Brad Templeton: 2019 Results for the Autonomous Car Industry

Waymo

2018 was the year when Waymo failed in an attempt to launch a real commercial product as originally planned. Instead, they launched an improved version of the limited access service in Chandler, Arizona, called Waymo One. 2019 also saw no commercial service, but they launched an expansion for autopilot, in the fall of unmanned operations, without accompanying drivers monitoring safety, there were more. To some extent, they did it in 2018 (but their announcement was overshadowed by the Uber failure a week later), but now it’s real. (They charge for travel, but not because they need money.)

Despite the fact that this is not a very big change, it remains the main topic because people do not understand what it costs for lawyers and engineers of a large company to subscribe to the deployment of unmanned vehicles and take all responsibility and risk upon themselves, if something happens will go wrong. This shows a very high degree of confidence. Waymo and a subsidiary of Nuro, which is engaged in small unmanned delivery robots that are unable to raise an adult, have agreed with California to release truly unmanned vehicles there, which is expected in 2020.

The change happened so gradually that my main coverage was the story of whether they did the right thing when an impatient videographer pursued them. It was an insignificant moment, the real story is that they are doing more than others, and on the way to do a lot, and no one else even stands by.

About the author: Brad Templeton – A software engineer, an evangelist of robotic cars since 2007, worked on Google in his early years. Founder ClarinetHonorary Chairman Electronic Frontier Foundation and director Foresight institute, founder of the faculty in Singularity university.

Winter

Another topic of the year was the slowdown in the development of robocars. Elements of the story included Waymo, who didn’t release everything as fast as promised, and retreat cruise from a plan where they launch paid services. Daimler backtracked toothat was most expected as well there have also been changes in other companies with less optimistic forecasts. At the beginning of the year, I emphasized that traditional manufacturers would probably welcome a slowdown, because in the first place they never wanted all this hype, and they are very happy to destroy their century-old business not so fast, but at a pace with which they can deal.

The slowdown was expected if you follow the model of the famous “cycle of hype” Gartner, and there is no doubt that the robocars too advertised and overestimated in this decade. For those who take this into account, things are not so bad. Usually, people expect a series of bold and desirable predictions to fail, and those who have taken advantage of this hype should now subside with it.

When asked when the robocars will appear, I gave two key answers:

1. When engineers, lawyers, and the board of directors realize that everything is safe enough to take on this risk, and
2. At different times in different places. In fact, the appearance of robocars in a given country can vary by about two decades, depending on economic factors and the work done.

No one named the exact year. Back in 2010, when I first started making any forecasts, I noted that most people were for 2020. They were pretty accurate when considering Waymo’s likely commercial launch next year as the start of all technology (or even a limited release this year). They were wrong only if you thought that in 2020 robocars will be everywhere, or that you can buy a car from the manufacturer and drive around most cities. The only people who predicted such a development of events were either poorly versed in technology or dreamed of advanced artificial intelligence that could cope with differences in all cities and countries at once.

Car manufacturers have always been mistaken in their forecasts, based simply on good appearance. The most common approach was to try to gradually make a robocar from an existing ADAS. This is not the wisest decision, but it makes sense.

Below I’m going to talk about Tesla, they had several of their own “winter” events, including the failed fully autonomous driving, which Elon Musk promised to customers in 2019. Tesla is still planning some sort of limited demo at the end of 2019 for several beta testers to at least somehow fulfill the promise. Another failure of Tesla was smart call too early, which undermined Tesla’s credibility, partly because it wasn’t really needed, and also because it could hesitate and sometimes cars could be damaged due to its mistakes. EU ordered Tesla turn it off.

At the other end of the spectrum is the Navya shuttle, the first sold robocar product that also slowed down.

What does it all lead to?

I am only slightly altering my past predictions. Soon we will see commercial pilot projects in several cities, limited only by the selected service area. They will test technology, security, state and public recognition and business models. In a couple of years, these projects will be deployed on a large scale. This will cause a fever, because if the manufacturer has a working technology, he should be the first one whose technology will begin to be profitable in the most interesting cities. They will rarely compete on equal terms. This fever will be in full swing in a couple of years, as soon as at least three companies take part in the United States. However, everything still depends on the decision of the teams about the sufficient security of their technology. Perhaps many will need much more time than expected. Success requires a lot of money, and going beyond the thinking of a regular automaker.

