Right? And “capable of” doesn’t mean it can do it all the time… In fact, Musk recently tweeted that he is using the beta version and he has been able to drive to work with “almost no interventions“.
While insurers and Thatcham support the introduction of more automation – and believe fully automated cars would be safer than human drivers – they said the current technology was a “quantum leap” away from what was needed.
The automated system would potentially need to return control to a human driver within three seconds to avert high-speed collisions – but insurers’ research found it takes 15 seconds for the driver to be sufficiently engaged to react to avoid a hazard, roughly 500 metres distance on a motorway.
Cars with ALKS as currently configured will not be able to change lanes, nor react to the red X signal on a smart motorway gantry indicating that a lane is closed ahead. Thatcham also believes they could fail to detect debris or even people on the road in an emergency situation.
“The idea of automation is something that insurers wholeheartedly support. But you can’t have steps towards automation – either the car is driving or it isn’t. We can’t have drivers sitting there watching Netflix, supposedly ready to take over.”
Here is a video of Cruise self-driving in San Francisco:
I enjoyed that, thanks.
I had a conversation like this earlier this month with some of my co-workers. The cars may be able to handle 98% of all the road conditions and hazards, but they won’t be able to cope with the person who is driving any of the other vehicles on the road.
Optimistically, I believe that a 100% reliable self driving car could be put on the market in 5 years. This would mean every researcher and developer would have to get it right the first time. 10 years is a more reasonable estimate. Still people and technology can surprise you…
Tesla has released its beta full self driving mode, and it is neither full self driving nor free of some scary defects.
Yeah, I found it curious that they had recently introduced the ability to recognize traffic cones. Ugh, that’s not even close to an edge case, and they’re just now adding it? Still don’t see how you’re going to hit this goal without something approaching strong AI.
The truth is, we don’t know if we can build a car that is truly self-driving in any random environment until and if we develop an AI with general intelligence. It’s one thing to allow self-driving on specific roads under specific conditions, and quite another to say you can just tell the car to go anywhere in North America or Europe and trust that the car can get you there safely.
We have yet to find out what happens when micreants decide to spoof the vehicles, say by blowing tin foil into the road to confuse the car’s sensors, or people game the cars by forcing them out of a lane when they want in, etc.
Also, in aviation you learn all about situational awareness, and how long it takes to ‘get your head in the game’ in an emergency situation. If car manufacturers are expecting to handle 95% of events automatically, but flip control to the driver in the other 5%, they are going to be very surprised.
Yes, and average drivers are not trained pilots. Plus there will always be a certain percentage of bad drivers.
Some drivers will be watching a movie on their phones, or turned round to chat with someone on the back seat, or half asleep, when the car suddenly beeps for their attention.
From the article I quoted above:
insurers’ research found it takes 15 seconds for the driver to be sufficiently engaged to react to avoid a hazard, roughly 500 metres distance on a motorway.
If you have systems to try to force the driver to always pay attention, then what’s the point of a self-driving car in the first place?
There seems to be a growing trend of people driving while recording themselves talking to their phone. Maybe they’re driving semi-automatic cars but otherwise it seems stupidly dangerous.
Uber is getting out of the self-driving car business (CNN article, New York Times article. They’re essentially giving the business to a start-up named Aurora.
"Self-driving cars without human drivers as a back-up are now being tested in China’s Shenzhen city, according to AutoX, a Chinese autonomous car technology firm.
The Alibaba-backed company said Thursday that this is the first time a “completely autonomous fleet,” that has no accompanying human drivers or remote operators, is on the roads in China."
My limited understanding of Chinese traffic is that it’s rather chaotic to put it charitably.
I wonder if self-driving might actually work better in an environment of total chaos rather than the US / Canada / japan / European condition of 99%(?) sensible / controlled / anticipatable and 1% bat-shit loony.
The company to which Uber gave its business is planning on concentrating on self-driving trucks, because automating highway travel is easier than trying to automate in-town driving.
It’s not surprising that this is happening in Shenzhen, which is modern and grid-like. And while it still puts them behind Waymo, which I think is now fully open to the public, they’re following the same general blueprint of closed ridership in a geofenced area & a goal of eventual open access.
What’s not clear to me is the whether they will still have remote monitoring like Waymo does, to get vehicles out of sticky situations. I assume so, but the articles don’t say.
That’s an interesting approach. Waymo has the opposite approach of staying in-town, because when a mistake is made (by the human or machine) the speeds are lower, and therefore mistakes are less dangerous.
I’m not certain of the calculus that lead them to pick a 40 ton truck traveling at 65 MPH over a 2 ton car traveling at 35 MPH. It seems to me that the truck is going to be a less forgiving test bed to make mistakes in.
Might be. OTOH …
If the significant error opportunity exists once every 10 seconds in the city and once every 10 hours on the highway you’ve got 3600:1 leverage there.
In a country whose cities are chock-full of pedestrians and bicyclists, but whose highways are all modern, it seems like a lot of the expensive high-consequence crashes are in town.
