Bits of Books - Books by Title
The Quest to Build the Driverless Car — and How It Will Reshape Our World
By Lawrence D. Burns with Christopher Shulgan
More books on Cars
More books on Inventions
When Burns moved to the Columbia U think tank (see below) he initiated research project to consider the implications of transport ind disrupted by three separate but inter-related factors - shared use vehicles,powered by electric motors, and driven autonomously.
2011 calculation that deployment of such an integrated system wd reduce the annual cost of travel in US by $4 trillion. For the individual, it wd reduce the direct and indirect (out-of-pocket and time costs) costs from $1.50 a mile to $0.25 a mile.
Better mobility for more people at a lower cost.
Will change the business model for car companies. Today, the average nett income per vehicle (everything except pickups) is between $1000 and $5000. That's selling millions of vehicles to millions of different customers. The future model will be running huge fleets of self-driving taxis earning $0.10 per mile for a 300,000 mile lifetime - ie a lifetime profit of $30,000.
Dave Hall famously built an amplifier at age of four. At college he patented a tachomemter, which gave him enough income that he didn't need to get a job after graduating. Instead he set up his own workshop. In 1979 he started manufacturing his own subwoofer invention,which finaced his battlebot hobby. He was a LIDAR pioneer, figuring out how to cram 64 lasers on to a single device. More importantly, his LIDAR spun, producing a 360°, dimensional picture of environment.
Hall hired Anthony Levandowski towork for his LIDAR company, Veladyne. AL had been in the first DARPA trial with a riderless motorbike called Ghost Rider (which was an ignominiuos failure - the operators forgot to turn on its gyroscope, and GR fell over ob the start line).
Sebastian Thrun cuthis teeth on a robot called Groundhog, which mapped abandoned mines in Pennsylvania for companies wanting to reopen them. In 1998 he built a robot docent for Washington's Smithsonian Museum. Solved navigation problems by creating a room map by sending robot around at night when place empty, and solved the don't-knock-a-child-over problem by coding it to interpret any change to the map as a human, and to stop and wait for it to move away.
Thrun ran the Stanford U team which won the second DARPA challenge (and a $2 million dollar prize) with Stanley, a converted 2004 VW Toureg. During their prep, they had attached multiple cameras on the roof to let them recap what had happened whenever something went wrong. Realized that it was interesting just going through the pics, so Thrun assigned an undergrad, Joakim Arfvidsson, to create a program to stich all the clips together to make seamless view.
Levandowski joined this team in 2007. He was responsible for creating a demo that would sell the startup to VCs. He hired a fleet of rentals (cheap if you rented by the month), and hired a fleet of drivers off Craigslist. Took two weeks to film all of San Francisco. Google bought the startup and team for an undisclosed sum, heading off VCs who were competing to invest.
Gm Autonomy concept car 2002 Auto Show - a skateboard with fuel cell powering 4 individual wheel motors. Everything below the platform, which allowed multiple cabins to be dropped on top.
GM, Ford etc sit on top of an 'intgrated auto industry'. They manage a supply chain of parts suppliers to assemble finished vehicle. In 2005 Burns and his boss, Rick Wagoner, called to the GM Vehicle Assessment Center. This was where GM disassembled their competitors' cars, right down to the last nut and bolt. At this visit, they were shown 3 vehicles: a Chevy Malibu, which was broken down into about 10,000 parts, a Toyota Prius hybrid, which had even more parts (bc double up systems), and finally, GM's electric car, the E-Flex. This had less than 1000 parts.
This was an epiphany, bc it meant the car was way easier to manufacture. Not only did it have a tenth the number of parts, it had a lot fewer moving parts. This meant the door was open for a lot more car companies, who wd not need GM's knowledge of organizing a huge supply chain.
And further, the crucial expertise was not going to be managing the hardware, but the software. Cars would be designed, engineered and controlled by programmers, not metal-bashers.
Analagous situation to the 1980's, when IBM outsourced the chips and the programming of PCs, not realizing that that was where the value lay.
Larry Page had to browbeat Sebastian Thrum into making AVs. Despite his experience with the successful DARPA challenges, Thrum considered normal street AV driving was impossible. But Page said "I've thought about this. Give me a technical reason why it can't be done. Not a societla reason. A technical reason."
Thrum reminded Page that he was the world authority on AVs, and if he said it was impossible, he should know. But he couldn't come up with a technical answer, and so, Oct 2008, first assembled a team to address the challenge. Chris Urmson charged with developing the software. Levandowski to run the hardware and Mike Montemerlo to develop the maps. The project was to be called Chaffeur.
