Updating this web page starting Feb 10, 2021
Autonomous cars will never exist says Waymo boss, AutoExpress, Jan 8, 2019
https://www.autoexpress.co.uk/car-news/105627/autonomous-cars-will-never-exist-says-waymo-boss
"CEO of driverless vehicle firm Waymo says fully autonomous cars will not be able to cope in all-weather conditions."
"Fully autonomous cars that can drive in any conditions and on any road without human input will never exist, according to the boss of a leading driverless vehicle firm."
"Waymo CEO John Krafcik told the Wall Street Journal’s D.Live conference “autonomy will always have constraints.”
"Krafcik explained how it will not only take decades for cars with advanced levels of self-driving capability to become common on roads, but even when they do they will not be able to drive themselves without any form of “user interaction.”
My Comments
I think autonomous cars will be on the roads around 2030.
Steve Wozniak: No Self-Driving Cars in My Lifetime, ExtremeTech, October 25, 2019
https://www.extremetech.com/extreme/300927-steve-wozniak-no-self-driving-cars-in-my-lifetime
"Apple co-founder Steve Wozniak believes in technology. But that doesn’t extend to believing autonomous driving is happening soon. Wozniak, now 69, says autonomous cars that don’t need a backup driver on board probably won’t happen “in my lifetime.” One culprit: Artificial intelligence probably isn’t intelligent or flexible enough to be better than even the worst drivers."
"But as a long-time user of Teslas, he’s been disappointed by overblown claims for self-driving that Tesla couldn’t back up. “They sucked me in,” Wozniak said. He also pointed to Tesla initially offering free charging, then backing away. While home-charging (240 volts) costs him about one-third the cost of driving a gasoline-fueled car, the surcharges now at Tesla and public charging stations bring EV operating costs closer to what gas costs."
"There are too many unexpected things, it might even be a tire lying in the road in front of you, [it doesn’t know to] steer around it, center my car, or what, and a human would do this easily. I don’t see Tesla possibly doing that. There are so many many navigation system [errors] and cars run on navigation systems."
My Comments
My estimate is there will be autonomous cars by 2030 and AGI within about 50 years from now (~2070).
My perspective on autonomous vehicles is that I am in favor of 4 High-level Autonomy.
On this web page I will be writing about various issues including:
Autonomous vehicle technologies
Road safety: fact and fiction
Cybersecurity
Social effects on people, roads, cities, industries
Societal acceptance
Other autonomous vehicles
Legal context and government regulations
Does AV technology fail gracefully on minor crash or does the car transform into a hurtling ton of metal with no steering wheel or brakes?
Effect of internet or GPS downtime?
Are maps needed/available?
Driving in bad weather?
Change course when traveling allowed or only at start of path. If yes how: voice, touch pad?
Valet parking as a step toward demonstrating the viability of autonomous driving in a relatively safe testing environment.
Why airplanes have pilots and why should self-driving cars differ?
Why buses have drivers and why should self-driving cars differ?
Airplane pilots and bus drivers have licenses that require more training, they drive more people, and the vehicle they drive is more expensive than a car.
Is self-driving car AI narrow or general? In my opinion, it's between narrow and general AI which is why people underestimate when self-driving cars will be safe.
These are my opinions on autonomous vehicles
Levels of Vehicle Automation
National Highway Traffic Safety Administration defines levels of Vehicle Automation
Federal Automated Vehicles Policy
http://www.nhtsa.gov/nhtsa/av/av-policy.html
My Comments
I think it is good that NHTSA defines 5 levels of vehicle automation.
Problems
1. Cars will need to read road signs and understand the hand gestures of police directing traffic. For example, can an automated car understand this road sign? Between the hours of 6:00-10:00 AM only commercial buses that carry at least 10 passengers can use the rightmost lane. The automated vehicle can use that lane at some times and at others it cannot.
2. How do Google cars handle disturbances in road conditions? If a lane has a lot of pot holes can the car swerve to avoid them? How about detecting broken bottles on the road. Can the driveless car see and avoid the glass?