I thought it might be time to follow up on Bo’s suggestion here to see where these companies stand on their self driving projects, since 2020 is wrapping up.
Now, I don’t remember why I picked a BusinessInsider “article” to start this discussion, because BI is pretty much clickbait trash, but let’s see what we can do with it.
Tesla is aiming to have its driverless technology ready by 2018.
2018 came and went, obviously. Earlier this year, Musk was still saying “by the end of 2020.” It seems like he’s always saying full autonomous driving is 6-12 months away, so take this for what you will.
Google has suggested it’s working on having the technology ready by 2020.
Waymo has not been one to make broad predictions. As I understand it, Waymo One is operating in a small area of Phoenix. I feel like that’s probably a win for Google having fulfilled it’s own expectation of where they could be by 2020, but it’s not exactly ubiquitous.
Toyota is looking to have a driverless car ready to go by 2020.
Toyota is still investing heavily in autonomous driving AI, but is still squarely in the R&D phase, without any promises of a commercial product on any particular timeline. I’m not sure where BI got this prediction so I can’t quantify its accuracy.
BMW will introduce a fully self-driving car in 2021.
The “iNext” that the article mentions is now the iX, an electric vehicle still slated to be released next year.
The iX was supposed to have SAE level 3 autonomous driving, even as of earlier this year, but that now seems to not be the case. Actual specifics on what it will have are scarce, but it seems that it will be whatever BMW’s current crop of level 1 and 2 driver aids are.
By 2020, Volvo plans to be well on its way to having a fully autonomous car.
Can’t really quantify this one, but Volvo is planning on selling cars starting in 2022 with LIDAR, but the cars will not be equipped with autonomous software at that time.
Nissan is committed to have a commercially viable autonomous car on the roads by 2020.
Nissan’s “ProPilot” branding is a fairly standard package of driver aids. I can’t find any promises that this will eventually become a fully autonomous package at any point in the future. I’m not sure if they were actually promising that back in 2017 or if BI made it up.
Ford is aiming to have its fully autonomous car ready in four to five years.
Ford is planning on launching a commercial service with autonomous Escapes in 2022, which would put it on the high end of the “4 to 5 years” prediction. I guess we’ll have to come back in 2 years on that one.
General Motors will have a fleet of self-driving cars available for employee use in late 2016.
Back in October, Cruise said that they were going to test self driving cars in San Francisco “this year.” I can’t find anything about whether or not those tests happened. Given the covid situation, it seems unlikely. It also seems strange that they thought it was going to happen. Regardless, they’re still clearly in the R&D phase.
Daimler, the maker of the Mercedes-Benz, plans to have its driverless trucks ready by 2020.
As of October, Daimler has partnered up with Waymo on this effort. To me, that says that they basically abandoned whatever in-house automation efforts they have, and they appear to be supplying the “truck” while they hope Waymo can supply the “self-driving.”
The Audi A7 drove 550 miles by itself in 2015, but there’s no word about when it’s hitting the market.
As of this year, Audi is “giving up” on level 3 autonomous systems in their production vehicles.
Baidu, a Beijing-based search company, is aiming to have a commercial model of its driverless car ready by 2018.
Baidu is now predicting 2025 for a commercial offering.
Honda is aiming to have fully autonomous cars on the road in 2020.
Like Toyota, Honda has not been in the predictions business, so I’m not sure where BI got this “aim,” but Honda is also still in the R&D phase, focusing more on their autonomous driver aid package than a fully autonomous vehicle.
Hyundai is aiming to have driverless features in its cars by 2020, but won’t have a fully autonomous car ready until 2030.
Perhaps the most realistic assessment.
There are 4 or 5 more predictions in that article but they’re even more speculative than the ones I listed, so I don’t think we need to address them.
I won’t rehash the news of AI/Autonomous driving difficulty that’s been posted in this thread over the last 3 years, but what seems to have changed most in that time period is peoples’ expectations. When Tesla first started rolling out AutoPilot, there was, I think, and expectation that people would be able to hop in their own personal vehicle and snooze on the way to work (i.e., level 4 or 5 autonomous vehicles), “in the near future.” While there are still a lot of predictions for level 4 or 5 vehicles “in the near future,” the expectation is now that they will be limited to taxis, most likely in geofenced areas, or long haul trucking routes.
I think the idea that people will be able to walk out to their garage and get into a robot car is slowly waning. That’s going to make this thread increasingly irrelevant, as the line between fantasy and reality gets more and more messy. That is to say, if Ford successfully launches a taxi service in 2023 in certain markets, and they truly are magic robot cars, albeit geofenced and with operators able to take over remotely in emergencies, how does that apply to my OP? Even I don’t know.
The CEO of Volkswagen expects them to be available between 2025-30. “Volkswagen Group CEO Herbert Diess expects autonomous vehicles to be ready for sale between 2025 and 2030. Diess cited the improving performance of computer chips needed in autonomous cars while developments in artificial intelligence are also speeding the process.”