Support the Arctic Sea Ice Forum and Blog

Author Topic: Transport and self driving vehicles  (Read 89004 times)

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Transport and self driving vehicles
« on: April 05, 2021, 12:42:15 PM »
After the challenge to vox_mundi about whether a post on purely self driving should be in the Tesla thread, I also wondered whether it belongs in the EV thread either.

Talking purely about self driving tech doesn't really fit with either because self driving is not exclusively in the EV domain, witness both Uber and Waymo vehicles are not EV.

However self driving does fit within the overall sphere of climate change and the impact on the climate and also policy and solutions.  Especially policy.  Self driving has the potential to radically reduce the number of vehicles on the road by providing cheap transport, where people need it, when people need it and at a cost which people will be willing to bear.  In fact, correctly costed, people may consider the price of insurance, maintenance and outright purchase cost against the number of times they actually really need to use a personal vehicle.  Especially with the growing prevalence of home delivery for so many articles.

So I thought that in my response to vox_mundi I'd start a new thread where we can throw in all the stuff about self driving which does not relate specifically to Tesla or other EV's.  Of course there will still be FSD stuff for Tesla because it is impossible to separate glory/failure from FSD with it being one of the pillars that Tesla is relying on to catapult the company into a position from which it will not fail.

To the response.

... There's every reason to think Waymo's competitors will face this same dilemma as they move toward large-scale commercial deployments. By now, a number of companies have developed self-driving cars that can handle most situations correctly most of the time. But building a car that can go millions of miles without a significant mistake is hard. And proving it is even harder.

Yes you need tens of billions of miles of driving with evidence that the Artificial driver is as good or better than the average human.  There is only one company doing this and, as it is proving on a daily basis, simulation is exactly that.  Simulation.  Not reality and reality trumps everything.

My take on this is "just how good is the average human anyway"?  For which I went looking for articles and there has been a study done on this.  However the question is much more nuanced than that.

The actual question should be "Just how good do humans think they are at driving and how will it be possible to compare real world AI driving against human perception".

Because as the linked study below shows, people are pretty crap at judging their own capabilities and some people will be scared by the firm driving stance of AI and others will be frustrated by how conservative it is.

https://www.researchgate.net/publication/331424175_Safer_than_the_average_human_driver_who_is_less_safe_than_me_Examining_a_popular_safety_benchmark_for_self-driving_cars

In the abstract is a clear statement about what is needed and there are two glaring statements in terms of perception.

Quote
Although the level of safety required before drivers will accept self-driving cars is not clear, the criterion of being safer than a human driver has become pervasive in the discourse on vehicle automation.  This criterion actually means “safer than the average human driver,”


Quote
At the level of individual risk assessment, a body of research has shown that most drivers perceive themselves to be safer than the average driver

Quote
Since most drivers believe they are better than average drivers, the benchmark of achieving automation that is safer than a human driver (on average) may not represent acceptably safe performance of self-driving cars for most drivers.

And there is the problem.  How do you actually categorise the Average human driver, independently, without self perception.

There was a study done on reaction times of drivers to situations.  They had differing reaction times depending on whether the incident was simple or complex and expected or unexpected.

This was real world testing done with test equipment in vehicles.

https://www.degruyter.com/document/doi/10.1515/eng-2020-0004/html

If you look at the tables, this is the average total response time.  So recognising the incident to moving leg and pressing brake pedal.



Now this is real world so, to some degree, the reactions will be slower.  However I have also been tested.  I was in the UK Army and in a drivers unit, we transported tanks around Germany.  ADAT came in with a testing rig, you sat in a seat behind a wheel and with your right foot on the accelerator and your left foot on the floor.  You watched a set of lights and reacted when they went red.

Reaction time was measured from when both, brake and clutch pedals had broken the sensors on them.

If you do a bit of digging, the actual average brain response time to visual stimulus is 250ms. So those leg responses, on top of average visual responses, would be around correct, you would think.  541ms being fast as an average, after all it only leaves just over 250ms to move the leg and brake after recognising the situation.  Closer to a second, as an average, is someone who is going to be struggling at over 70mph.

So my results?  210ms total response time.  To recognise the lights, move both my legs and depress both clutch and brake pedals.  The ADAT guy said I don't need to worry about what is in front of me, I need to worry about what is behind me because they'll run into the back of me if I have to emergency stop.  The unit was 650 drivers, average was about average for this table above.

OK so let's ignore me as a person and just look at this with a range of drivers and also put that result into the Complex/Unexpected category.  Results showed that Complex Unexpected is about twice as long as simple expected.  So 420ms in this case.  Which is still under the lowest bracket for this table.

Just how do you get a "perceived" AI which is better than the "average" human driver.  Especially when human drivers with response times in the 1,000 ms consider themselves as well over "the average".  Just consider how a human driver with 1,000 ms response time would react to a Tesla, with microsecond response times, driving positively down the road. They'd be scared out of their wits.  Then consider someone in the "fast" bracket.  Let us say 400ms to 500ms.  They're going to be bored out of their skulls and think that the AI is overcautious and wasting time.

Somehow we're going to have to get over these perceptions or we are never going to get acceptance and if we cant get acceptance we'll never get regulatory approval.

Whilst I think Elon is correct about stats and showing the number of people killed by human poor driving, as opposed to AI driving, I think there is another dimension.  I think there needs to be a very extensive study on comparing people's personal perception of their driving skills to their actual performance.

I think people will get a real shock when they understand what their actual driving capability is.  When you get your license/permit to drive, you don't get a % capability in driving mark.  You get a pass or fail.  That automatically makes drivers assume that if they passed they must be better than the median?  Surely?

In fact nothing could be further than the truth.  Gaining your license/permit just means you passed the absolute minimum BASE requirements in order to control your vehicle on the road.  You know absolutely nothing about how well you will drive in stressing situations or how you will react when things go wrong.

We are then told that these people must be able to "judge" whether AI is good enough or not.
Being right too soon is socially unacceptable.

Robert A. Heinlein

greylib

  • Frazil ice
  • Posts: 168
    • View Profile
  • Liked: 85
  • Likes Given: 184
Re: Transport and self driving vehicles
« Reply #1 on: April 06, 2021, 02:04:21 PM »
I don't drive - I haven't been behind a wheel for over forty years now, having realised during lessons that my reflexes are completely wrong.

This makes me an ideal candidate for FSD technology. Some people aren't cut out to be passengers. I see them wincing, pressing imaginary pedals, covering their eyes... I'm much more relaxed to the idea of sitting back and reading while a piece of computerised technology carries me to wherever I need to be.

Twenty years ago my wife and I had a bet. I said that before I reached eighty, I'd ride in an autonomous vehicle. Seven more years to go, and I live in hopes of a win.
Step by step, moment by moment
We live through another day.

vox_mundi

  • Multi-year ice
  • Posts: 10165
    • View Profile
  • Liked: 3510
  • Likes Given: 745
Re: Transport and self driving vehicles
« Reply #2 on: April 12, 2021, 07:25:25 PM »
Thanks for opening this thread, NeilT.

Your study on the average driver brought to mind the closing words of the monologue on Garrison Keillor's Lake Wobegon  ... "Well, that's the news from Lake Wobegon, where all the women are strong, all the men are good-looking, and all the children are above average."

https://en.m.wikipedia.org/wiki/Lake_Wobegon

------------------------------------------------------

Here's another self-driving entrant ...

----------------------------------------------

Intel’s Mobileye Will Launch a Fully Driverless Delivery Service In 2023
https://www.theverge.com/2021/4/12/22375759/intel-mobileye-udelv-autonomous-vehicle-delivery-2023



Mobileye, the company that specializes in chips for vision-based autonomous vehicles, announced that it will launch a full-scale, fully driverless delivery service starting in 2023. The company, a subsidiary of Intel, is joining forces with self-driving delivery startup Udelv to run this new service.

Deliveries will be made using a new type of cabin-less vehicle called The Transporter. While manufacturing plans are still in flux, Mobileye and Udelv say they will produce 35,000 Transporters between 2023–2028 — a signal of their seriousness to launch a driverless delivery system at scale.

The vehicle will be compatible with DC fast charging and will have a top speed of 65 mph. A spokesperson said that Udelv will announce its manufacturing plans for the Transporter at a later date.

Mobileye’s turn-key self-driving system features a full-sensor suite of 13 cameras, three long-range LiDARs, six short-range LiDARs, and six radar. It also includes the Israeli company’s EyeQ system-on-a-chip and a data crowdsourcing program called the Road Experience Management, or REM, which uses real-time data from Mobileye-equipped vehicles to build out a global 3D map.

