ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange

# What is the difference between global planning and local planning?

Do I understand correctly that at the global level we plan a path from the current pose to the desired. And at the local level - moving towards the desired point with detecting obstacles? Sorry for my English

edit retag close merge delete

Sort by ยป oldest newest most voted

These terms are vague and some research communities claim/assume that one or the other is enough...

Usually the idea is that global planning sees "the whole picture"/current world state/current assumed world state and local planning only has to work in the close vicinity of the start point.

Some projects leave the handling of dynamic changes in the environment to a "local planner" that can adjust the plan produced by a global planner in an online fashion.

My usual (over-the-top) analogy is:

• Solving Rubik's cube from any given state is a global planning problem.
• Turning a single side might work with a local planner (depends a lot on the setup)
more

does obstacle avoidance happen in the global or local planner? I have noticed that the global plan can be updated when there are new obstacles and it seems that the role of the local planner is merely to follow the global plan as close as possible...?

( 2018-07-04 17:03:56 -0600 )edit

The global planner plans a global path around obstacles and any new obstacles based on a frequency specified by the planner_frequency parameter. The obstacle avoidance (collision check) happens in the local planner where the cmd_vel is produced and is based on the controller_frequency parameter. The local planner tries the follow the global plan as close as possible meaning that it takes into consideration a part of the global planner at a time

( 2019-08-02 08:26:28 -0600 )edit

A two step planning approach of global planning and local planning (also called as Hierarchical Planning) is explained well in Spline-based RRT Path Planner, The authors state that :

• In the global planning stage we try to find a collision free kinematically feasible path from start to goal while skipping the differential or dynamic constraints (so this is where obstacle avoidance should be happening)
• In the local planning stage, we use path smoothing to meet the differential/dynamic constraints
more