Autonomous Vehicles Are Quickly Improving

Despite two fatal accidents involving (partially) autonomous cars this year, many companies continue to work on what most experts consider the car of the future. The problem with autonomous vehicles is that they cannot be fully developed in a lab. Artificial intelligence and machine learning are fueled by huge amounts of data, which is why for self-driving cars to improve, they need to be tested on public roads in order learn from real-life situations.

Several U.S. states have programs in place that let companies test their autonomous driving technology on public roads, with California, home to Silicon Valley’s tech elite, an obvious forerunner. In 2017, autonomous test vehicles completed more than half a million miles in self-driving mode on Californian roads, with Waymo (formerly Google’s self-driving car project) and Cruise (a GM subsidiary) the busiest test drivers.

Based on a Quartz analysis of data from mandatory reports filed with the DMV, all that testing is bearing fruit. As the following chart shows, the frequency of human interventions, i.e. situations in which the human driver decided or was forced to take control, was significantly lower in the second half of the year compared to the first. Waymo’s cars for example drove an average of 4,847 miles between two human interventions in the first six months of the year. Between July and November, the company’s backup drivers only had to step in every 7,527 miles.