While the Transporter will be fully autonomous, it won’t be completely alone in the wilderness. Udelv says the vehicle will have ultra-low-latency teleoperation capabilities, meaning a remote operator can offer a prompt to the vehicle if it gets tripped up.

The company is also testing autonomous vehicles in a variety of cities around the world for the eventual launch of a robotaxi service and has said it would bring its technology to personally owned consumer vehicles by 2025 as well.

Udelv is an interesting choice for a partner for Mobileye. One of the few AV startups that has yet to be acquired by a larger company, Udelv has been testing autonomous delivery vans in a variety of markets across the US over the last few years, including Oklahoma City, Arizona, and the Bay Area in California. Udelv said it has completed 20,000 deliveries for merchants in the cities where it has operations.

----------------------------------------------------------

GM's Cruise to Operate All-Electric Driverless Cars In Dubai
https://techxplore.com/news/2021-04-gm-cruise-all-electric-driverless-cars.html

Dubai said Monday that U.S. self-driving car company Cruise will become the first to operate autonomous taxis and ride-hailing services in the United Arab Emirates city of more than three million people.

It is the first venture outside of the U.S. for General Motors-owned Cruise, which has been operating electric vehicles in San Francisco for years, although it just recently began testing without the safety of a backup driver in the car. Cruise plans to begin offering a fully autonomous ride-hailing service there but it has not said when that will begin.

The partnership, Dubai officials said, aims to reduce traffic accidents, cut pollution and save money while converting 25% of all trips in the city to driverless transport by 2030.

As part of the agreement, Cruise will establish a Dubai-based company that will be responsible for the fleet of roomy vehicles, which have four passenger seats and no visible driver's seat or controls.

The fleet is expected to start operating in 2023 in limited numbers, before scaling up to as many as 4,000 vehicles by 2030, Al Tayer said.



-------------------------------------------

Your Next Pizza From Domino's Could Be Delivered By a Robot
https://www.forbes.com/sites/alanohnsman/2021/04/12/pepperoni-for-the-robot-dominos-nuro-test-autonomous-pizza-delivery/
https://www.restaurantbusinessonline.com/technology/dominos-putting-its-delivery-robot-work



The chain and autonomous vehicle maker Nuro will start testing the R2 vehicle at a location in Houston this week.

Customers can choose to have their pizza delivered by Nuro’s R2 robot—Nuro’s first autonomous, occupantless on-road vehicle with regulatory approval from the U.S. Department of Transportation.

... “This program will allow us to better understand how customers respond to the deliveries, how they interact with the robot and how it affects store operations.”

For the pizza chain, Nuro could provide a solution to perpetual challenges in finding drivers, a challenge that has only intensified in recent months. Nuro has also recently received some attention from other companies—Chipotle Mexican Grill invested in the company last month.
« Last Edit: April 12, 2021, 09:22:31 PM by vox_mundi »
“There are three classes of people: those who see. Those who see when they are shown. Those who do not see.” ― anonymous

Insensible before the wave so soon released by callous fate. Affected most, they understand the least, and understanding, when it comes, invariably arrives too late

Sigmetnow

  • Multi-year ice
  • Posts: 25763
    • View Profile
  • Liked: 1153
  • Likes Given: 430
Re: Transport and self driving vehicles
« Reply #3 on: April 13, 2021, 10:06:34 PM »
Quote
Mobileye’s turn-key self-driving system features a full-sensor suite of 13 cameras, three long-range LiDARs, six short-range LiDARs, and six radar. It also includes the Israeli company’s EyeQ system-on-a-chip and a data crowdsourcing program called the Road Experience Management, or REM, which uses real-time data from Mobileye-equipped vehicles to build out a global 3D map.

Removing the crutches:  Tesla is transitioning to pure vision as its self-driving method, figuring that the remaining vision software problems must be solved, regardless of any other sensors on the vehicle.

Elon Musk explains Tesla’s pure vision approach to Autopilot and Full Self-Driving 
Quote
“When radar and vision disagree, which one do you believe? Vision has much more precision, so better to double down on vision than do sensor fusion. Sensors are a bitstream and cameras have several orders of magnitude more bits/sec than radar (or lidar). Radar must meaningfully increase signal/noise of bitstream to be worth complexity of integrating it. As vision processing gets better, it just leaves radar far behind,” Musk explained. ...
https://www.teslarati.com/tesla-autopilot-fsd-camera-approach-explained-elon-musk/
People who say it cannot be done should not interrupt those who are doing it.

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #4 on: April 13, 2021, 10:20:59 PM »
I was looking at the article on the future of Waymo with the new co-CEO's.  I was interested in one of the pictures.

It is clear that the Waymo vehicle is using extensive pre mapped data to project a very clear picture of where it is.  It recognises certain types of vehicles and obstacles but, really, doesn't see very much at all.

There has already been one cross map of FSD images on the road video.

Also the article talks about AI and the need for deep AI skills then it says.

Quote
The AI technology is not ready, and despite the lidar, radar, and other sensor technologies used to complement deep learning models, self-driving cars still can’t handle unknown conditions in the same way as humans do. They can run thousands of miles without making errors, but they might suddenly make very dumb and dangerous mistakes when they face corner cases, such as an overturned truck on the highway or a fire truck parked at the wrong angle.

https://bdtechtalks.com/2021/04/08/waymo-ceo-reshuffling-self-driving-car-industry/
Being right too soon is socially unacceptable.

Robert A. Heinlein

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #5 on: April 13, 2021, 10:23:45 PM »
And, of course, new entrants know who the competition is.

Quote
Huawei’s autonomous-driving technology has already beaten Tesla’s in some domains by enabling cars to independently cruise for over 1,000 kilometers (621 miles), as per Rotating Chairman Eric Xu.

https://finance.yahoo.com/news/huawei-targets-over-1b-self-161517027.html
Being right too soon is socially unacceptable.

Robert A. Heinlein

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #6 on: April 13, 2021, 10:35:18 PM »
Meanwhile NVIDIA have unveiled their Drive Atlan chip concept, due in 2025 and with 1,000 tops.

https://electronics360.globalspec.com/article/16583/nvidia-unveils-new-ai-processor-for-self-driving-cars

Nobody is talking about how much power it will draw but with a data throughput of 400mbits/s and the ability to host dozens of input devices, I doubt it will be low.

Also Nvidia is delving into heavy duty AI training including self drive AI training.

https://www.forbes.com/sites/davealtavilla/2021/04/12/nvidia-unveils-powerful-grace-ai-supercomputing-cpu-and-atlan-secure-autonomous-vehicle-platform/?sh=6310eefc5d15

Being right too soon is socially unacceptable.

Robert A. Heinlein

vox_mundi

  • Multi-year ice
  • Posts: 10165
    • View Profile
  • Liked: 3510
  • Likes Given: 745
Re: Transport and self driving vehicles
« Reply #7 on: April 14, 2021, 12:27:33 AM »
Israelis May Ban High Tech Cars From Bases: ‘Perfect Espionage Vector’
https://breakingdefense.com/2021/04/israelis-may-ban-teslas-other-high-tech-cars-from-military-bases-the-perfect-espionage-vector/

TEL AVIV: Everyone who spends time around the military or embassies knows about tight restrictions on cell phones, smart watches, and cameras.

Well, there’s a new threat in town. The Israeli Defense Forces (IDF) are evaluating the danger from connected cars with cameras and global positioning systems (GPS) and may ban their entry into its bases. Senior IDF officers get leased cars and use them to drive to work. Many Israel Air Force officers drive near advanced systems, such as US-made fighters and other aircraft.

Several experts here say that civilians work in many IDF bases, adding to the potential threat.

... “Most connected cars have cameras and audio recording capabilities that can be activated remotely,” Evans said. “This is by design. Sometimes within the confines of a base, people will discuss classified things that they shouldn’t be discussing outside of a classified facility, such as a Sensitive Compartmented Information Facility. When they see other people, they are more careful. But cars are generally viewed as non-threats. Add to that the fact that you don’t really even need to compromise the car, but just the owner’s car account or cell phone, and it creates literally the perfect espionage vector.”

Elon Musk’s car company Tesla began selling its cars in Israel three weeks ago, and the demand is high. “The advanced cars are equipped with cameras and GPS systems, and they can be hacked and the data can therefore go to the enemy,” one security expert told Breaking Defense.

“Some of these vehicles boast 4k or at least HD camera capabilities,” Evans notes. “That is not just a fancy name for a resolution. You can take a seemingly harmless photo of a building 1,000 feet away, but at 4k, zoom in really close with the resulting video or picture, and you can see secrets not meant to be seen from outside the building. 4k cameras are much, much better than our eyes at spying for this reason.”

