This is the 2nd post in a series about self-driving vehicles and it explores how autonomous cars could become a reality. Self-Driving Vehicles: The Future Always Takes Longer to Arrive is the 1st post and covers the state of the vehicle autonomy (circa mid 2016) and how we’ve gotten here.
So what are the different paths towards commercially available self-driving cars? The way forward includes not only advanced vehicles themselves but also potentially shifts in road infrastructure, laws and regulations, and even business models for “mobility.”
In my first post earlier this summer, I highlighted the fact that we’re still a ways off from truly autonomous vehicles, despite many decades of technological advances to assist drivers. The “Auto-Pilot” capability in Tesla’s Model S is currently the most advanced semi-autonomous (NHTSA Level 2) system you can actually buy, but it still has many limitations and to me, it’s closer to “awesome cruise control” than it is to a “self-driving” experience. If you go on YouTube, you can find examples of lots of antics involving Tesla’s Auto-Pilot, but in reality, driver involvement is still an absolute must with this system.
Sadly, news came out shortly after I published my first post regarding the first fatality involving a Tesla in “Auto-Pilot” mode. While all traffic fatalities are tragic of course, I’m hopeful that events like these will not be a setback for technical and regulatory efforts to bring truly autonomous vehicles to market. It’s also a reminder that, for the time being, driver attention is required even as more advanced aids or semi-autonomous systems are developed (all categorized as ADAS – advanced driver assistance systems). Tesla has already released updates to its Auto-Pilot software to further refine situations like this.
FWIW, the first true competitor to Tesla’s Auto-Pilot ADAS system, Mercedes “DrivePilot”, has gotten terrible reviews. Despite its limitations, Tesla’s system remains the clear best in terms of performance and driver comprehension. And semi-autonomous/ADAS has applicability beyond passenger vehicles, as we have seen with companies like Otto focused on retrofit ADAS for semi trucks (Otto is being acquired by Uber).
There’s no one certain path towards autonomous cars. But the way forward will undoubtedly require meaningful changes to some combination of the following:
The most obvious area which requires change is in the cars themselves. For vehicles to be truly autonomous, they need to be able to perceive their environment in great detail, map their route, make decisions about what to do, and coordinate their actions at some level.
Arguably the biggest advancements in recent years have been in the area of perception. Self-driving cars currently in development rely on some combination of laser / LIDAR, radar, ultrasonic proximity detectors, and cameras / computer vision to perceive their surroundings. At their most basic, these systems help a car determine distance and speed relative to other objects (surrounding traffic, pedestrians, highway guardrail, etc.) or points of reference (painted lane markers on roads). In their more advanced forms, they can make determinations about the nature of an object in the surrounding environment (Is that thing 100 feet away a car, a large truck, a pedestrian, or a building?) and interpret context or meaning (the red octagon with lettering 50 feet away is probably a stop sign).
Many of the sensors and systems for vehicle perception are made by large Tier 1 suppliers like Bosch and Delphi, but new players like Mobileye (founded in 1999 and now a $10 billion market cap public company) have also emerged. There are also significant startups like Velodyne, Quanergy, and others that are seeking to become major component suppliers of various sensor technologies.
Perception isn’t the only area of vehicle development that will require significant advances. There are also significant efforts in AI / decision-making systems, improved mapping, vehicular coordination systems, and other technologies which will be needed to enable level 4 autonomy. These are all mutually dependent. For example, if autonomous vehicles can effectively coordinate their actions (e.g. “convoying” with other nearby cars or detecting transponders of other vehicles), then theoretically they could be slightly less sophisticated in terms of their individual ability to perceive the environment.
Various existing and aspiring vehicle manufacturers are taking different paths towards autonomy. Some like Tesla, GM, Mercedes, Volvo (both on its own and in partnership with Uber), and others are developing ADAS features that will enable level 2 or level 3 autonomy first, presumably as a stepping stone to fully autonomous vehicles. Others like Google and Ford have talked publicly about a strategy of focusing on level 4 autonomy completely, without intermediary development of ADAS.
The $500+ billion dollar question* is of course this: When will fully autonomous vehicles be available for use by the general public? The honest answer is nobody, even vehicle OEMs, really knows for certain. Part of the problem is that, for all the focus on the technology and vehicle development, autonomy also requires fairly meaningful changes to road infrastructure, laws and regulations, and other non-technical matters…
Road Infrastructure / City Planning
Most of the focus in terms of press, investment, or consumer enthusiasm is currently focused on vehicle technology development. But the pathway to fully autonomous cars and trucks will almost certainly require substantial changes to our road infrastructure and city planning.
If you think about a road in the broadest sense, it’s in fact a system that encompasses the tarmac surface itself, lines or pavement markers (raised reflectors, Botts’ dots, etc.), various signs conveying context, traffic control mechanisms (stop lights, etc.), and more. And this entire system was engineered for human-driven vehicles. If one were starting from a blank sheet of paper designing for a world with autonomous cars, you probably wouldn’t have most of this stuff. Theoretically, an autonomous vehicle could be programmed to stop at any intersection if the car detected cross traffic. The intersection would be conveyed by GPS map or “smart” road transponders, while the detection of traffic could be handled via computer vision, vehicle coordination sensors, or several other viable options.
Additionally, designing from scratch for autonomous vehicles would mean that, instead of a series of hanging lightbulbs, an intersection could feature a sensor to detect approaching vehicles and a transmitter to send instructions to each vehicle: “slow,” “stop,” “maintain speed,” “safe to turn left,” and so on.
Of course, in reality, we’re not going to rip up the millions of miles of existing road infrastructure and start with a new system of “smart” roads embedded with sensors, transponders, and such. But in places where new development is taking place, you may see “smart” road infrastructure deployed instead of conventional roads (just as you saw some developing areas “skip” copper telephone lines in favor of wireless telecoms).
In all of this, there’s also a key role to play for city planning in enabling vehicle autonomy. We will likely see certain stretches of road or certain zones within a municipality created to facilitate various forms of autonomous vehicles even before we truly have “anytime/anywhere” full autonomy.
When it comes to vehicle autonomy, we are and will remain in a state of transition for many, many years. I believe you will see more widespread adoption of semi-autonomous / ADAS features in cars over the next 0-5 years, but it’s likely true that anytime/anywhere autonomy is much further out into the future.
While history doesn’t repeat, it does rhyme, and we forget that cars and horses coexisted on public roads for several decades in the early 20th century. So even when fully autonomous cars finally become commercially available, they will likely have to coexist with human driven vehicles for some time. Even then, self-driven cars may only be usable in autonomous mode under certain circumstances based on road situations, local laws, weather conditions, etc.
There is indeed a pathway forward where a fully autonomous vehicle could be created just through tech development and advancements. But without other system-level changes like road infrastructure, vehicle coordination, and laws and regulations following suit, it could be a much steeper path.
* Sales of new cars and light trucks in the US were approximately $576 billion in 2015.
In my third post, I will tackle how the legal framework, insurance, and business models for mobility will change with the rise of autonomous vehicles. The best way to keep up with this series is to follow NextView’s Medium publication, Startup Traction, here or to all our content via email here.