Friday, January 24, 2020
4th edition of the 8th year of SmartDrivingCars
The Disengagement Myth
Kyle Vogt, Jan 17, "In a few weeks the California DMV will release disengagements data from Cruise and other companies who test AVs on public roads. This data is really great for giving the public a sense of what’s happening on the roads. Unfortunately, it has also been used by the media and others to compare technology from different AV companies or as a proxy for commercial readiness. Since it’s the only publicly available metric, I don’t really blame them for using it. But it’s woefully inadequate for most uses beyond those of the DMV. The idea that disengagements give a meaningful signal about whether an AV is ready for commercial deployment is a myth. …" Read more Hmmmm… Amen! This is a MUST read. As with everything, details matter. It is true that figures don’t lie, but but it is easy to game systems such that figures, without the underlying details, do lie. As Kyle points out, there are important details associated with disengagements. These need to be well understood for disengagements to be a proxy for safety and market readiness. The when, where and associated details of each disengagement is critically important if the objective is safety and market readiness.
What is also most important here is the underlying objective of the companies doing the tests and reporting the data. As has happened in our secondary education where students are taught what is in and how to take the SATs rather than just learn. The objective is not learning , but getting 800s on the SATs so that they can get into ‘Princeton’. This is perpetuated by the ‘Princetons’ of this world that don’t look into the details of the student’s academic qualities and capabilities. In the academic world, we know these students as ‘box checkers’, gamers of the college admission process. The gaming is continued by the ‘banks and med schools’ that use simplistic GPA (Grade Point Average, aka ‘disengagements’) cutoffs. The ‘box checkers’ then take ‘underwater basket weaving’ courses and become grade grubbers. It is lazy and irresponsible to use simplistic measures as proxies to very complex concepts such as intelligence, creativity, compatibility, and all the other details that make a good student, a good employee, a good citizen, a good mobility system.
In our case, testing is assumed to be about safety and market readiness; however, for some, it may be about trying to "make a silk purse out of a sow’s ear" or "putting lipstick on the pig". It is easy to game the metric ‘Disengagements’ by simply testing in easy places, under easy conditions, instead of really trying to find the corner/edge cases that you don’t know in places and conditions of the Operational Design Domain that you are actually going to serve and make a business out of all of this technology; rather than just trying to get good press, or flipping it to someone else or putting it on an academic self. The details would readily divulge the real objective of the company doing the testing.
I hope that Kyle, in his next post, will divulge what he, GM’s lawyers and GM’s board are requiring of his system for each of them to sign off and begin to operate an economically viable mobility service to the general public in some ODD. Each will demand that it be safe. The board will also demand that it be profitable. What details are they requesting that will make each comfortable signing on the bottom line? Alain
Smart Driving Cars Podcast Episode 139- Randal O’Tool
F. Fishkin, Jan 25, "Adaptive cruise control and self driving tech may lead to more urban sprawl. But the Cato Institute’s Randal O’Toole says maybe that isn’t a bad thing. He joins Princeton’s Alain Kornhauser and co-host Fred Fishkin for that plus Subaru’s tech, GM’s Cruise, Tesla and more on the Smart Driving Cars podcast. This edition is sponsored by the SmartETFs Smart Transportation and Technology ETF, symbol MOTO. For more information…head to www.motoetf.com " Just say "Alexa, play the Smart Driving Cars podcast!". Ditto with Siri, and GooglePlay … Alain
Comment on SmartDrivingCar.com/8.02-USDoT_v4-011220
R. O’Toole, Jan 13, "Congratulations on your new Subaru Outback. I’ve had one with Eyesight since 2015 and it has made driving much more pleasant for me (a die-hard cyclist). It is also much safer: …
I agree with you that self-driving cars will lead to urban sprawl, but I don’t see what is so unfortunate about it…
Another name for urban sprawl is affordable housing… " Read more Hmmmm… Ouch! Smart Villages/Communities along with Smart Cities Read on and listen to PodCast. Alain
Comment on SmartDrivingCar.com/8.03-HD_Maps-011820
M. Hornung’79, Jan. 17, "So, please, explain to me why I need super accurate info about the stationary things
see attached 4 of the 5 presentations from the Monday TRB session 1164 GIS Data Needs for Connected Autonomous Vehicles" … DeepMap…, Geospatial…, HD Maps…, HERE’s…
"The gist of what I heard is the map data customers think if they had a database of road furniture (poles, trees, curbs, overpasses, etc.), in conditions when the camera image and/or the LIDAR cloud is less-than-ideal (driving into the sun at sunset, rain, snow, hail, fog, etc. …. One also has radar and one should never drive in fog. slow down, pull over or just stay home… ), they can make a higher-probability guess as to what’s in front of them (is it an overpass they can drive under, or a truck stopped across the road?) or ahead on the side (is it a vertical pole or tree, or is it a person that might suddenly start jaywalking?)…" Read more Hmmmm… HERE’s DeepMap offering has data on building outlines along the streets as well as road furniture. Why do I need or want all those data while I’m driving down each road?? I know… It is because the road furniture data is useful only if I know WhereAmI to the same level of precision. Then I can take the difference of the two position vectors to know precisely how far I am from any of the road furniture.