As for GPS, Evans said, “This one is a little more simple, but knowing the GPS location of a military base is not hard information to find. But the GPS location of a specific secret building on that base is a different story. Imagine you work at a nuclear research part of a base. Its location is secret. But you have a smart car. Through other espionage activities, I found out you work there. I hack your phone or your car’s online account. I track your location as you go to work every day. Now I know the specific GPS location of your work facility. It goes downhill quickly from there.”

... An attacker will look for the weakest link in the chain, so instead of targeting a hardened vehicle, an attacker may target the backend or mobile app instead. Consequently, besides securing the connected car itself, it is imperative that automakers also employ approaches to secure their backend systems and relevant mobile apps.”
“There are three classes of people: those who see. Those who see when they are shown. Those who do not see.” ― anonymous

Insensible before the wave so soon released by callous fate. Affected most, they understand the least, and understanding, when it comes, invariably arrives too late

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #8 on: April 14, 2021, 01:19:46 PM »
Eventually bases may need underground car parks on the periphery of the base with shared transport (driven by a person), within the base.  Possibly the equivalent of golf carts for more senior personnel.

We live in a connected and media rich world.  It is a fact, the military need to assess the risk and then cope with it.  Even a pager, which can receive codes looking like phone numbers, is a significant risk and pagers work inside buildings and at some depth in the ground.
Being right too soon is socially unacceptable.

Robert A. Heinlein

Sigmetnow

  • Multi-year ice
  • Posts: 25763
    • View Profile
  • Liked: 1153
  • Likes Given: 430
Re: Transport and self driving vehicles
« Reply #9 on: April 14, 2021, 04:42:44 PM »
Quote
The AI technology is not ready, and despite the lidar, radar, and other sensor technologies used to complement deep learning models, self-driving cars still can’t handle unknown conditions in the same way as humans do. They can run thousands of miles without making errors, but they might suddenly make very dumb and dangerous mistakes when they face corner cases, such as an overturned truck on the highway or a fire truck parked at the wrong angle.

Well, humans make “very dumb and dangerous mistakes” every day while driving — even when not dealing with a corner case. ;)

No system will ever be perfect.  But “safer than the average human” is a good first goal.
People who say it cannot be done should not interrupt those who are doing it.

Sigmetnow

  • Multi-year ice
  • Posts: 25763
    • View Profile
  • Liked: 1153
  • Likes Given: 430
Re: Transport and self driving vehicles
« Reply #10 on: April 14, 2021, 05:42:34 PM »
Tesla started out with cameras, radar and ultrasonics... but is now moving to pure vision.  Years ago Tesla also extolled their new high-definition maps... but later dropped them. 

Xpeng is currently in ‘the more sensors, the merrier’ ;) phase.  (Yet it seems that their autonomous driving is limited to highways.)

Tesla’s Chinese rival Xpeng ups the self-driving game with world’s first mass-produced LiDAR in P5 sedan, defying Elon Musk
Quote
Xpeng Motors has launched the world’s first mass-produced electric car with LiDAR technology, upping the ante against competitors including Tesla by making the self-driving technology a standard feature in its P5 sedan.

The four-door car will be equipped with 32 perception sensors to augment its laser-based radar to detect its surroundings, including pedestrians, cyclists, static objects and road works – crucial in making autonomous driving possible. The car maker, which completed a weeklong, 3,675km autonomous test drive last month from Guangzhou to Beijing on its P7 sedans, will offer drivers what it calls navigation-assisted autonomous driving on highways and enhanced memory parking.
...
The P5, whose pricing is yet to be announced, is Xpeng’s third production model after the G3 sports-utility vehicle and the P7 sedan. Powered by Qualcomm’s Snapdragon flagship SA8155P computing platform, it will be fitted with two LiDAR units, 12 ultrasonic sensors, 5 millimetre-wave radars and 13 high-resolution cameras.
...
During Xpeng’s weeklong driving test last month, the carmaker reported 0.65 time of human intervention for every 100km driven. The success rate for lane changing and overtaking stood at between 86.05 per cent and 97.91 per cent during the six-day period. ... 
https://www.scmp.com/business/companies/article/3129530/teslas-chinese-rival-xpeng-ups-self-driving-game-worlds-first
People who say it cannot be done should not interrupt those who are doing it.

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #11 on: April 15, 2021, 12:24:17 AM »
People who report on any sort of self drive technology should at least know what they are talking about...

https://www.kbb.com/car-news/bluecruise-fords-new-hands-free-driving-system/

Quote
BlueCruise is what the Society of Automotive Engineers (SAE) refers to as a Level 2 driver assistance system. Fully autonomous driving, in which the driver can stop paying attention to the road and trust the car to drive, would be Level 5. There are no driver assistance systems beyond Level 2 for sale in the U.S. today. Other Level 2 systems include GM’s Super Cruise and Tesla’s Full Self-Driving system (a name which has created confusion over just what it can do).

Actually the Tesla Autopilot is the competitor for Blue Cruise and it is on sale, is more competent and has the same requirement of the drive to pay attention.  Although Tesla also requires that the driver actually have their hands on the wheel.

FSD is something else entirely and it is still only beta with no actual release date nor regulatory approval.

Something these reporters should state is that there is currently no legislation for Level 5 for personal vehicles anywhere in the world.  Busses, trucks, small delivery vehicles.  But not for personal vehicles.  Even the Waymo taxi's have a teleoperated backup driver.
Being right too soon is socially unacceptable.

Robert A. Heinlein

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #12 on: April 15, 2021, 12:37:39 AM »
Competition for Dojo.

https://www.streetinsider.com/PRNewswire/dRISK+emerges+from+stealth%2C+presents+breakthrough+performance+in+retraining+autonomous+vehicles+to+detect+high+risk+and+quot%3Bedge+cases+and+quot%3B/18260418.html

Quote
About dRISK

Using dRISK for retraining, Autonomous Vehicles can detect and contend with edge cases 6x sooner and with 2x greater accuracy. dRISK has built a taxonomy of edge cases derived from massive and heterogeneous data focused on high risk -- millions of hours of CCTV footage trained on high-risk intersections, full-text accident reporting, and extensive expert input from both transportation specialists and NASA experts in failure mode analysis. Integration is easy, and data can be delivered exclusively for perception retraining on fully annotated simulated and real-life data, or for full-stack AV risk assessment with hardware in the loop. dRISK's customers include AV developers and the world's largest insurers and transport authorities
Being right too soon is socially unacceptable.

Robert A. Heinlein

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #13 on: April 15, 2021, 12:52:19 AM »
The competition is hotting up for powerful processors.  It took a bit of digging to find the power levels, but here they are.

https://news.in-24.com/business/news/article8674/Autonomous-driving-Nvidia-Drive-Atlan-delivers-twice-the-power-per-watt.html

Quote
As Danny Shapiro, Senior Director of Automotive at Nvidia, breaks down, Atlan plans to have platforms with 400 TOPS at 50 watts, 800 TOPS at 100 watts and the 1,000 TOPS at 150 watts mentioned for the keynote

But Nvidia are not the only competitors.

Quote
Because it was Qualcomm at CES in January 2020 that wanted to beat Nvidia with twice the performance per watt. Qualcomm’s variant with 720 TOPS at 130 watts comes to 5.5 TOPS / W and would clearly beat Orin from Nvidia. At this year’s CES, Qualcomm’s portfolio was broadened, so that it now scales from 10 to over 700 TOPS. The first cars with Snapdragon Ride were originally scheduled to hit the streets in 2023, but Qualcomm has brought the schedule for the first solutions forward to 2022



Being right too soon is socially unacceptable.

Robert A. Heinlein

Sigmetnow

  • Multi-year ice
  • Posts: 25763
    • View Profile
  • Liked: 1153
  • Likes Given: 430
Re: Transport and self driving vehicles
« Reply #14 on: April 15, 2021, 04:01:15 PM »
The importance of AI
Quote

Elon Musk
@elonmusk 4/14/21
Major improvements are being made to the vision stack every week. Beta button hopefully next month.

This is a “march of 9’s” trying to get probability of no injury above 99.999999% of miles for city driving. Production Autopilot is already above that for highway driving.
 https://twitter.com/elonmusk/status/1382458022367162370

Elon Musk:
While perhaps obvious in hindsight, it turns out that the only way to solve self-driving is to solve a big part of real-world AI. That is what Tesla is doing, both in hardware & software.

Unless a company is good at AI with tight compute, it’s hopeless. This is insanely hard.
https://twitter.com/elonmusk/status/1382549229726498821   

But regulations will be the penultimate barrier (behind customer acceptance):
Quote
< Fixing the EU autopilot restrictions needs higher priority. Its putting off a bunch of people from buying a tesla in Europe. Those cars are currently feature-limited.

