Google’s self-driving cars: Implications for the auto industry and the key role of machine perception

Google’s self driving car effort is a threat to the auto industry. The company is the clear leader in autonomous vehicle technology and several years ahead of all other auto makers, including Daimler and Volvo. By presenting an all-electric prototype of a fully autonomous two-seater in May, Google has also made clear that it is serious to become a player in individual mobility and intent on reaping the rewards of its investment in this project (which so far has likely cost a few hundred million Dollars – not an enormous amount by the standards of the auto industry for developing a new car model).

What are the implications for the auto industry? They have much more experience in all aspects of mobility and are also working on autonomous vehicles. Could Google really be a signficant threat?

The standard answer to this question has been denial: Last year the main argument was something like: They may be able to build great software but they don’t know how to design a car. Now that they have designed a steer-by-wire two-seater with redundant layout of all safety-critical components and skillfully navigated the regulations – including limiting the speed to 25mph – , the argument is updated: They may be able to build a slow-moving two seater, but they can’t build a real car. And even if they could, they could not produce it in any meaningful volume.

As they overcome each objection, denial becomes harder, and additional time is lost. The argument that Google would not be able ramp up production is misguided. Google has no intention to challenge the auto makers on their playing field. It will change the game by providing autonomous mobility services rather than selling cars. Each Google autonomous car will then reduce the demand for privately owned cars by a factor of 5 to 10. This will have an impact on auto makers. It will affect their strategies, stock prices and make production capacity much easier to acquire.

Instead of denial, auto makers need to understand the magnitude of the threat. Self-driving cars will be a disruptive force; they will change the business model of the auto industry and bring hard times to most auto makers because demand for passenger cars will fall significantly. From a global perspective this is a good thing because resources will be used much more efficiently, alternative propellants can be used much more readily within autonomous mobility services and the strain on the environment (both pollution and land-use) will fall.

But it will be hard for the auto industry to adapt to these changes. Cars have been produced for more than a century. The requisite knowledge is widely available. The same does not apply to a key ingredient for self-driving cars: Teaching a machine to perceive its environment. Perception is the core problem which determines the success of a self-driving car.

Perception is a multi-faceted problem. It has to do with sensors, with prior knowledge, machine learning and is sensitive to action and context. Unfortunately, perception is not limited to the context of driving. Self-driving cars need to understand the behavior of people and things that may be relevant to the driving context – even if their behavior has nothing to do with driving a car (e.g. kicking a ball).

Because perception is hard, it requires considerable financial and human resources to solve the problem. Google has not only the financial resources but has recruited many leading experts in this field. Even the leading auto makers would find it hard to build teams that match Google’s expertise. Joining forces with other auto makers may be the only viable strategy.

Because perception is a general capability, it is applicable to many fields beyond driving and consequently it can generate returns in many fields besides driving. This is an advantage for Google because it allows cross-fertilization with its other business areas. Google has recently bought several leading robotics companies. Advances in perception by the self-driving car group could also benefit these business areas and vice versa. Google has also started a mobile phone project (Tango) which aims to use a high end Android mobile phone to create 3D maps of the environment in real time. Advances in this space may also be useful for the self-driving car project.

As a consequence, auto manufacturers who want to beat Google to a fully autonomous car, will need to carefully consider the additional opportunities which advanced perception could bring and determine how to integrate these opportunities into their strategy. Instead of narrowing the perception task to specific driving scenarios, auto makers should consider whether they could leverage their perception activities in additional ways.

Machine perception is the core competence for succeeding with autonomous cars. Auto makers need to give this capability top priority if they want to recover the ground already lost to Google.