Driverless Features Will Be An Evolution, Robotaxis A Revolution

Are driverless cars a revolution? They are not. Like many innovations that preceded them, they are merely improvements to the existing state of the art.

Usually, technological improvements happen in the context of large companies, who leverage their balance sheets, brands, market reach and economies of scale to experiment with new products. Sony’s Walkman, Nintendo’s Gameboy, and the Toyota Prius come to mind. Startups have the flexibility to innovate not only on the product, but also through the business model through which it will be offered. Tesla’s electric vehicles and accompanying charging system, Uber’s two-sided driver/passenger marketplace, and Nervana’s deep learning accelerators and accompanying deep learning framework, were better suited as startup efforts.

Building autonomous cars has required recruiting a new type of engineer in a new type of organization. GM acquired Cruise for the prototype it had built. Zoox and Aurora are building the blueprints for the vehicle, as well as the fleet operations to manage them.

Thirty years ago, the oil crisis almost killed many car companies. Competition has only gotten more intense, and manufacturers know that to survive, they must offer self-driving features. Luckily, car companies and their suppliers have invested heavily in self-driving technology, and some have the luxury of choosing the markets in which it will be offered. Audi, for example, doesn’t offer many of the Level 2 automation capabilities in its North American-bound A6, to direct interested buyers toward its pricier flagship A8 model. It’s only a matter of time that top Tier-1 suppliers such as Bosch, Continental and Aptiv, will offer Level 2 self-driving features as a box a manufacturer can check in a spec sheet, as they would LCD instrument displays.

Automakers now have their sights set on full-autonomy and are still figuring out their place in the autonomous future. Do they offer transportation services? Or do they partner with the likes of an Uber or Lyft, relying on technology from Aurora or Waymo? Meanwhile, they know that they will be selling millions of Level 2 capable cars.

Autonomous cars are maturing from excitement-inducing novelty to an expectation, or dare I say, an entitlement. Billions have been invested, ink has been spilled, and noise has been made, and the closest thing we have to autonomy is Level 2 capability in our cars and big promises from autonomous car CEOs whose companies are sitting at lofty valuations. As evidenced by Drive.ai’s struggle to raise money and subsequent sale to Apple, talent in the space has become abundant. Top engineers are flocking to computer vision and universities are churning out record numbers of roboticists. The challenge is no longer showing a proof of concept; the challenge is showing that you can build a profitable business. Investors are saying “show me the money” and passengers are saying “get us around cheaper and safer!”

The mood at the leading autonomous car companies has shifted from one of experimentation to one of delivering product. Leadership is less interested in basic scientists and more interested in engineers that can reliably deploy and predict the performance of autonomous cars. Rather than build dazzling demos, efforts are concentrated on extreme corner cases. At Zoox, where my firm Lux is an investor, vehicle design has matured to crash testing and homologation. Software developed in well-understood geofences is being thrown into the wild in incomplete geofences and crude maps to test the technology at its limits. This is reminiscent of my days at General Motors where more time was spent validating engineering specifications, testing technology under extreme circumstances, and understanding mechanisms of failure.

Whether you are an investor, engineer or executive at an autonomous vehicle company, be aware that the basic technology has rapidly transitioned from scarcity and is close to becoming mainstream. The most important players in autonomy are looking for talent to take them past the finish line, not help them build an autonomous car. For Level 2, the problem has been solved. Level 3, which is probably the topic of another article, may never happen. Level 4 requires a new business model and a level of predictability to determine geofences, availability and support to help an operator generate margin by deploying robots as opposed to recruiting human drivers. 

Think about techniques used in the military, airline industry and space missions to predict risk on missions involving many known and unknown unknowns. The pioneers of the autonomous revolution may have been the roboticists, but a very different group of engineers and experts working behind the scenes will likely be responsible for their success.