Elon Musk
4/14/21
Hopefully addressed this summer 
https://twitter.com/elonmusk/status/1382459271774834690
People who say it cannot be done should not interrupt those who are doing it.

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #15 on: April 15, 2021, 05:42:38 PM »
But regulations will be the penultimate barrier (behind customer acceptance):

My take is slightly different.  Regulations will be the Ultimate barrier.  Long after the problem with level 5 is solved, regulations will drag on.  Because it is in the hands of politicians and they don't like to make decisions unless they can guarantee it won't come back and bite them in the ass.

The way the press has been having a feeding frenzy on any single AI driven vehicle that has a problem, the ass munching is a given even if it exceeds expectations.
Being right too soon is socially unacceptable.

Robert A. Heinlein

Tor Bejnar

  • Young ice
  • Posts: 4606
    • View Profile
  • Liked: 879
  • Likes Given: 826
Re: Transport and self driving vehicles
« Reply #16 on: April 15, 2021, 06:04:09 PM »
Competition for Dojo.

https://www.streetinsider.com/PRNewswire/dRISK+emerges+from+stealth%2C+presents+breakthrough+performance+in+retraining+autonomous+vehicles+to+detect+high+risk+and+quot%3Bedge+cases+and+quot%3B/18260418.html

Quote
About dRISK

Using dRISK for retraining, Autonomous Vehicles can detect and contend with edge cases 6x sooner and with 2x greater accuracy. dRISK has built a taxonomy of edge cases derived from massive and heterogeneous data focused on high risk -- millions of hours of CCTV footage trained on high-risk intersections, full-text accident reporting, and extensive expert input from both transportation specialists and NASA experts in failure mode analysis. Integration is easy, and data can be delivered exclusively for perception retraining on fully annotated simulated and real-life data, or for full-stack AV risk assessment with hardware in the loop. dRISK's customers include AV developers and the world's largest insurers and transport authorities

As I know nothing about the specialties involved, here is a suggestion for those who actually know something  :) 
Near high-risk locations, electronically emit the expert data, etc. (a service of the Town or Highway Dept.), so that AI-driven vehicles have this 'extra' support when entering a particular 'fray'.  I've seen Highway advisory radio signs that say something like, "Tune your AM radio to 1570 for updates on traffic conditions ahead."  Alternatively, each AI system, Tesla, Waymo, etc., could communicate with every vehicle (of their manufacturer) approaching a 'known danger zone' with the valuable data needed for safety.  A vehicle that frequents such a location could save it for future reference, just in case the 'transmitter' is down.

I presume Tesla autopilot learns how to drive into the unusual parking spaces a car owner frequents.  A car would have to learn, "We back in at home and park on the right side of the paved apron, with two tires on the brick 'extension' at John's, or else there would be many frustrated owners: "The car drove me from Atlanta to Tallahassee all by itself (275 miles - 440 km), and it can't even remember that I back in. I always back in!"
Arctic ice is healthy for children and other living things because "we cannot negotiate with the melting point of ice"

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #17 on: April 15, 2021, 06:43:54 PM »
The core problem is to get the vehicle to drive as a human would.  Niceties such as remembering your parking space and how you park in it, are refinements which will come in time.  After all, you "know" that.  You will need to "tell" the vehicle what you know.  There is no place for that right now.

For the vehicle data, I'm not sure how much use live streaming data is.  Right now FSD is at the stage where it is learning to deal with immediate situations, just as we do.  Whilst we might like to know about downtown snarls, this kind of information is mainly so we can avoid it.

dRISK is all about trying to de-Risk teaching an AI to drive.  They do this by simulation.  It's a bit like sitting in a classroom and having someone talk to you day in day out about the edge cases and dangerous situations you might find yourself in and how to deal with them.  Except the AI has near perfect recall.

There comes a point, though, where data, for the sake of data, is counter productive.  It just means the AI has to sift through a larger and larger data set and that takes time and processing power.  The more you teach AI about "things" the more it has to remember.

The way we need to teach AI is in terms of how to respond, not millions of dry facts.  Think about a human driver.  Humans do not drive down the road trying to go through every situation to select the "best" response.  Humans watch the road, look out of obstacles, try to avoid oncoming or merging traffic, bikes, cycles and humans/animals.  We do this by experience in the real world and what does and doesn't work.  We discard the does not work and apply a filter to the "does work" which can then be used to solve new and unknown situations in the future.

dRISK is condemned to fail if this is their approach.  AI may be fast, but you can't load it up with every possible edge case in the world.  How would it deal with a huge pothole opening up underneath it?  It would have no clue because the one in a trillion edge case will never be taught to it.  However if you teach the AI how to respond to events and danger signals, it will automatically know what to do, if anything at all can be done.

This is not just about Data per se.  This is about the right data at the right time and the learned experiences which go with it.

Just as Waymo has not progressed in the last 6 years with their level 5 taxi service in geofenced areas, all the others who believe that the 1,000 TOPS of the Nvidia chipset alongside the provision of data centre speed connectivity to storage, will resolve the problem for them by just having a bigger data set and "hoping" that they have got all the edge cases, are going to have a rough time.
Being right too soon is socially unacceptable.

Robert A. Heinlein

Sigmetnow

  • Multi-year ice
  • Posts: 25763
    • View Profile
  • Liked: 1153
  • Likes Given: 430
Re: Transport and self driving vehicles
« Reply #18 on: April 15, 2021, 10:02:02 PM »
Quote
Near high-risk locations, electronically emit the expert data, etc. (a service of the Town or Highway Dept.), so that AI-driven vehicles have this 'extra' support when entering a particular 'fray'.  I've seen Highway advisory radio signs that say something like, "Tune your AM radio to 1570 for updates on traffic conditions ahead."  Alternatively, each AI system, Tesla, Waymo, etc., could communicate with every vehicle (of their manufacturer) approaching a 'known danger zone' with the valuable data needed for safety.  A vehicle that frequents such a location could save it for future reference, just in case the 'transmitter' is down.

Most(?) vehicle trip software apps use current traffic information to suggest routes that avoid heavy traffic or road closures, so “permanent” electronic warnings for select locations are certainly possible.  Teslas are even learning to avoid potholes. :) What owners are looking for next is instant local updates such as debris on the road at a specific location, or a disabled or slow-moving vehicle, which is witnessed by one car and transmitted to other cars in the area.
People who say it cannot be done should not interrupt those who are doing it.

greylib

  • Frazil ice
  • Posts: 168
    • View Profile
  • Liked: 85
  • Likes Given: 184
Re: Transport and self driving vehicles
« Reply #19 on: April 16, 2021, 12:42:30 AM »
Decades back, when autonomous vehicles were being discussed, the common belief was that there would be a single master computer controlling all the traffic in a city. Most commentators quickly realised that the only way this could be accomplished is if ALL vehicles were under control. A single rogue vehicle being controlled by a human driver would, fairly literally, throw a spanner in the works. Most SF stories of the time, therefore, had a system of automatic control - when a vehicle is driven into the city, control is ceded to the computer. Only when it leaves the city can the driver take over.

Pretty much the reverse of the situation right now, where autonomous driving is looking fairly safe on the main highways, but difficult and dangerous in town. Though I'd guess that if all vehicles were under computer control, most of the problems would disappear.
Step by step, moment by moment
We live through another day.

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #20 on: April 16, 2021, 09:40:04 PM »
Pretty much the reverse of the situation right now, where autonomous driving is looking fairly safe on the main highways, but difficult and dangerous in town. Though I'd guess that if all vehicles were under computer control, most of the problems would disappear.

Decades back the authors of such literature could not envision a car with a computer with the power that even the Tesla HW3 has.  It is hundreds of times faster than mainframes with thousands of processors from 1993 and hundreds of thousands of times faster than equivalent mainframes from the 80's.

That was all people knew then.  So they assumed it would take a computer larger than a building to run all those vehicles.

Now we know better.  Each vehicle working like a human is far more efficient than some huge monster computer trying to run everything.
Being right too soon is socially unacceptable.

Robert A. Heinlein

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #21 on: April 16, 2021, 09:43:57 PM »
Tesla is now starting to modify AP to be worse than it is today because...

Tesla limits self-driving in Europe due to new UN/ECE regulations

Quote
The European Commission has ruled that the Tesla company violated Regulation 79 of the United Nations Economic Commission for Europe.

https://www.greendrive-accessories.com/blog/en/tesla-limits-autopilot-in-europe-cause-of-new-un-ece-rules/

The article states that:

Quote
The countries concerned by this limitation are : Austria, Belgium, Bulgaria, Croatia, Denmark, Estonia, Finland, France, Germany, Gibraltar, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Monaco, the Netherlands, Poland, Portugal, Romania, Russia, Slovakia, Spain, Sweden, Switzerland, Turkey and the United Kingdom. Consequently Tesla is obliged to adapt its Autopilot system.