The maps were the key for the cars to locate themselves in the world. First, cars equipped with LIDAR and cameras would drive the same streets multiple times day and night. By compring things that changed position between scans with those which didn't, the software created a list of staionery objects (curb edges, buildings, street signs, telephone poles etc). Then, when AV navigated same territory, it wd scan its current surroundings and match it with its database of maps. This gave a very precise location with a margin of error of two inches.
Basic navigation was straightforward. The team ran a demo obstacle cource which they put their AV through, then challenged the fifty Google execs watching to beat the time. None could.
From there, the first step was teaching robot to recognize objects. So thousands of images of pedestrians, wheelchairs, cats and dogs, skateboarders, balls etc. The most challenging was a traffic cop directing traffic, with their multiple idiosyncratic ways of signalling.
Then they had to develop a behavioural engine - predicting what allthose objects cd potentially do, including whatother vehicles might do. The trickiest one here was guessing whether a cyclist hitting an intersection on yellow light wd run the red, and so AV would have wait, even though it had a green light.
When Google tried to interest Detroit in its AVs, the execs were dismissive - "People like to drive" and "People's self-image is tied up with their car". But these old white guys had missed the change. Kids now were more interested in buying the latest iPhone than in having a cool car.
Zimbabwe's casual transport 'system' - if you wanted a ride you just stuck out an arm and the nearest car with an empty seat would pull over. You just paid a small sum as gas money.
An American driver's cost of car. The AA annually estimates the out-of-pocket cost - depreciation, fuel, insurance, maintenance and finance. They estimated that thetypical cost was $0.65 per mile, incl parking. Burns then calculated the cost of the time spent driving. (Simply divided av wage of $24 an hr by the av driving speed, which was roughly 28 mph). That cost came out at another $0.85 per mile, so total cost of owning an operating a car was about $1.50 per mile (in 2011). So Americans spent around $4.5 trillion per year driving.
Burns' team ran another set of calculations to figure out how many AVs wd be needed. Then ran multiple iterations with different scenarios, and kept coming up with the same numbers. You needed just 15% of vehicles currently in a town or city to simultaneously achieve the three vital goals - high fleet utilization, low empty miles (between fares) and fast response times. And to cover rush hour surges you wd need a maximum of another 5%.
Driving costs wd be dramatically lower. The car cd be built for lot less than $10,000, excluding the self-driving system. That, once in mass-production, will cost less than $5000. Based on 250,000 miles travel (which is what most taxis do), the cost wd be around $0.05 per mile. Cost of electricity to power it, $0.01 per mile. Maintenace costs set at $0.05 to allow for cost of replacing batteries. Insurance at $0.02 per mile bc expect very few accidents. And another $0.02 for parking and financing. Total cost $0.15 per mile v $1.50 for ICE car today.
So even though the car companies were blind to the coming disruption, Google could see this as a massive business opportunity. Conservatively estimating that AVs cd capture 10% of the 3000 billion miles driven each year in Amerca. If you charged $0.10 margin over cost, you wd still only be charging $0.25 a mile, but your profit would be $30 billion pa.
Google found a regulatory loophole - low speed vehicles. A LSV can weigh up to 3000 lbs and have a top speed of 25mph. So they built 100 prototypes called Firefly, which, sensationally, had no steering wheel or brake or gas pedals -just an ON and OFF button.
For Detroit, the game changed at beginning 2015. Insiders had been quietly challenging the petrolhead mentality of "Everybody likes to drive" (except nobody syays they enjoy their commute) and "people's self-image is tied up with their cars, particularly young guys" (except that today's youth would rather spend their money on a better cellphone than a car, and are quite happy to use Uber to get around).
The attention grabber was Uber swooping in and poaching virtually thewhole staff of the National Robotics Engineering Center at Carnegie Mellon. It got a lot of attention from Detroit bc basically an endorsement of Google's AV concept and a belief in first-mover advantage. The coup set off a stampede, as auto makers realized that the AVs were coming very soon, and that their business model was going to be overwhelmed.
Owning a car is inflexible - people buy them for "the occasional but imperitive" use. They'll buy a pickup or an SUV so that they can carry large loads once or twice a year, but then is massively over-engineered and equipped for what they usually use it for. So car fleets would offer renters small two-seaters for most of the time, but also have a small fleet of bigger, special purpose vehicles to cope with that occasional use.