But how do I know WhereAmI? … SLAM! That’s where the building outlines etc come in. But I need to be able to range those objects from my nose. Granted, because of the law of large numbers I can get a high precision value of WereAmI with a ranging capability that is of lower precision. However, when I take the difference of two high precision values, the precision of the difference, the precision of the range of the object from my nose (which is the value that I need), is very much lower than the precision of WhereAmI and the HD map data, especially when the object is nearby. It is the nearby objects which I care about the most but are the objects whose location relative to my nose has the worse precision.
This is not an a good situation… as I approach an object, the precision of the ‘SLAM/HD_Map’ method degrades. Moreover, to use the HD data I’ve had to be able to range the HD objects. Granted at a lower precision, but at a precision that actually improves as I approach the object. Then why don’t I just focus on ranging the objects of interest rather than objects for SLAM and bypass using SLAM and HD maps completely? Duh!! (God must have realized this early on when she gave all of us (including dogs, cats, flees, fish, … ) capabilities to do somewhat precise ranging for collision avoidance rather than filling our memory with HD maps and a SLAM algorithm. The world is composed of living objects that don’t know where they are yet are very capable of not crashing into things by simply ranging them. I fail to appreciate why AVs should be different (unless I’m selling HD maps). Alain
G.M.’s Cruise Unveils a Self-Driving Car. Don’t Look for It on Roads
E. Griffith, Jan 22, "Cruise, the autonomous vehicle division of General Motors, unveiled an ambitious new vehicle on Tuesday that its executives said was “the beginning of the future beyond the car.” Emphasis on “beginning.”
The futuristic electric vehicle, the Cruise Origin, has a long road to travel before any passengers will be able to hail a ride in it. Cruise would not name a time frame for its availability. It provided no details on how many vehicles it planned to produce, or whether it has begun test drives on closed tracks. It has not obtained state or federal regulatory approval to drive on roads.
“Our work is far from done,” Dan Ammann, chief executive of Cruise, said during a presentation in San Francisco.
Cruise emphasized that the Origin was more than just an idea, however. In an interview, Mr. Ammann said the company would begin producing prototypes of the Origin and test driving the car “in the near future.”…" Read more Hmmmm… Looks like a good "beginning". This is a vehicle that is focused on providing Driverless Mobility-as-a-Service. Congratulations! ( see also Fred Lambert’s take…"GM Cruise unveils self-driving electric car for ride sharing" Alain
Scooter-sharing service Revel is racking up injury lawsuits months after launch
A Denney, Jan 19, "Moped-rental service Revel has only accelerated chaos on the streets of Brooklyn and Queens, endangering riders and other travelers alike through shoddy training and faulty parts, a series of lawsuits claim.
With just $5, a valid driver’s license and Revel’s smartphone app, users can ride off on one of the 1,000 electric-powered scooters the company has rolled out in the boroughs since May 2019 — no training required.
