Home > marl > world > world_info.m

world_info

PURPOSE ^

Template. Generates a data structure with information about the world.

SYNOPSIS ^

function [info] = world_info(w)

DESCRIPTION ^

Template. Generates a data structure with information about the world.
  [INFO] = WORLD_INFO(W)
  Returns information about the world that could be useful to the agents.

  This function is a template and performs no operations.
  A new world type must implement this function under the name
  <world_type>_indexmaps. This functions' input and output argument(s)
  must conform to the specifications below.
 
  Parameters:
   W           - the world
  Returns:
   INFO        - the world information structure

  INFO must contain the following fields:

  - 'imaps' - an index maps matrix that specifies where in the world
  state vector each agent finds its own state and the common state of the
  world (i.e. state  elements independent of other agents), and where on
  the joint action and reward vectors the agent finds its own action and 
  reward. Agent n finds its index map on the nth column of the IMAPS matrix, 
  which has the structure:
     [AS1 AS2 CWS1 CWS2 AIX1 AIX2 RIX]', where:
       - STATE(AS1:AS2) represents the agent's state.
       - STATE(CWS1:CWS2) represents the common world state.
       - ACTION(AIX1:AIX2) is the agent's action in the joint action
       vector.
       - RIX is the agent's index in the joint reward vector.

   - 'stattespacesize' A matrix containing the sizes of the state 
   components, one row for each state component. The first column
   specifies the type of component, 1 for discrete, 2 for continuous
   For continuous components, columns 2-3 specify the ends of the
   interval inside which the state varies. The 4th column is a binary flag
   specifying if the state component "rolls over" from one interval end to
   the other. For components that do not roll over, infinite bounds are
   allowed, given that they are supported by the agent algorithm.
   Discrete components have the cardinality of the component on the 2nd
   column. The rest of the column is not taken into account.

   - 'actionspacesize' the size of the joint action space. Formed in
   the same way as the above.

   - 'statenames' a cell array, each element containing a unique string
   name for the corresponding state component.

   - 'actionnames' a cell array, each element containing a unique string
   name for the corresponding action component.

  See below for an example of how to form these arrays.

  The size descriptors must have the same length, thus, if at least one of
  the states or actions is continuous, all state and action size
  descriptors must have 4 columns.

  Naturally, the order of the state and action components in the arguments
  must be consistent, such that e.g., the agent finds the size of its own
  state space at the position specified in its own index map.

  See also world, agent_initlearn

CROSS-REFERENCE INFORMATION ^

This function calls: This function is called by:
Generated on Wed 04-Aug-2010 16:55:08 by m2html © 2005