Tesla used Mobileye to run its driver-assist system. But Mobileye's makers, an Israeli firm, repeatedly told tesla it was wrong to call it an Autopilot, bc it was designed purely for use on freeways to monitor other vehicles travelling in the same direction. It simply wasn't designed to detect another vehicle crossing the road at right angles (which is how the tesla driver in Florida died)
The car, notes Lawrence D. Burns in his book Autonomy, is terribly inefficient. The internal combustion engine converts less than a third of a gallon of gasoline into actual kinetic energy - the rest is wasted as heat and sound. And most of the kinetic energy goes to simply moving the (increasingly large) car itself. Only about 5 percent of the gasoline energy is used to move the driver. Most people drive to work alone, in cars with too much capacity and engine power for that purpose. Then there's the fact that most cars sit unused 95 percent of the time. Not to mention the societal costs in death and injuries, the environmental costs of the cars themselves, and the infrastructure supporting all this inefficiency. If this sounds insane, as one observer put it, Burns reports that he couldn't agree more.
This might read like the buzzkill PowerPoint of some progressive transportation think tank analyst. But Burns was, for many years, a vice president at General Motors, whose workforce, he points out, once exceeded the combined population of Delaware and Nevada and whose fortunes were intrinsically yoked to the country at large.
Burns did eventually go to work for a think tank - he's the director of Columbia University's Program on Sustainable Mobility at the Earth Institute - but for much of his life, this Detroit-born-and-bred engineer, schooled at a General Motors-run college before taking his first job with the company, worked firmly within the system, crafting massive 10-year plans and optimizing production processes. But, he admits, he never felt like a car guy. He was, at heart, a mobility guy, and mobility guys did not necessarily think that single-occupant-driven SUVs (still the bread and butter of the auto industry) were the best way to move the most people around most safely and most efficiently.
The story he sets out to tell in Autonomy, aided by the writer Christopher Shulgan, is one of increasing disenchantment with the status quo in Detroit. The car, after all, had barely changed since the Model T: Gas-fueled, run by an internal-combustion engine, rolling on four rubber tires, with the passengers protected by a windshield and four doors. Sure, there was plenty of incremental innovation, but the car was an entrenched technology stubborn to change. However, an epiphany was to come - from the desert where teams of roboticists were competing to come up with a way to reduce the number of soldiers dying while driving Humvees. The challenge from the Defense Advanced Research Projects Agency (DARPA) had rival groups working to get their driverless vehicles across the finish line on a course in the Mojave.
Burns later agreed to sponsor a team from Carnegie Mellon as it embarked on the biggest challenge yet: piloting an autonomous vehicle through an urban environment. This is not exactly the realm of 'The Right Stuff.' Here we have slow-moving cars bumbling through parking lots and patiently navigating four-way intersections. But you can sense the excitement in Burns, an engineer at heart, as the team works through the night in unheated trailers in Pittsburgh winters on what would eventually be the winning entry: a modified Chevy Tahoe, named Boss, that successfully completed the 60-mile course.
For Burns, this was more than a proof of concept, it was a nascent revolution, the sort of mobility disruption he hoped to see: He longed not only to take the internal combustion engine out of the vehicle but to remove the driver. The concept was launched as the Internet sharing economy took off. Now humans could be freed from parking - cars would do it themselves. And freed from driving - a driverless car ordered online would come pick you up and take you wherever you wanted to go. But, as Burns tells it, the industrial story is a familiar one: brave innovators running up against organization men. Unlike the 'move fast and break things' ethos of Silicon Valley, Detroit took years to move from concept to prototype. And the car guys mostly just didn't get why people would not want to drive. During the Urban Challenge, Burns notes, Google sent a 'planeload' of senior executives; GM, by contrast, sent only him. In Silicon Valley, the human driver was regarded as a bug - not a feature - in a car. For decades, the car guys in Detroit had sold cars on the promise that driving was freedom. Now the fear was that robot cars were the beginning of the end of humans. An ad for the 2011 Dodge Charger shows the human-driven vehicle picking up speed as a voiceover intones: "Leader of the human resistance."