That level of ease has turned outer-borough traffic into a high-stakes road test, according to lawyer Daniel Flanzig, who represents two people bringing personal-injury suits against Revel in Brooklyn Supreme Court.
“New York City is not the proper place to be learning for the first time how to ride a scooter,” said Flanzig. “It’s not all that far off from a motorcycle.” …" Read more Hmmmm… Revel must know if they have a novice problem. Revel knows the who, when and where of at least some if not all of their mopeds crashes. They also must be saving some use characteristics of each of their customers. Those data sheds substantive light on the relationship between user experience and crashes. If revel hasn’t analyzed those data internally, they should or simply release the data and one of us will do the analysis. Some lawyer will definitely seek these data in discovery. Alain
The Tesla Skeptics Who Bet Against Elon Musk
D. Hull, Jan 22, "….Hothi is part of the counterrevolution—an informal yet obsessive global fraternity of accountants, lawyers, hedge fund managers, former Tesla employees, and some randos who just love trolling. For a few years they’ve been posting sick burns and negative indicators in equal measure with a #TSLAQ appended. They research executive departures, lawsuits, customer complaints, and accidents, adding the occasional allegation of financial chicanery and a passel of puns to suggest that Tesla’s end is nigh….
Lately, Musk and his company have had the upper hand… Tesla short sellers lost more than $2.8 billion in 2019, according to Ihor Dusaniwsky, a managing director at analyst S3 Partners, and have already lost about that much in 2020.. " Read more Hmmmm… Is TSLA now a Bubble? Alain
Tesla defends against claims of ‘unintended acceleration,’ says TSLA short is behind it
F. Lambert, Jan 20, "Tesla released a blog post to defend itself against claims of “unintended acceleration” that is being reviewed by NHTSA. The automaker says that the petition was started by a TSLA short seller who is betting against the company…." Read more Hmmmm… You decide. Alain
Car buyers shun electric vehicles not named Tesla. Are carmakers driving off a cliff?
R. Mitchell, Jan. 17, "Regulators are demanding zero-emission vehicles. And manufacturers are scrambling to provide them, spending billions on electric-car development. Ford says a third of its vehicles will be electric by 2030. Volkswagen plans to sell a million EVs annually just two years from now. At Volvo, half its offerings will be electric by 2025. By the end of this year, most major automakers will be offering at least one EV.
And buyers? So far, they’re not on board — especially those in the United States…." Read more Hmmmm… A very good read. Add to the reasons non-Tesla’s aren’t selling… (inside) competitions!
Tesla only sells EVs. All other OEMs have both gas and EVs. Now consider how cars are sold. If you land on the Tesla site, or a Tesla store front, you only see EVs. If you go to any of the other OEM’s showroom, you have a choice of gas or EV, PLUS you have a salesperson in the loop. Given the EV challenges that Russ lists and the salesperson’s overwhelming desire to sell you a car (any car), unless you are dead set on buying an EV, you are going to drive home with a gas powered car. They are so much easier for the salesperson to sell. (Plus if you are dead set on buying an EV, you’ll just go to the Tesla site. Because of the way they’ve always sold cars, the other OEMs don’t have a chance.) Alain
Tesla Tops Volkswagen to Become Second-Most-Valuable Auto Maker
T. Higgins, Jan 22, "Elon Musk has navigated Tesla Inc. TSLA +0.28% into new territory, as the electric-car maker’s market value topped $100 billion Wednesday and overtook Volkswagen AG as the world’s No. 2 most valuable auto maker.
Tesla shares rose 4.1% to $569.56, lifting its market cap to $102.7 billion, according to FactSet. Crossing the $100 billion threshold could start unlocking a more than $50 billion pay package for Mr. Musk. Volkswagen shares fell 1.19% in German trading, putting its market cap around 90 billion euros ($99.6 billion)….
Tesla, when it reports 2019 earnings in the coming days, is expected to post $2.6 billion in earnings by that measure and $24.19 billion in sales, according to analysts surveyed by FactSet….