[Correction].  GB is in the list.  Missed it.

Ah well.  This is how it is going to go.  Laws are going to be made for systems which are 10 years, or more, behind current tech.  Then they will limit current tech to their understanding.

Such is life.
Being right too soon is socially unacceptable.

Robert A. Heinlein

gerontocrat

  • Multi-year ice
  • Posts: 20384
    • View Profile
  • Liked: 5289
  • Likes Given: 69
Re: Transport and self driving vehicles
« Reply #22 on: April 16, 2021, 11:27:45 PM »
Tesla is now starting to modify AP to be worse than it is today because...

Tesla limits self-driving in Europe due to new UN/ECE regulations

The European Commission has ruled that the Tesla company violated Regulation 79 of the United Nations Economic Commission for Europe.

https://www.greendrive-accessories.com/blog/en/tesla-limits-autopilot-in-europe-cause-of-new-un-ece-rules/

Ah well.  This is how it is going to go.  Laws are going to be made for systems which are 10 years, or more, behind current tech.  Then they will limit current tech to their understanding.

Such is life.
A long time ago in the UK a man with a red flag had to walk in front of a Horseless Carriage.

Tradition rules , OK?
"Para a Causa do Povo a Luta Continua!"
"And that's all I'm going to say about that". Forrest Gump
"Damn, I wanted to see what happened next" (Epitaph)

Sigmetnow

  • Multi-year ice
  • Posts: 25763
    • View Profile
  • Liked: 1153
  • Likes Given: 430
Re: Transport and self driving vehicles
« Reply #23 on: April 18, 2021, 08:00:15 PM »
Tesla is now starting to modify AP to be worse than it is today because...

Tesla limits self-driving in Europe due to new UN/ECE regulations

Quote
The European Commission has ruled that the Tesla company violated Regulation 79 of the United Nations Economic Commission for Europe.

https://www.greendrive-accessories.com/blog/en/tesla-limits-autopilot-in-europe-cause-of-new-un-ece-rules/
...

That article is from December 2020.  Something may happen in a few months?
Quote
4/14/21
< fixing the EU autopilot restrictions needs higher priority. Its putting off a bunch of people from buying a tesla in Europe. Those cars are currently feature-limited.

Elon Musk
Hopefully addressed this summer
https://twitter.com/elonmusk/status/1382459271774834690
People who say it cannot be done should not interrupt those who are doing it.

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #24 on: April 19, 2021, 12:17:55 PM »
Unfortunately you don't just "fix" EU decisions.  They may be able to modify the UN decision but that is also a huge undertaking.
Being right too soon is socially unacceptable.

Robert A. Heinlein

Sigmetnow

  • Multi-year ice
  • Posts: 25763
    • View Profile
  • Liked: 1153
  • Likes Given: 430
Re: Transport and self driving vehicles
« Reply #25 on: April 21, 2021, 08:31:44 PM »
What’s the Difference Between Ford BlueCruise, GM Super Cruise and Tesla Autopilot? 
https://www.insidehook.com/article/vehicles/whats-difference-ford-bluecruise-super-cruise-tesla-autopilot

—-
Walmart Invests in Cruise, the GM-Backed Autonomous-Car Startup 
Quote
The retailer is preparing for a future when shoppers don’t have to drive to the store and is trying out a number of different technologies, including drones. The cost of getting deliveries to customers’ doorsteps across the nation is often prohibitive, and it’s a big reason why Walmart’s domestic e-commerce business is still losing money. ...
https://www.bloomberg.com/news/articles/2021-04-15/walmart-invests-in-cruise-the-gm-backed-autonomous-car-startup

—-
China Welcomes Autonomous Vehicles on the Country's Roads 
Quote
Baidu's autonomous driving division, Apollo, has said it received the rights to deploy a commercial self-driving bus program in western China. The self-driving bus will be launched in the Chongqing metropolis with a population of over 30 million. Baidu has previously conducted a limited non-profit pilot in this area. Now, the commercial bus will operate a limited 6.2-mile route. Passengers must book trips through one of Baidu's many transportation apps.
...
Autonomous driving in China has been strongly encouraged in recent years. For example, Beijing announced that more than half of new cars sold in 2025 will have autonomous driving capability. ...
https://www.tesmanian.com/blogs/tesmanian-blog/china-welcomes-autonomous-vehicles-on-the-countrys-roads
People who say it cannot be done should not interrupt those who are doing it.

Sigmetnow

  • Multi-year ice
  • Posts: 25763
    • View Profile
  • Liked: 1153
  • Likes Given: 430
Re: Transport and self driving vehicles
« Reply #26 on: May 02, 2021, 05:49:52 PM »
Super Cruise's Biggest Problem Is Us
Over 380 miles on I-95, Super Cruise drove for 346. It also canceled 25 times, in the process revealing the biggest hurdle for autonomy: unpredictable humans.
By Ezra Dyer Apr 17, 2021
https://www.caranddriver.com/features/columns/a36118292/super-cruises-biggest-problem-is-us/
People who say it cannot be done should not interrupt those who are doing it.

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #27 on: May 03, 2021, 12:06:16 AM »
The biggest problem I have with the post is this.

Quote
Champions of autonomy, people who think we’re 10 years away from Level 5 driverless cars, should take a ride in a Super Cruise Caddy

Super Cruise is a level 2 autonomy system. It was designedd to be level 2 and it operates as level 2.

People who want to know what Autonomous Driving is all about need to drive an Autonomous system designed to be Level 5.  Only then will they truly understand Autonomous driving.

Being driven by a smarter idiot does not make you aware of the technology which is being designed to be able to gain a driving permit.
Being right too soon is socially unacceptable.

Robert A. Heinlein

Sigmetnow

  • Multi-year ice
  • Posts: 25763
    • View Profile
  • Liked: 1153
  • Likes Given: 430
Re: Transport and self driving vehicles
« Reply #28 on: May 03, 2021, 08:59:15 PM »
Ford Mach-E Co-Pilot360 driver monitoring system needs an update ASAP
https://www.teslarati.com/tesla-autopilot-ford-mach-e-copilot360-test-video/
Hard to tell when the system engages/disengages.  And the system does not care if you unfasten your seatbelt.

—-
Quote
It’s comical to watch my [Tesla] (with supervision) navigate to and from my job and witness other drivers yawning, eating, using their phones, swerving, and being generally distracted.
Then consider the intense media focus on Autopilot, which greatly reduces traffic accidents
  —
In the name of clicks, the media is willing to slow or stop progress on life saving tech. Meanwhile, more people will die on the roads today and no one will care unless it’s a Tesla. It’s quite disturbed.
https://twitter.com/28delayslater/status/1389171924329172992

—-
Quote
Elon Musk:
A major part of real-world AI has to be solved to make unsupervised, generalized full self-driving work, as the entire road system is designed for biological neural nets with optical imagers
 https://twitter.com/elonmusk/status/1387901003664699392

Lex Fridman:
Driving is a real-world Turing Test.
 
< A Touring Test, if you will.
People who say it cannot be done should not interrupt those who are doing it.

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #29 on: May 04, 2021, 03:06:11 PM »
Lex Fridman:
Driving is a real-world Turing Test.

Not quite I hope. The Turing test was to determine, by question and response, whether the personality at the other end was capable of both comprehension and able to respond to the questions in such a way that the human posing the questions believes they are communicating with a human.

Or in short, for the machine intelligence to be indistinguishable from the human.

In the world of self driving I would hope that the machine would be more vigilant, more cautious, more compliant and, overall, a better driver than most humans.

In other words distinguishable by achievement.

Somewhat better than "Simon Says" or "Monkey see monkey do".  Which is by far the largest portion of self driving AI today.
Being right too soon is socially unacceptable.

Robert A. Heinlein

Jim Hunt

  • First-year ice
  • Posts: 6268
  • Don't Vote NatC or PopCon, Save Lives!
    • View Profile
    • The Arctic sea ice Great White Con
  • Liked: 893
  • Likes Given: 87
Re: Transport and self driving vehicles
« Reply #30 on: May 07, 2021, 10:42:08 AM »
This doesn't seem to have been reported in here yet, so please see:

https://www.gov.uk/government/news/government-paves-the-way-for-self-driving-vehicles-on-uk-roads

Quote
Motorists could see self-driving vehicles on British roads for the first time later this year, the Department for Transport has announced today (28 April 2021).

Following a landmark call for evidence, the government has set out how vehicles fitted with Automated Lane Keeping System (ALKS) technology could legally be defined as self-driving, as long as they receive GB type approval and that there is no evidence to challenge the vehicle’s ability to self-drive.

