But not every carmaker is going at the same speed. Toyota, one of only three car companies that sells over 10m vehicles a year, has made no equivalent commitments. The Japanese firm is instead concentrating on using artificial intelligence (AI) and automation to make conventional cars safer and more enjoyable to drive.
The immediate aim is to extend the age at which it is safe for older people to drive themselves, by using technology that can catch their mistakes. Software that processes data from on-board cameras and radar units will watch out for impending crashes and try to stop the car before impact, or correct for the slow out-of-lane swerve of a tired driver. Other software will guide the car in slow traffic, so that drivers can relax.
Helping older drivers is a particular concern in Toyota’s home market, where over a quarter of people are over 65. But similar demographic crunches are coming elsewhere. “Imagine a car, one day, that is so good that it will never be responsible for a crash, no matter what the driver does,” says Gill Pratt, chief executive of the Toyota Research Institute (TRI), the carmaker’s research hub in Silicon Valley.
This incremental approach will not necessarily leave Toyota in the dust. As the Uber crash showed, fully automated driving is difficult, and is progressing slowly, despite the billions being thrown at it. Rodney Brooks, a roboticist who sits on TRI’s advisory board, recently predicted that no unrestricted robotaxi service would arrive in a big American city until 2032. Toyota’s caution may let it avoid waves of self-driving hype and disappointment, while still giving it the tools to develop fully autonomous cars in future.
A slower approach also lets Toyota build the high cost of gathering driving data into its existing business. Before their cars can drive in a particular area, robotaxi firms must map it in exquisite detail, manually and at great cost, by driving mapping cars around the area they wish to service. Those valuable data are used to teach AI algorithms about human behaviour in the area, as well as about road layouts.
Toyota plans to gather similar data cheaply through its fleet of consumer-driven cars (by 2025 this will number some 50m cars). Outward-facing cameras and radars, now being installed in all its new cars to make them safer, will also gather on-board data that can be used to train fully autonomous driving software. Information gathered on such a large scale will allow Toyota’s AI to learn to handle traffic events that are extremely unusual, the sort which robotaxi firms gathering data in lesser quantities may never see.
Lack of “lidar” (light detection and ranging) sensors in Toyota cars could prove a hindrance, however. Lidar works by emitting pulses of laser light and watching for their reflections, thereby building a precise 3D map of the surroundings—essential for training today’s automated driving software, since video and radar do not capture the environment in sufficient detail. Robotaxi firms gather lidar data in every patch of city in which they deploy their cars, but Toyota will not, for the foreseeable future, be able to do so. The firm will either need to find a way to add expensive lidar sensors to the cars it sells, or to advance its machine-learning software to the point where it can learn to drive without it.
Toyota also needs to ensure that all the cars it sells have internet connections to transmit data in real time. A new arm, Toyota Connected, is aiming for that by 2020 in the firm’s two main markets, America and Japan.
All this adds up to a bet that massive scale and patience can beat being first to market. Toyota is not chasing the robotaxi dream directly. But it may nonetheless end up in the right place at the right time, and with the relevant data to cash in.