President Trump marveled at the change in fortunes during a Wednesday interview with CNBC. He praised Mr. Musk and said he recently spoke with the CEO, suggesting the auto maker planned further expansion in the U.S….
On Wednesday, Tesla was greeted with more good news. The state of Michigan announced a deal to settle a lawsuit with the auto maker over a years-long dispute that prevented the company from selling directly to customers in the state…. " Read more Hmmmm… And the hits just keep on comin’ Alain
Issues in Autonomous Vehicle Testing and Deployment
R45985, Nov. 22, "Autonomous vehicles have the potential to bring major improvements in highway safety. Motor vehicle crashes caused an estimated 36,560 fatalities in 2018; a study by the National Highway Traffic Safety Administration (NHTSA) has shown that 94% of crashes are due to human errors. For this reason, federal oversight of the testing and deployment of autonomous vehicles has been of considerable interest to Congress. In the 115th Congress, autonomous vehicle legislation
passed the House as H.R. 3388, the SELF DRIVE Act, and a separate bill, S. 1885, the AV START Act, was reported from a Senate committee. Neither bill was enacted. In the 116th Congress, interest in autonomous vehicles remains strong, but similar comprehensive legislative proposals have not been introduced. The America’s Transportation Infrastructure Act, S. 2302, which has been reported by the Senate Environment and Public Works Committee, would encourage research and development of infrastructure that could accommodate new technologies such as autonomous vehicles…" Read more Hmmmm… Indeed, Washington’s #1 interest in regulating transportation is safety. This report does focus on safety (although I’d rather suggest that it is human misbehavior, some/much of which is premeditated, that is responsible for the fatalities, not just "error"). However, Washington should also have interest in improving mobility for all. This report essentially doesn’t seen recognize that objective. It is largely focused on sustaining the ‘do-it-yourself’ personal car ownership model of mobility rather than the opportunities of ‘shared mobility-as-a-service-for-all’ opportunities of the driverless version of this technology.
I continue to rile against the SAE levels, especially ‘Level 5’ (… "The automated system performs all driving tasks, under all conditions that a human driver could"…, p.2) Not in my life-time (which almost ended 12/5/19) nor in the life time of anyone alive today will the technology be as good as a responsible alert human driver today. Our ‘safety problem’ is not about responsible alert drivers. They’re responsible for only ‘6%’ of the crashes. They are damn good! Our challenge is to
1. put an automated safety net around the human driver that doesn’t let the driver evolve into a crash (Safe-driving car) and
2. provide much improved affordable mobility to many (Driverless car, or call it Level 4 if you wish with an Operational Design Domain (ODD) which serves real mobility needs now and has the opportunity to grow and scale. Who knows, ‘Level 4’ may eventually get as close to all as we’ll get to Vision Zero. Alain
Most lidars today have between 1 and 128 lasers—this one has 11,000
T. Lee, Jan. 16, "Lidar sensors work by bouncing laser light off surrounding objects to produce a three-dimensional "point cloud." The first modern three-dimensional lidar was created for the 2005 DARPA Grand Challenge, a pivotal self-driving car competition. Today, many experts continue to see lidar as a key enabling technology for self-driving cars.
That original 2005 lidar, made by a company called Velodyne, contained a vertical array of 64 lasers that spun around 360 degrees. Each laser had to be carefully aligned with a corresponding detector. This complexity contributed to prices as high as $75,000…. There are now dozens of startups trying to build cheaper lidar. Many of them try to reduce costs by using a single laser beam that’s scanned in a two-dimensional pattern.
But other lidar companies are taking things in the other direction: building lidars with thousands of lasers. A company called Sense is selling a lidar with 11,000 lasers for around $3,000 each. Another company called Ibeo is working on a lidar that will have more than 10,000 lasers.
