Given the initial pose of the bot (say P1) and a goal pose (say P2), the global planner creates a path (global plan) from P1 to P2. This path is essentially an array (vector) of discrete points (poses) that can be followed by the bot to reach the goal (a.k.a P2).

The local planner computes the control inputs that need to be sent to the bot in order for it to keep tracking the global plan (provided by the global planner) while avoiding collisions with (dynamic) obstacles.

I hope that makes sense.

These two components do different things, and part of the confusion comes from the poor naming:

1. The global_planner really is a planner. Inputs are a start pose and an end pose. The output is a path.
2. The local_planner is really a controller (ROS2 actually properly renamed this!).The inputs are a path and the current robot pose. The output is base velocities (the controls for the base).

Better check out this page: http://wiki.ros.org/dwa_local_planner And the source code for DWA: https://github.com/ros-planning/navigation/blob/noetic-devel/dwa_local_planner/src/dwa_planner.cpp

A simple pipeline should be: A* send a path to DWA. DWA will generate a series of trajectories according to the params, velocities, vx, vy, vtheta samples, sim_time, etc.. then calculate the cost for each trajectory according to path_cost, goal_cost, occ_cost. In the end, find out the best trajectory.

DWA initializes with local costmap. local costmap contains map and sensor information.

Yep, I do have same question. Due to the original dwa take the breaking distance into accout. But the ros dwa don`t have this function.