Pulling that permit might have been legally questionable, but the real story is way more complex than that, because the legality of the permit was in question and had been under scrutiny by the courts for the entire period in question. Their decision to start building in spite of the permit potentially being illegal was a mistake, and the losses from such a mistake were entirely their responsibility.

I find it difficult to imagine how they could have come to such a dubious conclusion. Oil sands are some of the dirtiest oil you can get, and encouraging the use of oil sands refining before other, cleaner sources of oil is not sound environmental policy. And making that oil easier to import into the U.S. would doubtless encourage more extraction.

A permit, once denied, isn't generally eligible for being reinstated without correcting the issues noted in denying it. They did not correct anything. Instead President Trump issued a permit himself outside of normal regulatory channels, overriding the decision of those regulatory channels, with a complete lack of environmental review, likely violating dozens of federal laws. The legality of such a presidentially issued "permit" is dubious at best, and that legality was being actively contested in the courts at the time, precisely because there's no precedent for a president having any legal authority to circumvent regulatory authority and issue a permit that violates environmental protection laws just because he wants to.

There's a reason that the oil companies did not bother to fight the Biden administration's decision to rescind the permit, and simply shut down the project. They knew that the legality of the entire project was highly questionable, and that they had spent money building parts of it with full knowledge that the permits were being challenged in court and could be found invalid, at which point they would have to tear it all down. They baked that risk into their calculations and decided to go forward anyway in hopes of a windfall, and they lost.

Nothing like that is the case for offshore wind farms, to the best of my knowledge. They were permitted through the usual regulatory channels, and there was no plausible reason to expect that such legally issued permits would be illegally rescinded on the whims of a wannabe dictator.

So it's not really the same thing. It's not even close.

IMO, given the amount of money involved and the patent absurdity of the government's behavior, the only rational thing to do is ignore their order and continue to build the wind farm anyway. The government has no legitimate legal right to take back a long-term contractual agreement like that. Once they signed on the dotted line, the lease is valid. Any national security concerns, if legitimate, should have been settled before the government entered into the agreement. Now, it's too late. Tough s**t, Donald Duck.

The government has only one option at that point, and that is to take the wind farm company to court. At that point, keeping the reasoning secret from the judge will not be possible, and the judge will see right through the farce and order them to do what the judge ordered them to do before — live up to their agreements. Realistically, national security concerns are implausible, and more to the point, even if a national security concern does exist, that's the government's problem to figure out how to prevent it from being a national security issue. They have no legal right to coerce a corporation to act on their behalf in doing so, absent a law being passed by Congress, which they have not done.

The only alternative is to waste years in court trying to get a judge to overturn the executive order and then wait for them to file another one in six months, resulting their use of the land being a constant yo-yo. The only rational thing to do, IMO, is to force the government's hand by making it clear that you won't be bullied, and making it clear that every future interaction along similar lines will end the same way — with you continuing to operate under the terms of your existing agreement and the government repeatedly and expensively failing to compel you to do otherwise.

Well... maybe. Tesla premium connectivity costs $10 a month or $99 per year when purchased annually ($8.25 per month equivalent). My Apple watch uses a fraction of the bandwidth by comparison, effectively doing nothing more than receiving text messages and notifications. (Phone calls on a watch replace a cell phone on the same line, so they really don't increase usage.) Yet service for my Apple watch costs $15 a month.

So judging by the state of the market today, buying service from a phone company is likely to cost considerably more than buying it from a car company, assuming the car company doesn't use it as a profit center, because car companies get a deep discount by buying service in bulk for a million cars.

Whether you can safely assume car companies won't use it as a profit center is another question, but there's always mobile hotspots if they do, so there are limits to how much they can overcharge.

Only when something goes wrong, which one hopes occurs with declining frequency. Of course, this past weekend probably cancels out the last five years in one day... :-D

It doesn't necessarily indicate that. You're speculating.

It is entirely possible that Waymo Driver truly doesn't know how to handle that edge case. That's probably not the sort of thing that you'd have a lot of training data for, after all. However, I can think of at least five other possible explanations that are also plausible.

Option 1: In the interest of safety, they required the cars to phone home to report a traffic light down and confirm before proceeding, but because so many lights went down all at once, the remote operator team became overloaded.

Option 2: Some human driver did something significantly unexpected, and the Waymo driver software concluded that a wreck was imminent and assumed that it did something wrong, and stopped rather than risk making the problem worse.

Option 3: Those intersections were in areas that were covered mainly by secondary cellular towers that lacked proper battery backup, and the network went down. The remote operators were unable to take control over the vehicles to get them started again after one of the two possible explanations above.

