MicroRaptor Navigation

People do not keep precise 3D maps of the areas they move around in. We do not have laser range finders as a sensory input.

We don't use precise odometry to move around, although in some limited cases (usually in the dark) we use a form of odometry to move around inside our houses.

Mainly, people do landmark based navigation. Lankmarks can include many things, like trees, doorways, fences and gates, boundaries (roads & paths), and lots of stuff like that. Landmarks can also include ground/floor cover (grass, pavement, gravel, wood, concrete, etc).

Of course, landmark based navigation can't be done the way traditional sensory input is done in a robot. We typically are looking for things on several different levels, and using sensory input for more than just navigation. We navigate, we avoid things, we maintain balance while walking, we notice things in our environment, etc.

Maps are another interesting topic. We don't keep perfectly scaled raster maps of our environment in our heads. MicroRaptor will build and keep vector-based hierarchical maps. Those maps will contain connection information, mainly of the type this area and that area are connected.

Since the map is hierarchical, the nodes have parents, and each parent node typically has a number of children. The top level node might be my property, and subnodes of that node could be the house, the barn, the fenced-in yard, the front yard, and the driveway.

As long as the robot knows where it is when it is switched on, it can navigate from one place to another in the world that it knows.

Like most other things, maps will be stored in long term memory, and navigation will be handled with a navigation engine, which will be instantiated into one of the short-term memory slots.