At the same time, some limited products will appear and improve, including an improved version of the Tesla autopilot and the so-called “level 3” products that drive personal cars in certain situations – traffic jams, highways, parking lots and, possibly, several selected urban areas.

Tesla Autonomy Day

Tesla is a raven among car manufacturers, unlike anyone else. But at the same time, it’s not a car outsider, such as Waymo, Zoox or even Cruise before it was bought out. Currently, Tesla has only one (best in class) driver assistance product, but it is very far from a real robot car in unmanned mode. Tesla claims to be very close to this, but most of the industry is skeptical.

Finally, we could seriously talk about Tesla’s efforts after their “Day of Autonomy”, when they suggested observing their experts in the field of autonomous driving. Someone was impressed, someone not, but many disputes continue to rage. Can the Tesla philosophy of introducing autopilot into an existing car without a lidar and detailed maps compete with the prevailing philosophy of using a more fashionable and custom-designed sensory kit, including lidar. I have suggested several articles on these issues, including:

• Elon Musk’s War On LIDAR: Who Is Right And Why Do They Think That?
• Elon Musk Declares Precision Maps A “Really Bad Idea” – Here’s Why Others Disagree
• Tesla’s “Shadow” Testing Offers A Useful Advantage On The Biggest Problem In Robocars

Debate over the Tesla approach will continue. 2020 could see the so-called fully autonomous driving. But it is not such, it is rather a version of the autopilot, which can drive along city streets and turn from one to the other in the presence of a driver watching the road. In fact, Tesla has released demo videos. It will be interesting to see how drivers react, whether they are relaxed or terrified, and how much they rely on autopilot.

US National Transportation Safety Board (NTSB) and Uber Errors

Uber’s failures sparked a lot of discussion in 2018, and they still haven’t subsided. The NTSB released its findings and held hearings on them:

• New NTSB Reports On Uber Fatality Reveal Major Errors By Uber
• NTSB Hearing Blames Humans, Software And Policy For Fatal Uber Robocar Crash – But Mostly Humans

The conclusions of the board of directors coincide with my assumptions, but a few interesting facts can be added. They did not blame Uber for the numerous errors in the software, no matter how bad they were, because the system was designed to predict these errors, and there was a driver who monitored security and took control if a failure occurred. Uber did a tremendous job of developing its safety culture, as well as hiring, training, and monitoring its drivers, and that night the driver watched a video on the phone instead of looking at the road, leaving this situation to a car that it should have been easy to to cope, but it ended tragically.

One noteworthy thing in the conclusions was focused on a too strong reliance on automaticity – people quickly get bored with automated systems that only occasionally require attention, and they completely stop paying attention to them. Uber didn’t take that into account at all. Some manufacturers have a camera that monitors the gaze of the driver. Tesla requires drivers to touch the steering wheel periodically to confirm their attention. Regulators are expected to follow the NTSB and begin to insist on more effective countermeasures to track driver actions during testing and in driver assistance mode.

Challenges continue

In 2020, manufacturers expect the following challenges, roughly in order:

• Testing: As always, proving that your system is safe — for you, your lawyers, the public, and ultimately the government — remains the number one challenge.
• Prediction: you need to better predict what others are doing on the road (or next to it).
• Perception: this is especially true for those who rely on computer vision without going around the right one on the road. For those who rely on lidar, one must learn to better deal with obstacles in the distance and small debris on the road.
• Sensors: Many companies are striving to fulfill the promise of a relatively inexpensive and reliable lidar.
• Regulation: it is necessary to open more locations for unmanned operation. ADAS and supervised systems may face demands for defeating drivers inattention.

Read more useful articles:

• Cameras or lasers
• Autonomous cars on open source
• McKinsey: Rethinking Software and Electronics Architecture in Automotive
• Another OS war is already under the hood of cars
• Program code in the car
• In a modern car, there are more lines of code than …