|
Getting your Trinity Audio player ready...
|
After the DARPA Challenges
Before 2004, Carnegie Mellon University (CMU) did the most to advance the idea of Autonomous Vehicles (A.V.) in the U.S. After 2007, the history of Autonomous Vehicles (A.V.) became a Silicon Valley story.
The three California DARPA Challenges created massive interest among sponsors, researchers, the media, and the general public. DARPA called it a success, thinking it was time for the automotive industry to step up and commercialize the technology. At this point, the car manufacturers should have jumped in and snatched up all those young, talented engineers and students, used the science and turned it into applied science. But somehow, that didn’t happen.
CMU still had their NREC (National Robotics Engineering Center) working on top secret projects with the military and the mining industry. The Stanford Team had moved on and worked on various mapping technologies. In fact, many teams just returned to their universities, continuing their research within the confines of their labs. Car manufacturers, traffic planners, and “mobility consultants” ignored the real thing and moved back to creating slideshows, simulations, and concept cars when explaining the ‘Future of Transportation’ and ‘Modern Mobility.’ Despite what they had just witnessed, everybody still treated the Autonomous Car as science fiction or something Michael Bay would make a movie about.
‘Move Fast and Break Things’
This is, however, where Silicon Valley is at its finest. Looking for outrageous ideas with potential and throwing tons of V.C. money at it seems to happen a lot around here.
In this case, two Silicon Valley entrepreneurs – Larry Page and Sergey Brin – made lots of money with their 6-year-old small search engine company named Google (this is still 2004), at least so much so that they could venture into some other hobbies and take on some risks. They were working on their browser, mobile phones, and operating systems. Larry Page had some personal history with automobiles – his grandfather worked for General Motors (G.M.) – but also some fascination with robotics and robot cars. Conveniently, the three DARPA Challenges happened in Silicon Valley’s playground: the Mojave Desert.
In 2004, Google made two weird acquisitions. They bought mapping/satellite technology from two Danish brothers and a software company with a fascinating tool – which I even had a subscription for – named Keyhole. These two acquisitions became Google Maps and Google Earth. They added another company named ZipDash to integrate real-time traffic analysis.
2004, 2005, and 2007 were the years of the DARPA Challenges, and Google sponsored all three. First, the CMU team, but then they were closer to Sebastian Thrun and the Stanford team. By 2006, Sebastian Thrun invited Anthony Levandowski (the Berkeley GhostRider guy from the 2004 challenge) to Stanford and worked with him on VueTool – a camera-based mapping project. This project, including all team members, was acquired by Google in 2007 to develop what is now known as Google Street View.
All these weird acquisitions and interests started to come together once Google released its own browser (Chrome) and mobile phone operating system (Android O.S.), which includes turn-by-turn navigation and real-time traffic information. But as it turns out, maps are also very, very helpful when you want to create and train an Artificial Intelligence (A.I.) or robot car.
Hollywood and Pribot are reigniting the Spark
Still, the idea of autonomous vehicles might have gone to sleep again if it hadn’t been for the Discovery Channel’s T.V. show Prototype This! The showrunners had a production office on Treasure Island and challenged inventors to prototype pizza delivery from San Francisco downtown to their Treasure Island office. There were several ideas through the air or by water, but Anthony Levandowski suggested using a robot car.
Previously, Levandowski worked with several DARPA Challenge teams due to his knowledge of LIDAR technology. He was also working with Thrun on the Google Street View technology. On September 7, 2008, he brought it all together when his silver Toyota Prius Pribot, delivered a pizza from North Beach Pizza to Treasure Island using the Embarcadero and Bay Bridge. He had a lot of help from Caltrans and SFPD and only one hick-up. The issue was the left-turn exit to get off the Bay Bridge and onto Yerba Buena, which – judging by all the skid marks that exit has – is also a significant problem for many human drivers.
For the small amount of time and money spent on this prototype, this first autonomous vehicle driving through San Francisco was seen as a historic achievement and a huge success.
Project Chauffeur – creating the World’s Best Driver
And still, there wouldn’t be self-driving vehicles in 2024 if Larry Page wasn’t so stubborn about it in 2008. Levandowski’s pizza delivery success inspired Larry Page to take on the challenge of self-driving vehicles for real this time. He and Sergey Brin created Google X (at least with this X, we don’t have to add ‘formerly Twitter’ each time) and put Thrun in charge of a project named Chauffeur. They brought together DARPA Challengers Anthony Levandowski (team UC Berkeley ‘GhostRider’ and project ‘Prototype This!’), Chris Urmson (team CMU), Mike Montemerlo and Dmitri Dolgov (team Stanford) and added a few more specialists Thrun and Urmson had worked with before (e.g., Dirk Hähnel, Russ Smith, Jiajun Zhu, Nathanial Fairfield, …). Bringing all that DARPA challenge knowledge and know-how from Google Maps and Street View together as one team helped immensely. Now, all these acquisitions have put Google in a unique and advantageous position.
