"the robot cars will not be unleashed without human oversight. Regulators are requiring that a backup test driver remain in the driver’s seat to take over if necessary. Zoox is also not allowed to charge passengers, keeping the prospect of a profitable business model elusive."
There is another permit for fully driverless operations, which California introduced in April this year. Waymo got their fully driverless operators permit in late October, and to my knowledge they're the only company to even apply for it.
Waymo is the only driverless car company that I even remotely trust.
I went to a conference where Uber was presenting, two months before the incident where they killed a pedestrian. Their entire presentation focused on providing a public service to improve safety. It was so over the top on feel-good-make-the-world-a-better-safe-place, you couldn't believe a word of it. If they really cared about safety, they would be adding safety features to cars already operated by humans (e.g., automatic breaking, warnings when you veer out of your lane, cameras, etc). They wouldn't try to take the giant step of removing the human.
Their presentation was such BS, that it was clear they couldn't be trusted. Then two months later, it was confirmed.
Everybody knows Uber is a shitshow. I'm not sure what it is that turned you off Zoox. Sebastien Thrun, who started the Google self-driving car project considered Jesse Levinson to be his brightest student, and Jesse is now CEO of Zoox.
The thing about Waymo is that they haven't made any real progress in about 3 years. I mean, they've rolled out hundreds of new vehicles in several cities, they've racked up 10 million real world test miles and 8 billion simulated test miles, and have discovered thousands of new edge cases, but they still can't reliably navigate certain everyday driving scenarios such as left turns at uncontrolled intersections.
Behavior modelling isn't sufficient to read deeply enough into the matrix of dynamic, open world driving environments for self driving cars to perform gracefully beyond a certain threshold of xomplexity. The ML knowledge of how to deal with these kinds of problems doesn't exist, that is, unless somebody has an ace up their sleeve.
If anybody has an ace up their sleeve, it is likely to be either Aurora or Zoox. Though it's more likely that in a few years time we'll have a half dozen companies all bumping up against the same glass ceiling Waymo has been trying to break through for years now.
>bumping up against the same glass ceiling Waymo has been trying to break through for years now
I'd be happy to be proven wrong but I strongly suspect that over the next couple of years it's going to become apparent that tests and proof of concepts in a combination of limited and easy conditions (e.g. Chandler AZ) and other environments with a human backup are a very long way from scaling and generalizing to selling a truly self-driving car to the public.
It's also possible we'll see assistive driving systems for certain conditions (e.g. selected freeways) that don't require human attention and that full autonomy will be relegated to a maybe someday future. From the perspective of the car manufacturers, full highway autonomy actually seems like a big win even if it's not as sexy as autopilot under all conditions.
If they could solve two scenarios I would be content: highways and congested city traffic (where cars mostly sit and advance a few meters at a time). Especially the congested city traffic scenario would spare me of a ton of stress. If I don't have to watch the roads I can detach emotionally and find something else to kill time.
Unfortunately, congested city driving is probably one of the toughest cases. Once you can drive through Manhattan or Boston you can probably handle just about anyplace with paved roads in the US [ADDED: assuming a sufficient degree of mapping and testing in those other areas].
Cruise has been operating in SF for a while. They seem to be doing quite well. I’ve never seen them make an unsafe or questionable move. I even did a live safety test of one, jumped off the curb at a speed and direction that would have taken me in front of the car if I’d continued. It slowed significantly, so it could stop if I had continued in front (which I did not). The safety driver was like “wtf car, what are you doing?” till he saw me and then he was kinda pissed.
The Uber cars were a shit show. They didn’t know how to yield to pedestrians in the crosswalk. I’m glad the ca dmv revoked their registrations.
"It was so over the top on feel-good-make-the-world-a-better-safe-place, you couldn't believe a word of it. If they really cared about safety, they would be adding safety features to cars already operated by humans (e.g., automatic breaking, warnings when you veer out of your lane, cameras, etc). They wouldn't try to take the giant step of removing the human."
Zoox presentation in the NeurIPS Intelligent Transportation Systems workshop was similarly cringeworthy. First half was them going over the numbers of auto fatalities worldwide and how they should be eliminated, followed by very little information on how they would even come close to human level performance on the task.
but hasn't it been shown that other drivers are more comfortable if they see a driver? I keep remembering back to Total Recall's Johnny Cab. Surely it will be legislated that other than not having an obvious driver autonomous cars should have some visible identification that they are indeed autonomous.
article points out that 62 companies are licensed to test self-driving cars in California currently. (testing is not the same as offering autonomous rides.)
so, if i judge by analogy to what's happened in the human-driven space (e.g. Uber, Lyft, and ? i don't even know who else), i'll go out on a limb and say that roughly 60 of these 62 will eventually just go away.
The list includes most of the world's major automakers, a bunch of tier 1s, several major chipmakers, and Apple. Things would have to get pretty bad for all but 2 to go away.
They will all license from the winner of the autonomous race. Regulation will come at them fast as failure modes get discovered. Wouldn't surprise me if they'd get faa level of certification required within the following two decade. That will strangle newcomers and stragglers, cementing the working solution in the market
>faa level of certification required within the following two decade
If that's the case (and I'm not saying it isn't), you then probably need maintenance and update processes on par with commercial aviation. And that probably implies certified organizations renting/leasing vehicles only. And, at that point, you're probably cheaper just to have a minimum wage human driver.
Mainstream self-driving implies that people can buy cars at a similar price point that they can today and maintain them in the same loosey goosey way that many people also do today.
Well yes, it is certainly going to be a retirement, a even just a dirty sensor is going to be a showstopper issue or a security risk, so I imagine those vehicles are going to need significant maintenance
> i'll go out on a limb and say that roughly 60 of these 62 will eventually just go away.
Don't think you have to go out on a limb to say that. In the early 20th century there were hundreds of car manufacturers in the US, but only three really survived. The low chances of success are already pretty well understood.
well the reason there’s only Uber and Lyft is because there are significant barriers to entry into ridesharing at this point. Yes, anyone could launch an Uber-like app fairly easily, but the barrier to entry is the supply of drivers. You wouldn’t be able to onboard enough fast enough to create a service with low ETAs that would be competitive with either in a mature market unless you had a ton of capital to incentivize them to drive for you with bonuses.
With self driving you don’t necessarily have a supply issue assuming you can deploy as many vehicles as you need in a market to achieve ETAs on par with or lower than Uber/Lyft. It’s all going to come down to ETAs, price, and marketing. I don’t think there will be 60 services in a single market but I can see more than the 2 today operating successfully.
I think the barrier to entry in unlicensed taxis is that there are two large companies currently committed to losing money doing it.
Attracting drivers may be hard, but it's not zero sum; drivers can be and are active in Uber and Lyft simultaneously, it's not inconceivable to swap one of those for a different app or add a third. But it would need to be compelling in some way, preferably a not so monetary way, because a race to the bottom doesn't usually benefit a small company.