Option 4: The network went down and the software has a bug that causes it to fail in some unexpected way when the car can't communicate with the central servers.

Option 5: The network went down and the underlying cars have disabling technology that causes them to shut down when they lose communication (for theft prevention reasons). However, I think that one would also require satellites to be unavailable, which is not too likely.

We won't really know for sure until someone at Waymo analyzes the data and publicly provides an explanation for what happened, which probably won't be until next year, because I suspect that a majority of Waymo employees are probably gone for the holidays this week and next. :-)

No, because Waymo cars are actually autonomous and thus don't have a human driver in them. The only thing they can do is phone home and ask a human to plan a path for them or otherwise tell them what to do. And that takes longer than if a human were in the car driving. There's no real-time remote driving (for obvious safety reasons).

It's very relevant. It means that they don't have to care about safety. They can drive through the down light without stopping and trust a human to intervene. They can stop without going and trust a human to intervene.

But more than that, they have the luxury of treating it as a stop sign without worrying about whether the traffic light really is out or just not being detected because of a regression in their image recognition model, because if they are wrong, a human will intervene.

So having a system that is human-supervised relieves Tesla and Cruise of a lot of the responsibility for mistakes and makes them able to enable riskier features.

Apparently you have never driven in San Francisco. These aren't freeways, and these aren't suburbs. They're mostly dense urban streets a la Manhattan. You either have a continuous row of cars parked along one side or both, or you'll have driving lanes. There are no shoulders. If you get very lucky, there might be one spot free, but not enough spots for four cars at a single intersection — probably not at any time, day or night, realistically.

So in situations where stopping in the middle of the street is safer than continuing, then it is, in fact, a safety feature, because there is rarely (if ever) a third choice.

The safe fallback on the freeway is typically to pull over onto the shoulder. Unfortunately, city streets in SF do not have those. If they did, they would have been turned into a parking lane or another driving lane already.

Let's get real here. I live near a light rail line, and that's still a 15-minute walk, plus five minutes at the other end. And the closest light rail stop to my office is about a ten-minute drive on city streets, and it takes twenty or thirty minutes to get there on the light rail. So add it up and yeah, I could walk for 20 minutes, then take public transit for 20 minutes, then take some kind of shuttle for 10 minutes.

Or I can spend fifteen minutes, ten of which are on the freeway, and drive myself. The walking portion of the trip alone exceeds the total length of my commute.

Rail only makes sense if traffic is so bad that cars are completely infeasible. Otherwise, they're the wrong tool for the job.

I'm not the person you're replying to, but for me, it depends on where I'm going. To a scheduled appointment? No. I plan it out. On a long trip? Also, no. To the grocery store? Yes, of course. It's a five-minute drive on city streets. I'm sure not going to spend a minute on Google Maps to see if traffic is bad and I could save one or two minutes by waiting an hour. That wouldn't make sense.

My last shopping trip involved stopping at three different Lowe's and Home Depot stores because the website availability was so detached from reality that they had zero of something that they supposedly had 16 of in one instance. It's not random, but that doesn't mean the plan doesn't have to change at a moment's notice. Doing it by car took an hour. Doing it by public transit would have taken the better part of a day, because the second and third stores would have closed before I could have gotten to them.

So yeah, having to make unexpected changes to your plans is more common than you think.

And for long trips, food stops and restroom stops tend to also be random (though long-distance train systems often have those onboard, making that not a particularly interesting point). On the flip side, for long-distance trains, the interval is usually anywhere from several hours up to a whole day, so if you do have a planned stop for some other reason, it's going to be a long stop, and will usually require a hotel stay.

Sadly, that's not most places with trains. Subways, maybe, but surface trains tend to be more like every 15 minutes or more.

I'm okay with drivers having to pay for cellular communications for updates or pay to have Wi-Fi access within range of their cars. But otherwise, yes, map updates should be free forever. If that means the government has to pay to maintain the servers that provide the map updates, fine.

What does that mean?

Doing something like that would already violate any number of state and federal laws, at least in the United States.

Won't work. Shouldn't try. There are fundamental security reasons why trusting data from outside the car is unsafe.

Assuming LED tail lights (nanoseconds from dark to light), camera latency is one frame (which should ideally be 1/60th of a second, but in practice, is usually closer to 1/30th of a second) plus processing time of maybe another 1/60th of a second, for a total of < 100 ms.

