This is the fourth article in a five-part series about how self-driving cars work. Visit the Back\Line self-driving car channel to read the full article, and subscribe to read the rest of the series, on computer vision, sensor fusion, and localization!
Path planning is like the brain of a self-driving car. It's how the vehicle makes decisions about how to move through the world. In our model of the self-driving car software stack, it comes after perception (how the vehicle understands the world) and localization (how the vehicle determines its position in the world).
The first layer of the stack is computer vision, which we explored in the first week of the series. Computer vision is how the vehicle uses camera images to understand its environment. Subsequently, we discussed the details of sensor fusion, which is how the vehicle uses other sensors, like radar and lidar, to augment that understanding of the world. Last week, we dove into localization, which is how a self-driving car determines where it is in the world, with extreme accuracy.
Next week, we will study control, which is how the vehicle actually adjusts the steering, throttle, and brake, to move through the world.
Path Planning
Path planning, sometimes called "motion planning", is really the brains of a self-driving car. This is the part of the vehicle stack that makes decisions about how to move through the world. The process has three key sub-components: prediction, behavior, and trajectory.
Here is an example of path planning from a student in Udacity's Self-Driving Car Engineer Nanodegree Program.
Follow us over to Back\Line to read the rest of this article!