Over the last decade, most of the buzz around self-driving has focused on the vision originally launched by Google (now Waymo) and later Tesla and others, of a Level 5 (fully, really, truly, autonomous no matter what) personal automobile that you could afford and blissfully relax in the back seat of from then on. You could even send it to fetch your children from school or tell it to wander around waiting for you or maybe even find itself a parking space. More recently, it has become clear that Level 5 personal autos are going to take much longer and be much harder to create, than the initial rosy projections. At CES 2019 last week, this was reinforced by the way vendors moved from pie-in-the-sky promises at previous CES shows to a much more step-by-step, practical approach this year.
Level 5 Is Getting Further Away, Not Closer
Last year, in particular, was chock full of warning signs and pullbacks for autonomous vehicle efforts. Most famously, a number of high-profile accidents (aka crashes) on the part of Uber and Tesla cars have called into question whether some of the companies involved actually know what they’re doing. Along with that, Waymo, the undisputed leader in progress towards Level 5, met its self-imposed deadline of a commercial robo-taxi launch in 2018 only with the very softest possible launch to a small number of testers in restricted environments with limited routes. Tesla pulled the “Full Self-Driving” option from its order page, citing that it was causing confusion (if you mean confusion that something you ordered years ago may never actually be available for your specific vehicle, I guess).
However, on the bright side, there have been some positive developments in the autonomous vehicle space that point to likely near-term use cases and winners. Three of these stood out to me.
Aptiv: There’s a Lot of Money to Be Made In Levels 2, 3, and 4
One problem for nearly everyone aiming to produce a Level 5 vehicle is that it’s a total money sink. Unless you are Google and essentially print profits, or Tesla and Uber, who seem to be able to raise whatever money they need so far, that makes it a very expensive proposition with no near-term payoff. But for industry-leading automotive supplier Aptiv (formerly Delphi) the march from Level 2 through Level 5 is simply the natural progression of its automotive electronics business. They have laid out an advanced architecture that they believe will pave the way for car companies to get started with the process by integrating various existing driver assistance systems and then progressively improving them over time.
As a compelling demonstration of how this is working, I got a test ride in one of the cars Aptiv has provided to the Lyft fleet in Las Vegas. The car itself was similar to the one they were using last year, but the system had been upgraded to provide improved functionality. For example, implementing RTK (Real-Time Kinematic GPS augmentation) has allowed the cars to locate themselves within 2.5 cm (instead of 10 cm). That makes the difference between not knowing and knowing whether a pedestrian is standing on the edge of the curb or in the crosswalk. Impressively, but not surprisingly, in the several mile test drive I took, no operator intervention was required. That included dodging buses, pedestrians, and making unprotected U-turns on a busy 6-lane street.
Nvidia also made a major move in this direction. Previously it had primarily touted how its high-end Pegasus computer was perfect for Level 5 projects. Perhaps realizing that any type of volume sales for Level 5 or even Level 4 are likely to be slow in coming, Nvidia launched a Level 2+ solution, DRIVE AutoPilot, based on the less-expensive Xavier SoC — aiming to capture the large and growing volume of vehicles that have various driver assist and automated safety systems.
Deere: “Hey, We Did Autonomous Before People Knew What It Was”
John Deere has been putting a variety of autonomous driving technologies into its tractors for around 15 years. We got to ride in the current version, and not only is the technology impressive, but it has a no-brainer use case. Many farmers spend 12-16 hours a day driving their tractors up and down rows of crops. Any mistake and plants are crushed. As our Deere test driver pointed out, this is like being told to drive your car down the white line on the edge of a road for a solid day and having it cost you $ 3 for every time you deviated. RTK and camera-enabled, Deere’s top-of-the-line tractors can do the job with 2.5 cm (1 inch) accuracy.
Deere has also implemented a clever “route memorization” feature. Because not all crop fields are nice, straight lines of crops, the driver can maneuver through their field once, and then the tractor can repeat the task when needed. This allows the farmer to perform other essential tasks, like controlling whatever farm implement is attached. If the right way to proceed is to move slightly to one side to avoid driving on top of previously planted seeds, for example, the route can also easily be nudged to one side or the other. We got to witness this for ourselves as the massive tractor we were riding in retraced our steps perfectly through a windy route.
There are many other similar use cases like large scale industrial facilities and mines. They have an immediate payback and are feasible with today’s technology. They just aren’t quite as attention-getting as the promise of personal self-driving cars.
Robo Taxis Are Happening, Just Not the Way We Thought
A number of companies have pared down the Level 5 challenge to make it more tractable. Some are limiting their locations to easier to navigate areas, like Voyage’s effort based in retirement communities. Others are looking at applications that don’t involve moving humans, like grocery delivery robot company Nuro, which has teamed up with Kroger to pilot autonomous delivery. Finally, other companies like Aptiv (with Lyft) and Cruise (part of GM) are tackling the broader challenge of Level 4, but with some operational constraints. They’ve limited their vehicles to areas that are accurately mapped and monitored, and where they can get needed infrastructure updates. They also don’t have the same cost constraints, as fleet vehicles have much higher utilization, and can, therefore, be much more expensive, than typical personal vehicles. They can monitor their vehicles from an operation center, and intervene as needed — something an individual can’t do if they send a personal vehicle off on an errand.
Robotaxis also provide a window into the likely future for transportation. Most industry experts argue that by the time we get close to Level 5 most people will not own cars, but instead will simply subscribe to a “mobility service.” Obviously, driving enthusiasts and those who live in rural areas will be exceptions, but for anyone in a reasonably dense area who doesn’t relish doing their own driving, or simply want their own, specialized vehicle, they probably won’t own a car at all.
Not everyone agrees with this prognosis. Zoox, for example, believes it can design, build, and sell a Level 5 car itself, and in a lot less time than most of the rest think is possible. We’ll know soon enough.