To be clear, Ibeo’s new lidar isn’t out yet, so we don’t know how well it will perform. And Sense’s current lidars aren’t close to matching the performance of Velodyne’s best lidar. They have a range of 15 to 40 meters, compared to more than 200 meters for some Velodyne units…." Read more Hmmmm… It is nice to see the progress in Lidars, but the competition is a CCD (camera) and its supporting image processing software. Alain
Simulation: Preparing Autonomous Vehicles For The Road
H. Laine, Jan 22, "… A number of companies (my own included) provide simulation software for ADAS and autonomous vehicles, so I’ve seen the benefits of simulation firsthand. It has become a key technology for virtually testing a vehicle in a highly realistic environment that can mimic all weather conditions, lighting and city roads, down to the actual cracks in the road. The purpose of virtual testing is for the vehicle to “practice” driving in an environment almost identical to real-world conditions…" Read more Hmmmm… In case you didn’t already know. Alain
On the road to Washington, D.C.
D. Burr, Jan 23, "We’re on a mission to deliver safe and reliable self-driving technology to Uber users around the world and create positive impact through self-driving cars that drive more than themselves. Uber Advanced Technologies Group has already started this journey in Dallas, Pittsburgh, San Francisco, and Toronto. Today, we’re adding another city to that list 一 our nation’s capital….
Starting on January 24, we will begin exploring a new mode of transportation in Washington, D.C. when we bring our self-driving cars to District roads for data collection. Importantly, these cars will be in manual driving mode 一 meaning a Mission Specialist (a specially trained vehicle operator) will maintain control of the vehicle at all times.
We are using a phased approach to develop and deploy our self-driving vehicles, taking the necessary steps to operate safely in every city. Manual data collection is the first step in our development process and part of how we are validating our self-driving expansion approach. This manual data collection allows us to:
- Develop high-definition maps, the foundational information layer for our self-driving system…"
Read more Hmmmm… I’ve argued above that this "foundational information layer" is a phantom foundation. Nothing in this layer could have averted the Elaine Herzberg crash. In fact, reliance on this foundational information layer could/may have contributed to the disregarding/discounting of real-time range data that had detected Elaine 6 seconds before the crash. Alain
C’mon Man! (These folks didn’t get/read the memo)
There are so many bad articles. I’m overwhelmed. C’mon Man! Alain
No one should ever buy a car again. Here’s why
C. Taylor, Jan. 23, "… Ever-growing fleets of smart vehicles will eventually make ownership obsolete. The wait for a car to be hailed will keep going down, until it becomes almost always faster and cheaper to get a ride anywhere than to bother with parking and walking to your destination. In many city centers, that tipping point has already been reached. Even now, buying a car makes little economic or environmental sense, and it will make less sense with each passing year.
So why would you ever do it again? … " Read more Hmmmm… Too Sunday Supplement-ish so I made a new section. Alain
INSIGHT: The Decade-Long Self-Driving Honeymoon Is Over. What’s Next?
E. Tanenblatt, Jan 22, "A decade of breathless self-driving hype and C-suite predictions came to a quiet, ignominious end recently when the industry failed to deliver on its promise of mass-market commercial deployment. … Self-driving is in mass-market commercial deployment at Tesla, Subaru and others….
At last, reality has sunk in: Driverless cars are nowhere near ready for the open road. … True, and thank goodness the industry is backing away from consumer-oriented Driverless and focusing on Feel operators who will use these vehicles to provide affordable high-quality mobility …
And, despite having already spent tens of billions on autonomous systems, the industry enters this new year knowing that additional massive investment will be required to reach true automotive autonomy. … another new term "true automotive autonomy" what is that??? Please stop…
Even firms like Waymo and GM, whose tech stacks are rated among the most robust in the world, have been forced to delay commercial offerings or dramatically narrow their scope to purpose-built platforms or carefully geofenced environments. … nothing wrong with carefully geofenced environments. Even though some car commercial’s suggest that I can drive down river beds, up the great wall, through rock canyons, in deep snow I really can’t and shouldn’t try. I’ve been geofenced to a roadway network….
With deadlines in the rearview and the sticker price ballooning, these trends will define the new year and decade…." Read more Hmmmm… See comments in line above. Alain
Calendar of Upcoming Events:
4th Annual Princeton SmartDrivingCar Summit
evening May 19 through May 21, 2020
On the More Technical Side