According to Toyota, “Due to the infrequent nature of data collection, maps generated in this manner are seldom updated, limiting their usefulness.” To that end, they will rely on information generated by existing vehicles, meaning that how you drive and your behavior could end up teaching self-driving cars how to drive.
The company hopes that by having more people map out the same routes repeatedly will allow them to keep their margin of error down to a maximum of 2-inches of a straight road. Of course this is just one way that self-driving cars could learn to drive better. As it stands there isn’t a perfect or unified system yet as many companies are still working it out.
For example it was recently that a new system was introduced that relied largely on images from preloaded maps to identify objects and obstacles on the road, as opposed to more expensive equipment like radars and lasers. We are still a fair bit away from self-driving cars being released commercially and legalized, but hopefully this is a step in the right direction.