Designed for use on a motorway in slow traffic, ALKS enables a vehicle to drive itself in a single lane, while maintaining the ability to easily and safely return control to the driver when required.

The technology could improve road safety by reducing human error, which contributes to over 85% of accidents. The driver will be able to hand control over to the vehicle, which will constantly monitor speed and keep a safe distance from other cars.

Today’s announcement comes as a consultation on The Highway Code rules is launched to ensure the first wave of this technology is used safely and responsibly. This consultation will conclude on 28 May 2021.

etc.
"The most revolutionary thing one can do always is to proclaim loudly what is happening" - Rosa Luxemburg

crandles

  • Young ice
  • Posts: 3379
    • View Profile
  • Liked: 239
  • Likes Given: 81
Re: Transport and self driving vehicles
« Reply #31 on: May 07, 2021, 01:37:30 PM »
Lex Fridman:
Driving is a real-world Turing Test.

Not quite I hope. The Turing test was to determine, by question and response, whether the personality at the other end was capable of both comprehension and able to respond to the questions in such a way that the human posing the questions believes they are communicating with a human.

Or in short, for the machine intelligence to be indistinguishable from the human.

In the world of self driving I would hope that the machine would be more vigilant, more cautious, more compliant and, overall, a better driver than most humans.

In other words distinguishable by achievement.

Somewhat better than "Simon Says" or "Monkey see monkey do".  Which is by far the largest portion of self driving AI today.

Just having superhuman calculating ability does not mean the computer has general intelligence.

So the question is when does something have general intelligence?

The turing test is one way to say it has general intelligence. There could be other ways but it is hard to define what they are so the turing test has become *the* test for general intelligence.

Self driving is a difficult test to pass and requires a bit more general intelligence than many specific tasks, but it isn't fully general it is just a specific task no different to superhuman calculating ability or chess or go ability.

Better than ability would be preferred to indistinguishable for a test for general intelligence.

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #32 on: May 07, 2021, 03:54:22 PM »

Self driving is a difficult test to pass and requires a bit more general intelligence than many specific tasks, but it isn't fully general it is just a specific task no different to superhuman calculating ability or chess or go ability.

There is a bit of a difference here.  The rules for Chess and Go do not vary.  They are fixed and well defined.  Strategy within the rules is what is needed.

For driving, the rules change from country to country and  from state to state.   Rules are not fixed and are constantly in flux.

Also whilst Chess and Go have other players, driving has any number of other participants and it is not a given that they will obey the rules.  Then there are environmental considerations.  Non road users using the road, debris on the road, whether from a vehicle or from the roadside, weather, temporary changes in both rules, laws and the condition of the road, such as roadworks and human traffic control (whether for roadworks, failed traffic lights or simply to manage a massively congested situation).

All of which must be known, understood and reacted to.

In short you need to give the AI driver the 17 years of human experiences in reacting to situations which may, or may not be, related to what you are doing at the time; then teach that AI to drive.

This is several orders of magnitude more difficult than teaching a computer to play chess.

A Turing test would expose the computer Chess player in short order.  Because the Turing test is designed to assess general levels of intelligence in an artificial intelligence.  It is these general levels of intelligence which we use to drive.  We layer driving experiences on top of that general experience.

This is why people who's general level of intelligence or mobility are denied a driving permit.  Because they can't reach the required base level in order to layer driving over the top.

This, to me, is the fundamental barrier which most AI driving solutions are failing at.  They started at the wrong level.  They tried to add driving skills to a subnormal intellect.  Not understanding that the base level normal intellect is a pre requisite for being able to drive.  It provides recognition skills, reaction skills and comprehension skills which are all critical core requirements to be able to drive and control a vehicle in the noisy world of human traffic.
Being right too soon is socially unacceptable.

Robert A. Heinlein

Sigmetnow

  • Multi-year ice
  • Posts: 25763
    • View Profile
  • Liked: 1153
  • Likes Given: 430
Re: Transport and self driving vehicles
« Reply #33 on: May 07, 2021, 09:51:38 PM »
Volkswagen says it can profitably sell a self-driving system for €7 an hour
Quote
Volkswagen’s Klaus Zellmer says that the automaker is planning to build every vehicle it has with all the bells and whistles, and customers can then “lease” each feature from VW, not locking them into a massive purchase or extensive program with a bulky initial cost.
..
Zellmer says that Volkswagen can charge as little as €7 an hour for a Level 4 Autonomous drive mode. Compared to a train ticket for a destination that is several hours away, this would make the Level 4 Autonomous vehicle a more affordable option.

“Our cost modeling says if we charge €7 an hour for Level 4 autonomous drive mode, this is a profitable business case,” Zellmer added.

“Think about getting from London to say, Southampton on the train,” Zellmer said. “How much is the train ticket? It’s certainly more than €7 an hour.” It ultimately comes down to the cost-effectiveness of allowing drivers to choose which features they’d like and only charging them for how much they use them. ...
https://www.teslarati.com/volkswagen-profitable-self-driving-system/
People who say it cannot be done should not interrupt those who are doing it.

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #34 on: May 07, 2021, 11:12:18 PM »
At the speed of the VW architecture, it will have run over the policeman, holding his hand up, before it even recognises there is a problem... ;D ;D ;D
Being right too soon is socially unacceptable.

Robert A. Heinlein

vox_mundi

  • Multi-year ice
  • Posts: 10165
    • View Profile
  • Liked: 3510
  • Likes Given: 745
Re: Transport and self driving vehicles
« Reply #35 on: May 07, 2021, 11:32:54 PM »


Honda 1AutoX autonomous taxi



Honda and AutoX have been operating a fully autonomous, no safety driver taxi service in China for a couple of months now.
“There are three classes of people: those who see. Those who see when they are shown. Those who do not see.” ― anonymous

Insensible before the wave so soon released by callous fate. Affected most, they understand the least, and understanding, when it comes, invariably arrives too late

vox_mundi

  • Multi-year ice
  • Posts: 10165
    • View Profile
  • Liked: 3510
  • Likes Given: 745
Re: Transport and self driving vehicles
« Reply #36 on: May 08, 2021, 02:09:38 AM »
Why Hasn’t Waymo Expanded Its Driverless Service? Here’s a Theory
https://arstechnica.com/cars/2021/05/why-hasnt-waymo-expanded-its-driverless-service-heres-my-theory/

Last October, Waymo did something remarkable: the company launched a fully driverless commercial taxi service called Waymo One. Customers in a 50-square-mile corner of suburban Phoenix can now use their smartphones to hail a Chrysler Pacifica minivan with no one in the driver's seat.

And then... nothing. Seven months later, Waymo has neither expanded the footprint of the Phoenix service nor has it announced a timeline for launching in a second city.

If Waymo is confident its technology is safe to operate in Chandler, Arizona, why wouldn't Waymo self-driving be equally safe in neighboring suburbs like Gilbert or Mesa? Those cities have very similar weather, road conditions, and traffic patterns.

Maybe Waymo One just isn’t profitable?

Waymo's self-driving technology isn't actually that automated. If a driverless vehicle gets confused, the safe thing to do is often to slow down and contact Waymo headquarters for assistance. Waymo has a staff of fleet operators that oversees the cars and provides guidance—such as confirming that the car has chosen a safe route through a construction zone

It's also possible that more mundane factors could be driving up costs. For example, with no drivers to clean up messes, Waymo needs a cleaning staff in its garages. Waymo also has a call center to respond when a customer pushes a button inside the car to call for help.

A less obvious but potentially significant expense was pointed out to me by Joel Johnson, a Phoenix-area college student who has taken dozens of Waymo rides and documented them on YouTube. Johnson has observed that Waymo's driverless Pacificas are frequently tailed by a Waymo roadside-assistance van. If a Pacifica gets stuck, a van can quickly arrive at the scene to fix the problem.

Johnson believes that in the early months of driverless operation, Waymo had a roadside-assistance van following every Waymo One vehicle, which would be a significant expense.

Suburban Phoenix is a terrible place to run a taxi service

Things start to make more sense if you think about Waymo One from the demand side. In other words: what are people supposed to use Waymo's taxi service for?

Waymo One is centered in Chandler, Arizona, which is extremely suburban. Most households there own a car. Most stores, offices, and homes have ample parking. As a result, most people in Waymo's service area don't need a taxi very often.

An infrequently used ride-hail service is likely to be unprofitable. A ride-hail network needs a minimum density of vehicles on the road to give customers a reasonable pickup time.

Indeed, this seems to be a general principle: the areas that are most profitable for a ride-hailing service are also likely to be the most challenging for self-driving software to navigate safely. Ride-hailing services tend to be the most profitable in dense areas with an army of potential customers and scant parking. But those areas also tend to have crowded and chaotic streets.

