Uber’s program is in the early days of development. Michael Wagner, a robotics researcher at Carnegie Mellon University, told me that building a successful autonomous fleet—one that handles unexpected driving situations reliably and safely—takes an enormous amount of technical work. Wagner runs a company that tests the robustness of complicated software, including in consumer electronics and aerospace systems. “You don’t always know you’ve built it until you’ve gone out and done a whole lot of testing on it,” he said. In Uber’s case, the testing has been happening on Pittsburgh’s streets. For the past several months, fourteen test cars have been making daylong trips around the city, collecting information about the surrounding environment as they go. (According to Uber, one three-hundred-and-sixty-degree sensor generates more than a million data points per second.)