Time to transmit that signal over a radio, do the public key crypto to verify that it was actually sent by a car and isn't noise, perform checksums on the packet, decode it, and figure out what to do is likely to be a decent fraction of that time, because it may require multiple retries to get the packet, you're going to have a nasty collision domain with multiple devices transmitting on the same frequency, etc. You *might* get average latency under 100 ms, but with exponential backoff, worst-case will still be well over 100 ms, realistically, given how many cars would be within the collision domain.

So you're not actually getting the information there noticeably more quickly by using a radio. What you are doing is making it possible for a malicious person to stick a radio transmitter on a bridge or light post and tell every car that passes by that the vehicle in front of them is braking.

This does not improve anything, but adds considerable risk, making it a bad idea.

And no, this is not a solvable problem. There's no way to prevent these transceivers from being in the control of the general public, because they are, by definition, in cars that are owned by the general public.

This one is at least theoretically feasible, as long as every device has a unique public key, and as long as stolen devices are immediately added to a revocation list that gets propagated to every car in the world.

However, unless there are zero human-driven cars (where you would gain a fraction of a second from not having to wait for the green light to light up), there's zero advantage to doing this over just using cameras and stopping at the stop line. If your car is stopping six feet into it, it is almost certainly because a human driver is incompetent. Even Tesla's nowhere-near-fully-self-driving tech doesn't make that mistake.

Mind you, when lights are out and other cars do things that are unsafe, self-driving cars may panic and phone home to ask for help, and that might occur inside the interaction, but that's not the same thing as making their initial stop six feet over the limit line. If there is a limit line visible on the street, self-driving cars stop at it, period. This isn't even hard.

Also, if the light is out, the transponder will also be out, so you'll be in the same situation that you're in now.

There's no advantage to this, either. First, it's a massive security risk. You're trusting data from the lead car to tell you what to do. If it says, "Speed up to 80 MPH," the back car is going to do it even if the car in front is going 20 MPH, and you're going to have an accident. Any hardware that is in the hands of consumers fundamentally cannot be trusted to provide valid data to hardware in the hands of other consumers. Such a trust model cannot reasonably exist in the real world.

Second, you have computational ability in the cars anyway. So the only advantage would theoretically be conserving the CPU utilization required for driving the car safely, and because you cannot trust that the data coming in is valid, you cannot reasonably stop doing what is necessary to avoid causing a wreck.

Even things like obstacle detection can't be trusted until you observe the obstacle. However, you can observe the car in front of you dodging an obstacle and take similar action. But you don't need communication to do that — only cameras. And unlike what you're proposing, cameras work without the car in front of you having to have special hardware, making them a far superior vehicle-to-vehicle communication system.

Therefore, again, this is useless technology that provides zero benefit over not having it.

The only possible advantage any of this expensive complexity could have would be enabling braking for cars that lack any other driving assistance, but even that would not fully work until the last non-transmitting car dies in 30+ years. Meanwhile, putting camera-based driver assistance in the car will fully work immediately.

This is the sort of thing that only makes sense to companies that sell overpriced electronics to car companies, plus the government bureaucrats who have fallen victim to their lobbying efforts. From a trust and security perspective, inter-vehicle communication makes zero sense. It is fundamentally impossible for such technology to provide actionable data without verifying that data using cameras, and if you already have the data from the cameras, you don't need a second source for the same data, which makes it completely and utterly useless. Thus, it provides no meaningful benefit over camera-based technology, but adds significant expense to *every* car, all so that the cheapest camera-free cars can be only slightly cheaper than they would be if they put cameras and a computer in them instead. It's false economy.

So sure, if you want every car to cost an extra thousand bucks for no reason, push for these silly laws. As for me, I'd rather have technology that actually does something useful, rather than padding the pockets of vehicle OEM component manufacturers.

It's post-prop-13 California. We're lucky there are paved roads.

Various news sites also reported that some people observed Waymo cars treating them as 4-way stops. So it probably isn't as simple as not being programmed to be able to fall back, but rather some combination of multiple factors, including the nonfunctional lights, that in combination spooked the cars.

Either way, though, requiring human verification of an outage once per traffic light is probably the right thing to do, if only to ensure that the non-detection isn't a bug. Multiply times a lot of cars at a lot of lights.

Tesla FSD beta relies on a human operator in the car. I don't know what it does when a light is out. It either treats it as a red light or as a green light. In the former, then it relies on the human driver to take over to get it going again. If the latter, then it relies on the human driver to avoid a fatal collision. Either way, it relies on a human driver in the car.

Tesla's robotaxis also have a human safety driver. And still reportedly crash 12.5x more often than human drivers. So my guess would be that they treat disabled lights like green lights and hope for the best. :-D