3. The assumption that people use cars only to drive from point 'A' to point 'B' is incorrect. Many people go driving to cruise and enjoy the sights. They also make impromptu driving decisions such as hey there's a place to stop and buy fruits or flowers, let's pick some up. Or look there's a souvenir store or museum let's stop and check it out.
4. Studies showing the dangers of a driver using a phone also show that having a person sitting alongside the driver makes driving safer. This is because the other person provides a second pair of eyes and can notice and point out dangers to the driver. The driverless cars now discussed do not have the ability to use those valuable inputs.
5. The Google self-driving car uses a Velodyne Lidar sensor with 64 lasers to build a 3-D map of the environment. The Velodyne Lidar sensor with 64 lasers can be a health danger to drivers and pedestrians. It could potentially momentarily blind other drivers causing road accidents.
6. The Google self-driving car uses a video camera to detect traffic lights and a Velodyne Lidar sensor. Both the video camera and the Velodyne sensor suffer from substantial degradation by rain, fog, or snow. In low precipitation the automated car can mistake rain or snow for objects, and in high precipitation the car becomes a door stop - it does not work. The solution to use higher power lasers isn't feasible for safety reasons.
Do people need to buy a Google car for sunny days and another car for bad weather days? What happens when a Google car user sets out for a drive on a sunny day and the weather becomes inclement. Are you supposed to leave the car at an arbitrary place when the weather becomes bad?
7. How do the Google car's algorithms work with complex road conditions such as in front of the car a pedestrian runs out to get a rolling ball, and behind the automated car another car is tailgating. If it brakes hard the tailgating car will crash into the self driving car. A human driver may resolve the dilemma by driving up onto the front lawn of a house. Can the Google car come up with the correct solution?
8. If the Google car drives down a narrow street that has parked cars on both sides a human might see that one of the parked cars has people in it and one of them might open the car door in the path of the Google car. Can the Google car detect people in parked cars and know there's a real chance one of them might swing a door open into the driving lane?
Driver's exam
The NHTSA should add to their regulations that automated cars must be able to pass a driver's exam.
The Google and other manufacturers' driverless equipment should be able to take instructions from the test examiner such as: 1) At the next junction, make a right hand turn, 2) Parallel park behind that silver Chevy over there.
In my opinion, cars should gradually progress in levels of automation according to the stages set out by the NHTSA.
A Turing Test For Driverless Cars
The well-known Turing Test evaluates whether an unseen human and computer are indistinguishable. It was intended to test general cognitive ability.
My suggestion is there should be a narrowly defined Turing Test for driving exams with the goal that a driving examiner should not be able to distinguish between the driving ability of a human and a driverless car.
A Driverless Car Turing Test
Using a windshield camera, record a complete driver's exam. The camera needs to be positioned so it doesn't show whether there is a person in the driver's seat sitting alongside the human examiner.
Passing the Turing Test For Driverless Cars
After the test, a separate examiner reviews the video. Without knowing if the driver's exam was taken by a person or a driverless car, the separate examiner decides whether the driving capabilities exhibited were sufficient to pass the driver's test.
If the driverless car passes the driver's exam, it obtains a driver's license and also passes the Turing test for driverless cars.
Passing the Turing test for driverless cars would be a significant AI milestone.
Conclusion
The NHTSA should add to their regulations that automated cars MUST be able to pass a driver's exam.
Driverless vehicles should be able to take instructions from a driving test examiner such as: 1) At the next junction, make a right hand turn, 2) Parallel park behind that silver Eldorado over there.
A driverless vehicle should, similar to a human, be able to understand signs and instructions from traffic officers. For example: Road construction ahead, from 6AM to 3PM use the detour on Lincoln avenue.
And a driverless car should be able to interpret hand signals from a traffic officer to pull over to the side of the road and park.
High-Definition Maps
Autonomous cars need to know their own location.
GPS
Prior Maps
Clever AI Turns a World of Lasers Into Maps for Self-Driving Cars, Wired, July 2016
https://www.wired.com/2016/07/civil-maps-self-driving-car-autonomous-mapping-lidar/
Forget GPS, Civil Maps gives self-driving cars street smarts
"Silicon Valley startup Civil Maps creates detailed street-level maps of cities using vector-based graphics, letting cars determine their position based on distance to known objects."