The more complex and unpredictable an environment is, the more likely self-driving software is to get frozen with indecision.

... And perhaps Waymo has calculated that expanding its Phoenix footprint or launching in Dallas or Las Vegas will be more of the same: the services not only won't be profitable, they also won't help Waymo make much progress toward mastering the most difficult environments.

And that may be why Waymo has increasingly shifted its testing efforts to San Francisco. In recent months, I've had multiple people in San Francisco tell me they've seen a lot more Waymo cars on the streets.

... Waymo's leadership has belatedly realized that Kyle Vogt—Cruise's founder, then-CEO, and now-CTO—had a point: that the fastest way to develop self-driving software that works everywhere is to focus on the hardest driving environments rather than the easiest ones. That's worthwhile not only because the software will improve faster but also because the software needs to work well in challenging environments in order for a driverless taxi service to achieve profitability.

This appears in other industries too SpaceX has been very focused on MVP type strategies that also tackle some of the hardest stuff upfront than expand from there. With F9 they went right for real cargo to orbit, rather than engaging in tons of suborbital work. As they pushed up capabilities they then could do landing. And for Starship, they're doing the most difficult part from the very get-go. The upper stage is what will have to deal with the most exotic return profile with aerodynamic control, orbital entry heating, bellyflop/flip up, and precise landing at the end. The Super Heavy booster in contrast isn't quite "just a bigger F9 1st stage" but it's got a lot more in common with that. It goes up suborbital and comes straight back down. If they'd done that easier piece first it'd be cool, but they might get trapped in a local minima if the harder piece than needed changes that disrupted what they'd already done (or else waste a bunch of time/money).
“There are three classes of people: those who see. Those who see when they are shown. Those who do not see.” ― anonymous

Insensible before the wave so soon released by callous fate. Affected most, they understand the least, and understanding, when it comes, invariably arrives too late

vox_mundi

  • Multi-year ice
  • Posts: 10165
    • View Profile
  • Liked: 3510
  • Likes Given: 745
Re: Transport and self driving vehicles
« Reply #37 on: May 15, 2021, 07:55:30 AM »
One of Waymo's Autonomous Taxis Got Stumped by Traffic Cones and Fled Its Fleshy Overlords
https://gizmodo.com/one-of-waymos-autonomous-taxis-got-stumped-by-traffic-c-1846900332

A comedy of errors unfolded after one of Waymo’s self-driving vans became confused by a construction zone, stalled in front of oncoming traffic while figuring out its next move, and then fled from Waymo’s Roadside Assistance team. And a Youtuber captured all of it on video.



About 12 minutes into the video, the car attempts to turn right onto a multilane road, but stalls after it sees the rightmost lane is closed for construction and blocked off by traffic cones. While the car’s internal AI system, aka Waymo Driver, ponders what to do next, a Waymo representative contacts Johnson and informs him that Roadside Assistance is en route to his location. While they’re still on their way, (16:45) the car peels out to complete the turn but promptly comes to a stop between lanes. A moment later, it backs up a few feet only to stop again, blocking the entire lane of traffic this time.

Right around this point in the video, a construction worker arrives and starts collecting the offending traffic cones that left Waymo Driver stumped. A Roadside Assistance rep pulls up behind the car not long after, but before they can enter the vehicle, it drives away.

After briefly fleeing its fleshy overlords, Waymo Driver comes to a stop again, this time long enough for Roadside Assistance to get in and take the wheel. The human driver completed the ride and Waymo waived the fee, Johnson said.
“There are three classes of people: those who see. Those who see when they are shown. Those who do not see.” ― anonymous

Insensible before the wave so soon released by callous fate. Affected most, they understand the least, and understanding, when it comes, invariably arrives too late

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #38 on: May 15, 2021, 01:12:45 PM »
Nice to know the competition works so well...

Didn't see it in the news though. Hardly a surprise there, Waymo isn't a threat to anyone.
Being right too soon is socially unacceptable.

Robert A. Heinlein

vox_mundi

  • Multi-year ice
  • Posts: 10165
    • View Profile
  • Liked: 3510
  • Likes Given: 745
Re: Transport and self driving vehicles
« Reply #39 on: May 21, 2021, 12:15:49 AM »
Cool Tech, Crazy Turns: A Reporter's Take On Driverless Cars
https://techxplore.com/news/2021-05-cool-tech-crazy-driverless-cars.html

The annoyed shopper paced around and knocked on the windows of a minivan blocking him from leaving his Costco parking spot. He didn't seem to notice, or care, that there was no one inside.

A colleague and I had called for the Waymo ride—our first in a fully driverless vehicle—and quickly encountered a hiccup: figuring out how to tell it to meet us at the curb.

We ended up spotting the minivan across the bustling parking lot, and hurried over. As we pulled away, the shopper raised his arm and extended his middle finger.

... During our rides, the minivans slowed for speed bumps and carried out a textbook right-on-red turn. Most impressive was a careful maneuver at a green light where a woman with a walker stood dangerously close to the corner.

But customers in crowded parking lots might find it hard to pinpoint pickup locations without drivers who can call, text or simply watch for them.

A Waymo minivan also made an aggressive turn at a green light that we would have never taken. With several oncoming cars zipping toward us, the minivan darted across the road into a parking lot. Though we didn't come close to crashing, the turn frightened us.

Waymo later examined the maneuver, saying the cameras and remote-sensing technology in a dome atop the minivan had detected the oncoming cars, knew their speed and understood that the vehicle could make the turn safely. [... they would probably brake in time]

Another failed to go the requested location, dropping us off about a four-minute walk away. Panigrahi believes the minivan might have been rerouted due to traffic or a road closure in the area. [... or it was just having a fuck-you attitude]

From there, we tried to catch a ride back to Costco, but Waymo canceled four trip requests and stopped accepting my credit card as a fraud prevention measure, because several requests had been made in a matter of minutes.

... The company said it is listening closely to customer feedback and acknowledges it needs to improve passenger pickups. It also said it's working to set the proper [... low] expectations with riders, and has launched a campaign that provides tips.

The program serves hundreds of passengers each week and offers prices in line with Uber and Lyft. The company runs 300 to 400 vehicles in Arizona for its ride-hailing service and testing.

... 400 vehicles to service ~800 customers per week? 1-2 customers per vehicle per week? $10-20 per ride. It's going to take awhile to cover expenses

“There are three classes of people: those who see. Those who see when they are shown. Those who do not see.” ― anonymous

Insensible before the wave so soon released by callous fate. Affected most, they understand the least, and understanding, when it comes, invariably arrives too late

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #40 on: May 22, 2021, 10:34:30 AM »
This is quite interesting because Tesla has been gaining huge volumes of data from smart summon with Musk stating that navigating mall parking lots as being one of the most challenging things it had to do.

By getting it out there early, FSD is able to do things others can't when it is finally approved and ships.

The press seems to act as if the Tesla FSD team are a bunch of amateurs.  They are not, the challenge is just way bigger than was originally appreciated.

Everyone underestimated the difference between simple highway lane following and actually driving in a city with people, objects, animals and ad hoc traffic systems.

The Waymo approach is if the environment doesn't match the map, go somewhere else. The Tesla approach is teach the AI to drive.
Being right too soon is socially unacceptable.

Robert A. Heinlein

Sigmetnow

  • Multi-year ice
  • Posts: 25763
    • View Profile
  • Liked: 1153
  • Likes Given: 430
Re: Transport and self driving vehicles
« Reply #41 on: May 22, 2021, 04:18:08 PM »
—- What does a Tesla see when a person “bear-walks” in front of the car?
Two short clips.
Quote
Jeremy Judkins
Seeing if I got the update that allows Tesla to identify dogs and other animals.
https://twitter.com/jeremyjudkins/status/1394671712852135937
5/18/21, 11:10 AM
Person bear-walks in front of Tesla. Display... :o

 I wanna give credit to this post for inspiration. They actually got it to work. ...
 https://twitter.com/jeremyjudkins/status/1394676060911022093
“Oh my God!  It F*ing did!”
People who say it cannot be done should not interrupt those who are doing it.

vox_mundi

  • Multi-year ice
  • Posts: 10165
    • View Profile
  • Liked: 3510
  • Likes Given: 745
Re: Transport and self driving vehicles
« Reply #42 on: May 22, 2021, 08:11:45 PM »
Ford CEO Says US Needs to Regulate Automated Driving Systems
https://techxplore.com/news/2021-05-ford-ceo-automated.html

Ford CEO Jim Farley, the CEO of America's second-largest auto company is calling for the federal government to set standards for fully or partially automated vehicles to tighten the safety of electronic driving systems.

