addressalign-toparrow-leftarrow-rightbackbellblockcalendarcameraccwcheckchevron-downchevron-leftchevron-rightchevron-small-downchevron-small-leftchevron-small-rightchevron-small-upchevron-upcircle-with-checkcircle-with-crosscircle-with-pluscrossdots-three-verticaleditemptyheartexporteye-with-lineeyefacebookfolderfullheartglobegmailgooglegroupshelp-with-circleimageimagesinstagramFill 1linklocation-pinm-swarmSearchmailmessagesminusmoremuplabelShape 3 + Rectangle 1ShapeoutlookpersonJoin Group on CardStartprice-ribbonShapeShapeShapeShapeImported LayersImported LayersImported Layersshieldstartickettrashtriangle-downtriangle-uptwitteruserwarningyahoo

Let's navigate the partially observable world (AIMA ch. 4).

Last time, we showed how to implement a navigation world full of convex, disjoint polygons; now we'll navigate it (see exercise 4.11b).

The nature of the percepts has been underspecified; in order to implement the world, for instance, we'll have to make a decision as to whether:

  1. The goal is always given relative to the agent; or
  2. the goal is only given when it is visible.

Either model has implications for search and discovery strategies.

Join or login to comment.

  • Michael M.

    We can use the Euclidean Distance of each Vertex in the state as the heuristic function cost = distance(vertex, goal) = SQRT(SQUARE(vertex.x - goal.x) + SQUARE(vertex.y - goal.y)) for each vertex.

    In addition h(n) can be Euclidean distance in using the A* algorithm to solve the problem.

    One approach:

    1. Modeler generates the initial state.
    2. Initial state is passed to the solver.
    3. Solver creates an initial search node,
    4. Solver calls the expand function to create all possible child nodes.
    5. For each child node, the expand function calls the modeler to generate a new state.
    NOTE: The coordinate system does not change for the new states, they are still using the
    origin of the initial state. The searhc node cost is added by the expand function via the
    Euclidean distance.
    6. A* algorithm is run on the fringe set of nodes. The modeler is called interactively from A*
    when expanding child nodes.

    Your thoughts?

    February 5, 2013

  • Peter D.

    Pretty good: we had some demos and clarified 4.11.

    February 5, 2013

4 went

People in this
Meetup are also in:

Sign up

Meetup members, Log in

By clicking "Sign up" or "Sign up using Facebook", you confirm that you accept our Terms of Service & Privacy Policy