I am currently enrolled in Sebastian Thrun's robot car course CS373. He's the Stanford professor and Google Fellow that headed the group that WON the DARPA Grand Challenge. My understanding, from taking this course, is that their self-driving car is not only able to navavigate to a goal-destination in unfamiliar territory (as in the Grand Challenge, which took place in a desert), but the car is able to identify urban obstacles: crosswalks, stop signs, traffic lights, and can also predict the motion of potential obstacles around it (i.e. cars and pedestrians). The robot car uses controllers with statistical models, so it is able to identify the probability of an obstacle entering the trajectory of the vehicle and respond accordingly -- slowing down like you would in that situation.
Watch some videos of the Stanford car.
Here's the class at Udacity if you're interested.