The ad, presumably, was at least partially tongue in cheek, but it does illuminate a shortcoming of Autonomy. The book is a passionately argued, you-were-there account of the birth and rise of the autonomous vehicle from an authoritative Detroit voice. Burns, a technological utopianist (and one of the first people to get a cochlear ear implant), makes a number of compelling arguments for why smaller, self-piloted, shared vehicles make sense. But we don't hear much about that other great engine of the car business: consumer desire. Do people want to be driven? Corporations can be intransigent, for sure, but so can consumers. Seat belts in cars, for example, were introduced almost 70 years ago, but they were quickly abandoned because consumers resisted them, and it took many decades for this standard safety feature of today to gain acceptance. As with seat belts, it may take more than simple availability for consumers to adopt the technology.
And the engineering hurdles are signficant. Making cars drive themselves is, as Uber's co-founder Travis Kalanick once put it, a “hard problem.” Human transportation, however, can be a wicked problem. Fixing one problem can often lead to unintended others (Uber, which was posited as a way to smooth traffic by reducing car ownership, has arguably taken people off public transit in New York and helped worsen congestion.) And even the much-publicized early fatalities attributed to autonomous vehicles - two Tesla drivers using autopilot and a pedestrian struck by an Uber test vehicle - may have been results of programming decisions made by humans. A larger question, not much discussed in Autonomy, is how much risk we are willing to accept to have autonomous vehicles. What role should ethics have in the programming: Do we prize safety over speed and efficiency? And who should be the ultimate arbiters of those decisions?
Burns is right to envision a better future with computer-assisted driving (simply eliminating the possibility of drunken driving would save thousands of lives in the United States alone). But like a parent handing the car keys to a newly licensed teenager, relinquishing autonomy to computers - which also seems inevitable — will come with a curious mixture of hope, fear, regret and more than a few mishaps. "Enjoy the ride," Burns offers as his final words - but do buckle up.
How has autonomy become a commonplace term in our conversations? It seems only a short time ago that autonomous vehicles seemed futuristic and a little far-fetched, doesn’t it? In Autonomy: The Quest to Build the Driverless Car… and How It Will Reshape Our World, Lawrence D. Burns describes a transportation future where we "safely and conveniently use autonomous vehicles to take us where we want to go" (p. 1).
How have we come so close to this reality of on-demand transportation autonomy? In Autonomy, Burns narrates how robotics teams have taken hundreds of thousands of steps to train self-driving cars to react to the same obstacles to which human drivers react. And the story of that technological roadtrip is fascinating - especially to those of us who have heard pieces of the autonomy story but didn’t understand the ramifications of the self-driving Big Picture.
The book points to a future in which our trips will primarily take place in a 2-seater electric vehicle (EV) hailed through a car-sharing company. As users, we'll pay a "monthly subscription fee in exchange for on-demand use of a company vehicle for a certain number of miles per month" (p. 228). Autonomy helps us to understand this likely future reality by taking us back to the beginning of robotic transportation research.
The book, however, is not an encyclopedic compilation of dry sequential facts. Burns' career spanned head of research and development at General Motors to consultant at Google. Likely not a confident writer himself, Burns was wise: he brought on a ghostwriter, Christopher Shulgan. As a result, the chapters unfold as stories with a mystery genre-like tone.
For example, Chapter One is the story of the Defense Advanced Research Agency (DARPA) challenge to stage a race for robot cars across the Mojave Desert. The adventures of the robot named Sandstorm are unveiled in real time and with a sense of suspense, urgency, and wonder - will it win? Had its team anticipated all technology glitches?
Of course, having famous names like Google co-founders Larry Page and Sergey Brin sprinkled throughout the chapters doesn't hurt, either. ("Larry's always been a robotics enthusiast," one of the team noted.) The authors blend dialogue from interviews; infuse historical contexts (i.e. the importance of drone imagery from Iraq and Afghan territories); trace a cast of unlikely, contradictory, and often combative tech team characters; and, unveil a play-by-play of robot races that reads like a whodunit.
When our protagonists (spoiler alert) fail to win the DARPA challenge, all is not lost. A second race, which, like the first, required "publishing for the rest of the robotics community the secrets of all competitor'’ approaches" (p. 52), brings in new characters like Sebastian Thrun from the Stanford AI Laboratory. His team heightens tensions among the original Red Team of Red Whittaker, Chris Urmson, and others - some of whom were drawn from a graduate-level seminar class called Mobile Robot Development.
Instead of Sandstorm's electrical motor and lever that pushed against the gas pedal, the new entry, HIghlander, has a throttle that was controlled electronically through a computer system. With less margin of error, HIghlander was becoming a better robot driver.