Whoever Owns the Maps Owns the Future of Self-Driving Cars, Popular Mechanics, July 2016
http://www.popularmechanics.com/cars/a21609/here-maps-future-of-self-driving-cars/
"Inside the battle to own the data that will drive the automotive industry of tomorrow."
Self-driving cars aren’t going to be so great until we make our maps way better, The Verge, August 24
http://www.theverge.com/2016/8/24/12628488/uber-maps-self-driving-cars-egress-problem
Road Safety
How are autonomous vehicles tested?
Are the testing results made public?
Important: One of the main flaws in the autonomous cars safety statistics is the following:
They compare Level 0 – No-Automation driving accidents to Level 4 - Full Self-Driving Automation driving accidents.
That is either intentionally or accidentally the incorrect measurement.
It is unfortunate that Google and other market participants use weakly thought-out arguments and metrics.
The correct comparison is Level 3 - Limited Self-Driving Automation compared to Level 4 - Full Self-Driving Automation.
Are there methods to build safer cars available other than full automation? Safer materials? Wheels?
Prevent driver disturbances (e.g. driver distracted by texting while driving)
Prevent driving when intoxicated
Measure cognitive ability before allowing car to start.
Allow remote driver (say, in Nevada, or Delhi) to drive the car.
Are autonomous car tests realistic?
Done in carefully curated location.
Autonomous cars driving less than 25 or 35mph
Important: Is it safe that autonomous cars will be dependent on connection to the 1) internet, and 2) GPS satellites?
If either connection was disrupted, autonomous cars might not be able to drive at all.
Autonomous cars use two different maps: 1) live map created by its sensors (radar, cameras, lasers) and 2) what is called a "prior" map created somewhat like Google street view.
Tesla plans to have all Teslas on the road contribute to data for the prior maps.
What is the latency, or staleness, of the prior maps? Will they be updated every few hours, or every few weeks or longer?
Miles traveled by autonomous cars statistic. Do these miles count disengagement from autonomous mode as driverless miles?
No one was driving Tesla before fiery crash that killed 2 passengers in Texas, authorities say, CBS News, April 19, 2021
https://www.cbsnews.com/news/tesla-crash-texas-no-one-dirving/ [Sic: 'dirving'].
"Harris County Precinct 4 Constable Mark Herman said authorities are certain that no one was driving the car at the time of the crash."
"They feel very confident just with the positioning of the bodies after the impact that there was no one driving that vehicle," Herman told CBS affiliate KHOU-TV.
My Comments
With self-driving cars who is responsible for a crash, Tesla hardware and software? Did that Tesla car have anyone driving? According to the Texas authorities no one was in the driver's seat.
Will Tesla allow forensic analysis on the Tesla algorithms, or will Tesla claim its software algorithms are a black box filled with secret proprietary lines of code.
MIT Center for Transportation and Logistics, January 21, 2021
https://ctl.mit.edu/news/real-self-driving-revolution-remains-years-away
"The real self-driving revolution remains years away."
"Bryan Reimer, a transportation researcher with the Massachusetts Institute of Technology's AgeLab, said it will be decades before people can buy truly self-driving cars in which humans ride solely as passengers."
My Comments
Conventional estimates on the time until autonomous cars are driving down the road, with passengers, seems to be far different than the actual roll-out schedule of these smart cars.
CNN Business, January 21, 2021
https://edition.cnn.com/2021/01/21/success/self-driving-car-technology-2021/index.html
"This year, new technologies will enable more drivers to take their hands off the wheel while on the road. But that doesn't mean their cars will be fully self-driving -- that day still remains far in the future."
"For the time being, these systems will only be used on limited-access divided highways with on-ramps and off-ramps. On these roadways, there are no pedestrians, bicyclists, or double parked trucks. Vehicles with this technology will be able to drive at relatively high speeds, but only in simple traffic situations."
