Self Driving Car Engineer Nanodegree Program -- Path Planning 3.2 Search

3.2 Search

3.2.1 Motion Planning

  Motion planning cames in multiple segments, It starts out with a predictive task where the motion planner tries to understant what are other vehicles about to do in the future. Then the vehicle will pick a behavior kind of simply speaking, whether to left or right. Eventually, it generates a maneuver trajectory as to where to go. The trajectory is not just a curve in space in X and Y. It also has a time dimension because we have to set the speed of the vehicle at the same time.

3.2.2 Introduction to Path Planning

  Path planning is how the vehicle generates safe drivable trajectories to get where we want it to go. We’ll use data from compute vision and sensor fusion in order to understand the environment around us. We’ll also use data from localization to understand precisely where we are in that environment. The path planning block uses all of that data to decide which maneuver to take next. Then it constructs a trajectory for the controller to execute.
  Behavior planning is where we decide at a hight level what the car shall do in the next 10 seconds or so.
  Trajectory generation is where we create smooth, drivable, and collision-free trajectories for motion controller to follow.

3.2.4 Motion Planning

  The process of finding a path from a start location to a goal location is called “planning”,

Planning Problem

• Given: a map of world; a starting location; a goal location; some sort of cost function; The simplest way to think of cost is just the time it takes to drive a certain route.
• Goal: find the minumum cost path.

We can look at the path planning problem as a search problem.

The path planning or search problem is to find the shortest sequence of actions that leads out robot from the start state to the goal state.

3.2.14 A*

  A* is a variant of the search algorithm that’s more efficient that expecting every node.
  The mechanism for searching is gradually expanding nodes in the open list. A* is almost the same thing but not quite.
  A* uses a so called heuristic function, which is a function that has to be set up. If it’s all zeros then A* resorts back to the search algorithmn already implemented. We can the heuristic fucntion h, then for each cell it results into a value.

  In a world without obstacles the heuristic fucntion would actually measures the distance to the goal. So the heuristic fucntion has to be an optimistic guess how far we are from the goal, which means a smaller equal to the actual goal distance from the coordinate.

Heuristic function doesn’t have to be accurate. This is a function that helps you understand where to search next in the case of ties. It underestimates or at best equals the true distance from the goal.

3.2.17 Dynamic Programming

• Given: a map of world; a goal location;
• Goal: The best path from any possible starting location.

3.2.23 Planning Conclusion

• A*: uses a heuristic to find a path,
• Dynamic Programming: finds an entire policy that has a plan for every location.