Sure, robot cars encountered obstacles they didn't understand, like rolling over on a test track when sandbanks resembled a computer scenario that called for acceleration into a curve. When the robot would encounter something it couldn't handle during research phases, someone on the team would code a fix. As the process repeated itself dozens, and eventually hundreds of times, the robot became sophisticated enough that it began to teach itself.
Robotics had been seen as novelties, curiosities that had little effect on anyone's day-to-day lives. Then the 3rd race, the 2006 DARPA Urban Challenge, sought robots that could navigate the chaotic urban environments of Iraq or Afghanistan - traversing 60 miles in 6 hours while obeying the rules of the California Driver Handbook. As part of that Challenge, Dave Hall became fascinated with LIDAR's potential to create a 3-dimensional scan of the world so that the robot could detect oncoming vehicles in all directions. And Anthony Levandowski - later to become well-known in 2017 in a lawsuit between Waymo and Uber - was selling LIDAR to as many teams as he could. Meanwhile, Thrun and Levandowski sold their VueTool technology to Google to accelerate the Street View project.
While Street View was in development, Urmson and Slesky were constructing situations that would wreak havoc on their vehicles computer algorithms - situations unlikely to happen in real life, which resulted in a robot that could drive in traffic at the speeds that were commonplace on public roads.
Problems did stymie continual forward progress in autonomous vehicles. As GM tried to separate itself from its health care obligations and with the specter of bankruptcy looming, company research into autonomy fell by the wayside. Ideas to design an autonomous vehicle based on the Segway idea seemed, well, silly to audiences. The idea that self-driving cars were a near-term possibility seemed too idealistic - that is, until Larry Page told Thrun that he should work on self-driving cars. With the budgetary backing that Google could provide, Thrun hired a team of about 12 engineers to create an autonomous vehicle that could successfully navigate 10 drives totaling about 1000 miles. These routes would duplicate the driving experiences of most any California street.
Using the Google Street View service, which the Autonomy authors describe as "a cartographic achievement unmatched since the days of Vasco de Gama and Magellan" (p. 173), the team had access to artificial intelligence and computer-vision software. Their autonomous vehicle, named Chauffeur, had an ability to locate itself in the world, conduct a similar scan of the area around it, and match its current surroundings with the list of landmarks in its 3-D maps. How did it work? The car would:
compare its preexisting list of stationary objects with the objects around it to discern which objects were likely to move
draw upon the 3-D maps, which could also help the vehicle discern the dotted or solid lines in the middle of the road
determine tricky but important parts of the world around it - like traffic lights (p. 174)
Optimism permeates Autonomy. When Google executives are invited to run the same route as Chauffeur but faster than the robot, not one exec succeeds. The authors exclaim that the team was creating a vehicle that could drive not only as well as a human being - but better than one. Faster, sure. But more importantly, more safely, with less of a tendency to get distracted or confused.
These tests were conducted on public roads, which shocked America when newspaper headlines captured the pilot project. Still, the authors insist that the Google car had sensors all around it. It knew what was happening ahead, as well as to the right and the left, and behind — at all times. However, pitching the idea to Detroit's major automakers didn't make sense to them, Urmson recalled. Self-driving cars, the authors argue, are "so transformative, in terms of safety, efficiency, the automobile’s environmental effects, that Detroit's appropriate response should have been a rapid embrace."
Undeterred, the team continued on, feeling "like we were working on something that would change the world, that would have a positive effect on many of society's most pressing problems, from pollution to the most basic challenge of just getting around our planet"” (p. 201)".
The remainder of the book chronicles:
potential calculations of profitability in autonomy
the difficulty in keeping the human driver's attention once lured into the calm of an autonomous vehicle - and accidents that occurred as a result of inattention
fleet transportation, historical analyses of job displacements (such as autonomy will create), the move toward commonplace understandings of self-driving technology with Tesla's self-driving software and other automakers’ versions
the dilemma of disruption on established corporations, among others
The book ends with the acknowledgement that the mobility disruption will affect different people differently. The elderly will experience liberated mobility, while some current autoworkers many lose their jobs as a result of autonomous technologies.
"If we can pull it off," the authors conclude, "and we will, we're going to take 1.3 million fatalities a year and cut them by 90%. We're going to erase the challenges of parking in cities. All of that land will allow us to reshape downtowns. People who haven't been able to afford a car will be able to afford the sort of mobility only afforded to those with cars. And we’re going to slow climate change" (p. 326).
Books by Title
Books by Author
Books by Topic
Bits of Books To Impress