• View
• Edit
• History
• Print

LocalNavigation

Overview

The term local navigation is used to refer to a simple motion control algorithm that will move a robot from point A to B without the benefit of state based search. So, obstacle avoidance is often expected but not the ability to avoid local minima. This delineation is descendant from the Three Level architecture. I will discuss it here in this context, but this separation of responsibility proves useful in a variety of architectures.

The local navigator typically accepts way-points from a Path Planner, and is expected to go to each one in sequence while avoid obstacles. In this way, the path planner can set a course at a higher level of granularity (often necessary to make the problem tractable).

Examples

There are many local navigation algorithms, and they may be specialize to specific robots and / or control mechanisms. The examples here are biased towards two wheeled robots.

Artificial Potentials

This method assigns repulsive forces to detected obstacles and attractive forces to goals. The robot's desired velocity is calculated as a vector with all goals and obstacles added to it (usually normalized). This vector is then converted into control outputs.

An excellent tutorial has been provided by TODO: link to that tutorial paper TODO: Maybe link to your own code for this one?

VFH

An improvement upon Artificial Potentials, the Vector Field Histogram is an efficient representation that incorporates previously sighted obstacles into the algorithm for a more robust navigator. TODO: Possibly link to the Player / Stage implementation

TODO: More examples?

Counter Example

How does RRT fit into this model that I am laying out? Does this model actually fit with what real systems are implementing? TODO: Expand