... "Absolutely," Farley said when asked whether federal regulations are needed. "Today, the regulations are state-by-state," he said of fully autonomous vehicles. "They're really oriented toward the development of the technology, not large-scale deployment of the technology."

He suggested that legislators and the National Highway Traffic Safety Administration were moving too slowly.

"Time is of the essence," Farley said, explaining that Argo AI, an autonomous vehicle company in which Ford is a major investor, is moving ahead quickly with technology that will let Ford start an autonomous ride-hailing service.

Argo, which is testing autonomous vehicles with human backup drivers in six U.S. cities, expects to be ready for Ford to offer ride-hailing without human drivers sometime next year. In the Phoenix area, Alphabet Inc.'s Waymo is already offering a limited fully autonomous ride service.

And later this year, Ford will offer "Blue Cruise," its own partially automated highway driving system that, like Tesla's Autopilot, keeps cars centered in their lane and a safe distance behind traffic in front of them. With Blue Cruise, drivers can take their hands off the steering wheel. But unlike Autopilot, they will be monitored by a camera to make sure they are paying attention.

... Farley's stance on regulation is unique in the auto industry, which has generally has favored voluntary guidelines over regulations. No federal regulations specifically govern electronic driving systems, although they do fall under safety standards that cover all vehicles.

The industry's biggest trade association, the Alliance for Automotive Innovation, last month proposed voluntary guidelines for partially automated systems. The government's National Highway Traffic Safety Administration has relied on voluntary cooperation, taking a hands-off approach so as not to discourage life-saving innovations.

In 2015, Volvo Cars CEO Hakan Samuelsson asserted that a patchwork of state rules and the absence of U.S. federal oversight could slow the development and introduction of autonomous vehicles.
“There are three classes of people: those who see. Those who see when they are shown. Those who do not see.” ― anonymous

Insensible before the wave so soon released by callous fate. Affected most, they understand the least, and understanding, when it comes, invariably arrives too late

morganism

  • Nilas ice
  • Posts: 1691
    • View Profile
  • Liked: 215
  • Likes Given: 124
Re: Transport and self driving vehicles
« Reply #43 on: May 22, 2021, 10:27:00 PM »
Now Your Car is a Cybersecurity Risk, Too

https://www.eetimes.com/now-your-car-is-a-cybersecurity-risk-too/

"One interesting data point is where the attacks are directed — the so-called attack vectors. The table on the right clearly show that there are two popular targets. Cloud servers are the entry points for nearly 33% of total cyberattacks, as hackers try to gain access to valuable data that can be used to compromise automotive cybersecurity. Insecure keyless entry or key fobs are frequently exploited to gain access and to steal cars. Mobile apps are in third place with nearly 10% of the cyberattacks.

It is interesting that remote attacks add up to nearly 80% of total cyberattacks while the physical attacks are around 20%."

"In June 2020, two new UN cybersecurity regulations were adopted—the WP.29 cybersecurity regulation. Both regulations are for all types of vehicles. The cybersecurity regulation was updated in March 2021. Deployment will start in a few countries in 2021 and 2022 with significant deployment in 2023 and 2024.

The first regulation focuses on cybersecurity and cybersecurity management systems (CSMS). The latest update of the CSMS document is available at: E/ECE/TRANS/505/Rev.3/Add.151 (unece.org)."

xposted on tesla thread

Sigmetnow

  • Multi-year ice
  • Posts: 25763
    • View Profile
  • Liked: 1153
  • Likes Given: 430
Re: Transport and self driving vehicles
« Reply #44 on: May 25, 2021, 12:04:13 AM »
Quote
Whole Mars Catalog
So the CEO of Steer says that you can’t detect cones with LIDAR and Radar. You need vision for that.
As time goes on, i’m more and more convinced that @karpathy and @elonmusk got it right from the beginning on vision being the fundamental constraint on whether it would work.
https://twitter.com/wholemarsblog/status/1396753409726771205
⬇️ Text image below.

Edit:
Tesla has removed references to radar on its website for the Model 3 and Model Y, moving closer to confirming an all-vision approach for FSD and Autopilot.
Tesla updates website to reflect pure vision, no-radar Autopilot and FSD approach
By Simon Alvarez May 24, 2021
https://www.teslarati.com/tesla-pure-vision-update-configurator/
« Last Edit: May 25, 2021, 12:53:28 AM by Sigmetnow »
People who say it cannot be done should not interrupt those who are doing it.

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #45 on: May 25, 2021, 02:33:57 PM »
The publication of the problem with the veracity of reporting sources and the conflict it causes in the AI should be enough for anyone.  Without an AI of human cognizance, it becomes extremely difficult to work with conflicting data and to make the right choice.

Unless you reduce both radar and lidar to range finding, to back up vision, the gain in vertical integration is not worth it.  You could create a situation where vision always outvotes radar and lidar and only takes radar or lidar data into account when vision does not produce a clear result.  But if vision meets 99.9997% of the driving requirements, the sheer cost of radar and lidar and the minimal lack of input, makes it non viable.
Being right too soon is socially unacceptable.

Robert A. Heinlein

NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #46 on: May 25, 2021, 03:42:47 PM »
Select car leasing has produced a graphic which shows the compute power of the Tesla HW3.



It all looks pretty impressive.  Until you realise that only the top two lines count.

What we are really talking about is giving the processing equivalent of a severely mentally limited person the task of driving our cars.

Of course the AI doesn't have to run the body, control life around the body, or do the 1 million and one thing the human brain has to do.  Which is good, because if it did it would not be intelligent enough to get a permit.

It does, however, when judging the difficulty level of driving and the ability so far achieved, put the magnitude of the task into perspective.

Being right too soon is socially unacceptable.

Robert A. Heinlein

Sigmetnow

  • Multi-year ice
  • Posts: 25763
    • View Profile
  • Liked: 1153
  • Likes Given: 430
Re: Transport and self driving vehicles
« Reply #47 on: May 25, 2021, 08:12:49 PM »
Waymo
Quote
“Wow!! This technology is going to change the world within the next few years!  I quit!” — John Krafcik, Waymo CEO

“Wow, completely agreed! I quit too!” — Waymo CFO

“Yes, amazing! We also all quit!” — Four Top Waymo Executives
https://twitter.com/wholemarsblog/status/1396983574075625472

Waymo eyes raising up to $4 billion as executives' exits prompt questions
Quote
Waymo, born as a moonshot inside the search giant over a decade ago, has long been considered the standard bearer in autonomous driving and last year launched a ride-sharing service with driverless cars in Phoenix. Yet critics have pointed out the company hasn’t delivered on pledges to bring that service to more cities and settled on a commercial model.
...
Since February, Waymo has lost six key executives in rapid succession, including its CEO, CFO and several key lieutenants. ...

Recently, Waymo has poured money into self-driving trucks. In late 2019, Krafcik told reporters in Detroit that trucking might be a faster path to commercialization than taxi services. Last October, Waymo signed an agreement with Daimler AG’s trucking division to begin equipping trucks with its self-driving software. Among the recent departures was Vijaysai Patnaik, an executive who ran Waymo’s trucking product. Patnaik didn’t return a message seeking comment.
https://www.autonews.com/mobility-report/waymo-eyes-raising-4-billion-executives-exits-prompt-questions
People who say it cannot be done should not interrupt those who are doing it.

kinbote

  • New ice
  • Posts: 17
    • View Profile
  • Liked: 14
  • Likes Given: 27
Re: Transport and self driving vehicles
« Reply #48 on: May 25, 2021, 08:36:36 PM »
That graph is not merely terrible, it's a great example of just how terrible bad graphs can be. 72 trillion is roughly 7% of one quadrillion, not half, and the bottom half of these entries would not show at all on a graph scaled to one quadrillion.

The second graph located at:
https://www.selectcarleasing.co.uk/news/article/the-power-of-teslas-on-board-brain
is just as ridiculous. The Hubble Space Telescope value is roughly 1% of the Apollo 11, yet graphed as ~60%.

With graphs this bad and no references or explanation to where or how the values were derived, it's difficult to take any of it seriously.


I plugged the values they provided from their first graph into excel and made this bare-bones bar chart.




NeilT

  • First-year ice
  • Posts: 6275
    • View Profile
  • Liked: 386
  • Likes Given: 22
Re: Transport and self driving vehicles
« Reply #49 on: May 25, 2021, 09:18:22 PM »
I should have scrutinised that more.

However it makes the point more clearly.  Perhaps why the Nvidia Drive AGX Orin reaches 2,000 TOPS.

The question is do you need that for an AI to drive a vehicle?  Time will tell.
Being right too soon is socially unacceptable.

Robert A. Heinlein