"Drivers still need to pay attention, however. A camera above the speedometer and tachometer makes sure the driver is looking at the road at all times. Or, at least, almost all the time. If the driver looks away from the road for more than a few seconds, the system will stop working."
My Comments
Car and driver for awhile yet travelling down the open roads together: person and machine. "If the driver looks away from the road for more than a few seconds, the system will stop working." Instills a zero confidence level of trust. Why is there so much hype about self-driving cars?
IEEE Spectrum, 6 Key Connectivity Requirements of Autonomous Driving
"While the future of the automotive industry is still unwritten, here is what we know: Industry leaders will need to master connectivity to deliver the V2X (vehicle-to-everything) capabilities fully autonomous driving promises."
My Comments
I am in favor of level four high automation. However, level five full automation for cars is probably an unachievable goal. A small exception may be autonomous trucks travelling on highways.
I do not understand why Google a search engine company got involved with autonomous cars.
See: https://sites.google.com/site/myronwtechnologynuggets/autonomous-vehicles
Vox February 28, 2020
"It’s 2020. Where are our self-driving cars?"
In the age of AI advances, self-driving cars turned out to be harder than people expected."
"When it comes to self-driving cars, the future was supposed to be now."
"In 2020, you’ll be a “permanent backseat driver,” the Guardian predicted in 2015. “10 million self-driving cars will be on the road by 2020,” blared a Business Insider headline from 2016. Those declarations were accompanied by announcements from General Motors, Google’s Waymo, Toyota, and Honda that they’d be making self-driving cars by 2020. Elon Musk forecast that Tesla would do it by 2018 and then, when that failed, by 2020."
My Comments
As I learn some AI and ML it seems according to Professor Ng that understanding hand gestures is a hard problem compared to knowing where other cars are located on a highway.
https://en.wikipedia.org/wiki/Waymo
"In December 2016, the unit was renamed Waymo, and made into its own separate division in Alphabet. The name Waymo is derived from its mission, "a new way forward in mobility". Waymo moved to further test its cars on public roads after becoming its own subsidiary."
Conclusion
Level #5 fully autonomous is a vague term and needs to be expanded to define whether the driver can influence the fully autonomous vehicle's path.
Are Autonomous Cars Really Safer than Human Drivers?, Scientific American, Feb 3, 2018
https://www.scientificamerican.com/article/are-autonomous-cars-really-safer-than-human-drivers/
It’s 2018, so where are the self-driving cars?, The Verge, January 30, 2018
https://www.theverge.com/2018/1/30/16950726/state-of-autonomous-car-self-driving-ces-detroit-2018
What If Self-Driving Cars Can’t See Stop Signs?, Bloomberg, Feb 2, 2018
Artificial intelligence is still too easily tricked to be trusted.
https://www.bloomberg.com/view/articles/2018-02-02/what-if-self-driving-cars-can-t-see-stop-signs
The Cyclist Problem, Slate, Feb 3, 2018
Self-driving cars aren’t good at detecting cyclists. The latest proposed fix is a cop-out.
Added Feb 27, 2018
Before Hitting the Road, Self-Driving Cars Should Have to Pass a Driving Test, Scientific American, Feb 22, 2018
"Human driving tests include some evaluations of a driver’s judgment and decision-making, but tests for self-driving cars should be more rigorous because there’s no way to rely on human-centered concepts like instinct, reflex or self-preservation. Any action a machine takes is a choice, and the public should be clear on how likely it is that those choices will be safe ones."
"Comparing With Humans"
"Self-driving cars’ algorithms constantly calculate probabilities. How likely is it that a particular shape is a person? How likely is it that the sensor data means the person is walking toward the road? How likely is it that the person will step into the street? How likely is it that the car can stop before hitting her? This is in fact similar to how the human brain works."
"That presents a straightforward opportunity for testing autonomous cars and any software updates a manufacturer might distribute to vehicles already on the road: They could present human test drivers and self-driving algorithms with the same scenarios and monitor their performance over many trials. Any self-driving car that does as well as, or better than, people, can be certified as safe for the road."