Google's self-driving cars have racked up considerable mileage on public roads, and each mile yields data that can make them more capable.
While the tech giant has already taught a car to drive through an autocross course at speed, there are many real-world scenarios its systems must be prepared for before the car can get its driver's license.
The car can now identify several common roadside objects and correlate them with certain actions. That means an autonomous car can navigate around a construction zone and stop at railroad crossings. Where humans see train tracks, the car sees a red fence, telling it to stop unless it can make it all the way across safely.
Google's self-driving cars also know to yield to cyclists, and can even identify an outstretched arm as a signal that the rider intends to change lanes. The car can also keep track of multiple cyclists, pedestrians, and vehicles at intersections and yield right of way accordingly.
While Google is continually improving its autonomous cars, it's also making plans to offer the technology to consumers.
Last year, Google announced a partnership with IBM and Continental to develop self-driving car tech. Continental has done some self-driving car work of its own, while IBM's forays into the automotive field include a concept lithium-air battery that could dramatically increase the ranges of electric cars.
However, none of these companies manufacture cars themselves, so another party may have to brought into the group. On the other hand, Google has discussed building its own cars, which would be very interesting indeed.