Google taught its self-driving car how to make three-point turns

Google hasn't thought of everything, but it's getting pretty darn close.

Instead of choosing the tight line to turn around, Google's car takes a wider arc that's more natural to humans.


Every month, Google releases a report on its self-driving cars and gives us some insight into how the learning process works. This month, it's talking about three-point turns.

Three-point turns are an art, especially in tight spaces that may turn three points into seven or eight. Google's autonomous tech can best humans in this regard, thanks to its 360-degree vision and constant calculation. The cars were taught to recognize everything from parked cars to trash bins when making three-point turns. Google claims its cars practice about 1,000 turns of this nature every week.

Since a computer might calculate that the most efficient turn is some weird menagerie of adjustments, Google also programmed its cars to try and mimic the natural feeling of how a human might execute a similar turn. As a result, the cars prefer wider arcs with more forward motion, which may be slower, but is more comfortable for passengers.

Google currently operates 24 Lexus RX crossovers and 34 of its own gumdrop-style cars across four states -- Washington, California, Arizona and Texas. To date, the cars have driven 2.2 million miles in autonomous mode, averaging about 25,000 miles per week.

Google also uses its monthly report to chart collisions. This past month, there was one. While at a stop sign, yielding to traffic, one of Google's gumdrop cars was rear-ended by a human driver at a net velocity of 3 miles per hour. The gumdrop had a bit of damage on the hatch, but everything was fine otherwise.

Autoplay: ON Autoplay: OFF