solve every possible weird event happening for automation to be useful.
The main two use cases as I understand are:
* A qualified driver sits in the car, just doesn't want to drive. They can be asked to assume control any time it's needed.
* The vehicle forms part of some sort of large fleet, and can be managed centrally.
You'll always want some amount of central control for a fleet. What if the car breaks down? What if somebody else crashes into it? What if somebody tries to steal it?
> There's an obstruction on the road. Brake, alert the driver, have them figure it out.
That is the whole crux of the problem. There is no driver, it was supposed to be self driving. Like many of the "self-driving taxis" in San Francisco whose passengers might not even have a driving license.
> No. Self-driving cars only automatically drive on a proper road.
Not good enough. All those human driven cars handled the situation is seconds.
Now all your "self-driving" cars are causing a huge traffic jam. In the meanwhile the police has just open a second lane for the human driven cars to overtake the AI ones....
> That is the whole crux of the problem. There is no driver, it was supposed to be self driving. Like many of the "self-driving taxis" in San Francisco whose passengers might not even have a driving license.
We have the internet. For such a problem, call the home base, have a remote operator figure it out.
> Not good enough. All those human driven cars handled the situation is seconds. Now all your "self-driving" cars are causing a huge traffic jam. In the meanwhile the police has just open a second lane for the human driven cars to overtake the AI ones....
I don't imagine we're going to have fully autonomous cars any time soon. There will need to be a human in the loop somewhere.
On a longer term, I can see a police officer having some sort of device or signal, hopefully authenticated, to give the car an understandable command.
> The big problem right now is more that Tesla is trying to do it on the cheap and trying to do the job with too few sensors.
I think you are oversimplifying the current issues with self-driving car and a bit too optimistic on industry's capacity to build reliable software for a self-driving car.
The question isn't whether it's possible to constrain the road environment such that self driving cars can work, but whether it's desirable. The article is simply answering in the negative.
"That line there at the end is the game. Self-driving cars are just worse trains, as evidenced by the time Musk had to actually attempt to deploy his vision, or lack thereof. These cars will require physically modifying the environment to make some sort of... let’s call it a “track." Had we spent those $100 billion dollars on a massive overhaul of American public transit, we could have already built the infrastructure we need to solve this crisis. Instead, we have made a few tech entrepreneurs very rich, and we are left with the same political problem as before."
This is a pretty succinct indictment of capitalism. I think it was good to grow out of feudalism/mercantilism and I think central planning was a failure, but the current problems are:
- bonkers inequality
- bonkers waste
Weirdly I think these problems are solved by taxation. It shouldn't have been possible to burn $100B on self-driving cars. Companies and billionaires aren't good at spending this kind of money, especially when we have so many humans in poverty and a looming climate change disaster.
SUVs are another example of extreme private waste.
I have a lot of sympathy for the individual decisions of my neighbours to purchase range rovers so they don't have to endure the discomfort of potholes. However, it'd be much more efficient if we could somehow crowdfund that same expenditure towards improving the road surface so that all vehicles feel stable.
Unfortunately general taxation is quite unsuccessful at fixing these infrastructure issues because resources always get allocated to whatever stakeholders have the most political power. No matter how big a government budget gets all the money seems to get absorbed into defence, healthcare, education, pensions and general administration.
> The message is clear – NPR expects its readers to be passive consumers of climate change solutions.
I get the opposite impression from its headlines, at least when applying Betteridge's law: NPR presupposes that the technological solutions will be insufficiently effective, and that there's nothing to passively consume. All we can do is cope and hope that the socioeconomic and political systems that created this problem in the first place will somehow fix it.
> SpaceX promises perhaps the dumbest solution to climate change possible: colonizing space. Elon Musk has convinced legions of people that it will be easier (or maybe better? more desirable?) to colonize Mars to save humanity than to avoid destroying the only planet where food literally grows on trees and water falls from the sky.
1. Elon Musk is not the (only) one doing that convincing; the idea that we could and should be expanding humanity's reach into the Solar System and beyond instead of confining it to a single planet that's already straining to support us predates Elon Musk's very existence.
2. It ain't just Mars. Mars is attractive because it's close and it has some far-future hope for terraforming, but as far as sustaining human life goes it ain't that great; there's a whole asteroid belt full of worlds chock-full of both water ice and hydrocarbons (meaning they've got the raw resources for food and water without needing to rely on Earth), and beyond that are the gas giants with dozens (hundreds, maybe) of moons with similar abundances of water and hydrocarbons.
3. Keeping the entirety of humanity confined to Earth is fundamentally unsustainable without either some very dramatic decreases in quality of life or mass exterminations that make the Holocaust look like child's play; even if we can freeze the population at its current number, the cat named "industrialization" is already out of Pandora's box, and the developing world is, well, developing. Space colonization is a vital pressure relief valve; the optimal end-game is to get the entirety of human industry and the vast majority of humans off Earth, so that Earth's biosphere can heal from the damage we've done to it over the last few centuries. The fact that Earth is the only planet where food literally grows on trees and water falls from the sky is exactly why preserving it is of utmost importance; anything that can be done beyond Earth must be done beyond Earth if we expect there to be any trees or water left on it in the coming centuries.
Colonization of other planets/solar systems/galaxies by human bodies is so ridiculously difficult that becoming post-human becomes the practical choice.
I disagree, based on the fact that we're continuing to make further and further progress on manned space exploration v. borderline zero progress on becoming post-human; it's readily apparent, given existing progress, that the former is the more practical choice in the relatively-near-term by a rather wide margin.
Climate change has to be technologically solved! (Or it won’t)
- The coming massive Electrification of Industries needs technological solutions, from batteries to hydrogen infrastructure and new industrial processes
- We need carbon capture, because not every CO2 is done by burning fossil fuels. For example concrete is a massive CO2 emitter which is not talked enough about it (aside from Bill Gates)
- Public transport surely can help and should be fixed in the US, but it won’t save us. That is like a wishful thinking antisolution. Because billions of people in Africa, in India and Asia want mobility and will use an individual vehicle for that. Technology has to provide that.
We shouldn’t bank on “silver bullets” like Fusion, but we do have silver bullets like Solar and Battery tech and Electrification.
I don't see the "pogo stick" thing as an unsolvable issue. You don't need to recognize every single weird thing that could show up on a road, you just need to have a way to deal with an unexpected obstacle showing up, and defaulting to gracefully braking.
If somebody decides to exploit this by randomly wandering into the road, they'll do it in front of multiple cameras per car, which can be easily used to fine such people until it stops happening with any frequency.
For some reason the self-driving car discussion gets bogged down in things that don't really need realistic consideration. Eg, the Trolley Problem. There's no need for a car to make some sort of AI driven decision whether "child" or "woman" is the best thing to run over, when it can just detect a generic obstacle and brake.
The big problem right now is more that Tesla is trying to do it on the cheap and trying to do the job with too few sensors.