Preface: As a consumer, I am excited by the prospect of autonomous vehicles and the individual and collective benefits they can provide. As an investor, I am also bullish about the innovation wave already in its early innings around mobility and autonomy. But a realistic understanding of the potential timeline for autonomous vehicle adoption is vital to decision making as a startup entrepreneur or investor in this area.
Rome wasn’t built in a day. In fact it wasn’t even built in a year or a decade, but rather a couple centuries. Autonomous vehicles will become widespread a lot faster than that, but it always takes longer than you think for the future to arrive. Fully autonomous vehicles (aka “Level 4” autonomy in NHTSA guidelines, where a car can navigate itself from A to B without any human involvement) won’t be generally available for “awhile”. IMO “awhile” is a minimum of 5 years from today, and quite probably a decade or more away.
It isn’t because of lack of interest or investment in vehicle autonomy. GM’s acquisition of Cruise Automation for $1 billion about 3 months ago has been a sort of Sputnik moment for self-driving cars. In the 90 days since this deal was announced, car OEMs like GM, Ford, VW, Toyota, and Fiat Chrysler as well as tech giants like Google and Apple and on-demand transportation companies like Uber, Lyft, Didi Chuxing, and Gett all scrambled to form strategic/investment partnerships with each other.
That said, there’s no way you will be able to buy / rent / hail a fully autonomous vehicle in the next couple of years. Level 4 autonomy on public roads is a nearly unbounded problem… driving on a highway at 60+ mph or being able to self-park is a totally different challenge than navigating a poorly marked, chaotic urban downtown. Autonomous vehicles that work anytime, anywhere, and in any conditions are still quite a ways off.
So where are we today? How might we get to fully autonomous vehicles in the coming decades? And what are the broad economic and societal impacts of autonomous vehicles? I am a technologist, a VC investor, and a car enthusiast so this is a subject that I’ve tried to understand from a variety of perspectives over the last year or two. In this first post I will try to tackle the current state of play and how we got here.
Driver Aids: 50 Years In the Making
People often don’t realize or take it for granted that the “machines” have been aiding drivers for some time. Cruise control was invented in the late 1940’s and was first adopted on a production car in 1958, the Chrysler Imperial. In fact it was marketed as “Auto-Pilot” and a Popular Science article from 1958 about the car carried the headline “Like it or not, the robots are slowly taking over a driver’s chores. The newest one is a power-operated accelerator pedal…”.
In more recent decades, additional technology has come to market to assist drivers. All of the following systems involve software / hardware to inform drivers or physically take action on a car’s throttle, brake, or steering systems.
- Anti-lock brakes –> again debuting on Chrysler Imperials in 1971
- Adaptive cruise control –> Mitsubishi & Toyota developed LIDAR systems in early ’90s, Mercedes first launched radar systems in 1999)
- GPS navigation –> pioneered by Mazda and Toyota in early ’90s, Toyota also had first CD-ROM based map system
- Lane departure warning –> first deployed by Mitsubishi in 1992 and broadly available in early 2000s from Japanese manufactuerers
- Collision avoidance systems (warning / automatic braking) –> demonstrated by Mercedes in 2003 as well as by Honda for Japanese market and Acura brand
- Electronic traction / stability control –> Toyota, Mercedes, and BMW all developed first systems in late ’80s
Like pretty much all innovation in cars, these features were only available in the most expensive luxury models initially. Today they are commonplace through a combination of consumer adoption and regulatory standards. My wife’s 2013 Ford Explorer, pretty much the quintessential mass market car for a suburban family like ours, literally has all of the features above.
But all of these capabilities merely assist a driver on an occasional basis during a given journey. They provide very significant safety and convenience benefits, but in the hierarchy of autonomy where Level 0 is a completely “dumb” car (functionally equivalent to nearly all cars produced 30+ years ago) and Level 4 is full autonomy, then each of the driver aids above is considered Level 1.
DARPA’s Robotic Cars to Today
DARPA has held a series of challenges in the last 10+ years involving autonomous / robotic vehicles. The first one took place in 2004 and was conducted in the Mojave Desert. None of the entrants could complete the 150 mile course, in fact the best vehicle (developed by CMU team) only went about 7 miles before it got stuck like the others. Subsequent challenges in 2005 and 2007 had better results, and the 2007 event was the first actually staged on roads… albeit an abandoned military base in California. But these challenges certainly sparked interest in vehicle autonomy.
The Tesla Model S auto-pilot system, which was released about 8 months ago (Oct 2015), is pretty much the first Level 2 autonomy available in the US. It combines adaptive cruise control with lane departure to enable a car to maintain speed and safe distance from cars in front, while also controlling steering input to keep the car centered in the lane. Other OEMs are preparing to release similar systems to market in the near future, but Tesla is the first both because they’ve led in technical innovation and frankly as small/new entrant to the global car market they’re willing to take more risk than incumbents.
But even these systems require a driver to continuously monitor the situation and be prepared to intervene at any time. They utilize a combination of laser, radar, and computer vision to assess the spacing of a car relative to surrounding traffic and painted lines on the road. Tesla’s auto-pilot is a remarkable achievement and a great convenience… I am fortunate to drive a Model S. But even in the somewhat controlled environment of highway driving (e.g. less unpredictability from intersections, traffic lights, pedestrians, etc), the system requires human intervention reasonably frequently. If there’s construction somewhere on the highway along your trip, if the lines on the highway have a gap (e.g. the right hand lane when approaching an exit), if vehicles around you veer unpredictably, then Tesla’s system will cut out and require the driver to take over with only a second or two notice.
Google, Ford, Volvo and others are experimenting with Level 3 autonomy where a vehicle can drive a portion of the time without a human driver’s input. None of these systems can work outside a fairly controlled set of conditions in terms of weather, speed, or geographic scope so there not truly go anywhere, anytime sorts of vehicles. And no one is that close to achieving true Level 4 autonomy.
So How Will We Get to Driverless Cars?
To be fair, the potential for autonomous or semi-autonomous vehicles has advanced tremendously in the last 5-10 years. The teams out of top engineering schools like Stanford, CMU, MIT, Virginia Tech, and others that competed in the DARPA Challenges in 2004 / 2005 / 2007 now make up the backbone of the R&D teams for most of the commercial efforts in this space. And obviously we talked about how the GM acquisition of Cruise was this Sputnik moment.
But the underlying technology for perceiving the environment, mapping routes, and making decisions about unexpected situations all have a lot more work that needs to be done. There are also many steps on the regulatory, business, and road infrastructure fronts that will need to be accomplished before we truly have autonomous cars. My next post will touch on various paths on all of these fronts which could get us there.