They used the Toyota Prius they had used for Street View and pizza delivery and faced many challenges. There was a bonus system depending on whether the team reached certain milestones. With that in mind, Team Chauffeur just set out and went on the road – this time without the help of Caltrans or SFPD. They basically went dark mode and rogue and drove their vehicles around Palo Alto, Mountain View, Lake Tahoe, Big Sur and up and down SR-101 and I-280. The California Vehicle Code was apparently totally fine with that as long as a human driver was ready to take over at any time.
Driver-Assist vs Self-Driving
Of course, one of the big questions was the philosophical difference between creating a driver-assist system and an autonomous vehicle. Just for legal and safety reasons, the project started as a driver-assist system, where a driver had to sit in the driver’s seat and be able to take over at any time. But Team Chauffeur wanted to rule out human error completely. Human error is still the biggest flaw of today’s driver-assist systems, including Tesla’s FSD. By putting humans still in charge of safety, the Tesla car only needs to make the right decisions 80% of the time, maybe 90%, perhaps even 95%. But there is the moment when the car wants the human to take over, but that human isn’t paying attention. The NHTSA did tests where people checked out once auto-pilot was engaged, and it took some drivers an incredibly long seventeen seconds to regain control of their car. One famous case was Elaine Herzberg, the first pedestrian killed by a robot car. Legally, Tesla and Uber will never take responsibility in these cases; the advantage of a driver-assist system is that they can blame everything on the driver.
Google X (aka Chauffeur or Waymo) took a completely different approach. They planned on having real autonomous vehicles, where the car must make the right decision 100% of the time and can never blame the driver—simply because there is supposed to be none. That is what driverless really means. This very different approach also led to the fact that in 2024, Waymo had a nearly perfect safety record while Tesla and the other car manufacturers, with all their driver-assist systems, did not.
Iterations of Waymo’s Most Experienced Driver
The team initially wanted to provide “a new WAY forward in Mobility” or Waymo. They aimed to get rid of the driver seat and basically have the customer treat the robot car like an elevator. Get in, name your destination, hit the start button, lean back, listen to some elevator music or an elevator pitch and wait till the car stops again. Like elevators, there would probably be a red button and a two-way radio to speak with a remote operator in emergencies. This takes the human error completely out of the system. However, this also required Waymo to create the world’s most experienced driver, which required the correct decision 100% of the time.
First, Chauffeur used a small fleet of Priuses to drive several million miles around Silicon Valley. But, to learn the car manufacturing craft, they decided to create their own robot car from the ground up. They introduced Firefly in 2015 by having a blind man take a ride in this two-seater with no steering wheel and pedals.
“Google’s self-driving technology is impressive; our short rides in its two cars clearly demonstrated that there is plenty more capability beyond what the team showed us. Still, there are moments when you can’t help but be impressed.” [Editor MotorTrend]
Firefly only lasted two years and another several million of vehicle miles driven (VMT). But eventually Waymo decided to let the professional car manufacturers do their job and focus their own efforts on making their A.I. safer. Currently, Waymo is equipping the Chrysler Pacifica Hybrid and the Jaguar I-Pace with radar, LIDAR, and all the other tools necessary to turn these cars into The Most Experienced Drivers in the World.
Detroit and D.C. messed up big time
2008 should have been the year Detroit and G.M. finally got the message. It was the year of the financial crisis, where cash became a scarcity. All these fossil fuel car manufacturers had to do their Walk To Canossa and beg for a government bail-out to keep them afloat. All the while, Exxon ($45 Billion) and Chevron ($23 Billion) were raking in record income. Even a car and fossil fuel fetishist must understand the irony in this. 2009 should have been the year that the automobile industry weaned itself off of bigger, heavier, and faster and turned to smaller, nimbler, and more efficient. If they did or the government bailouts had forced them to, then today, we would celebrate MPGs over acceleration and pedestrian safety over larger grilles that kill 509 people per year.
When Washington, DC, handed over the bail-out money, they should have insisted that the automobile industry finally agrees to follow specific Pedestrian Safety Standards. D.C. did not, and now the next best hope for safety might be Waymo’s Robotaxi.
More Information:
- A.V.s: The Jurassic Period
- A.V.s: The 20th Century
- A.V.s: The Race in the Desert
- Motor Trend: Why Firefly Mattered
- History Lesson: Walk To Canossa
- Lawrence Burns at Google
- Book “Driven: The Race to Create the Autonomous Car”
- Book “Autonomy: The Quest to Build the Driverless Car”
Editor’s Note: The views and opinions expressed in all blog posts are those of the authors and do not necessarily reflect those of the Redwood City Pulse or its staff.
