GridWorld

Module that implements a simple grid world with obstacles and rewards/reinforcement. Movements are given either as absolute or relative directions in the MOVE input.

In absolute mode set by move = "max", the maximum direction in the MOVE array sets the direction of movement. The four components in the array correspond to north (up), east (right), south (down) and west (left). In absolute mode with summation, set by move = "sum", the movement is the result of adding all directions in the MOVE input before finding the maximum direction.

In relative mode, set by move = "relative", the array MOVE is interpreted as having three components: move-ahead, turn-left, turn-right. With mode = "relative_sum", the components turn-left and turn-right are subtracted before the maximum is found.

No action is performed if MOVE is the zero vector.

The agent is automatically moved to the initial location when a reward location is reached.

Example XML definition

A simple example

  <module
      class = "GridWorld"
      name = "GridWorld"
      start_x = "2"
      start_y = "2"
      move = "absolute"
  />

Parameters

Name	Description	Type	Default
class	The class name for the module; must be "GridWorld".	string
name	The name of this instance of the module.	string
move	How to use the MOVE input	choices: max sum relative_max relative_sum	max
start_x	Starting x coordinate for the agent	int	1
start_y	Starting y coordinate for the agent	int	1
normalize_coordinate	Map coordinate output to the interval 0..1	bool	no

Module Connections

Inputs

Name	Description
MOVE	Move in absolute directions (N, E, S, W) or relative (Forward, Turn-Left, Turn-Right).
OBSTACLES	Matrix with obstacles
VALUES	Matrix with values

Outputs

Name	Description
LOCATION	Matrix with location of agent
COORDINATE	Array with location of agent as an x, y value
LOCAL_OBSTACLES	3x3 grid around agent
LOCAL_VALUES	3x3 grid around agent
REWARD	The current reward value
COLLISION	The action makes the agent collide with a wall
IMAGE	An image of the world with agent. Can be used by a viewer to visualize the world

Limitations

It should be possible to turn off the automatic movement of the agent from goal to start locations.

Author

Christian Balkenius
christian.balkenius@lucs.lu.se
Lund University Cognitive Science

Files

GridWorld.h
GridWorld.cc
GridWorld.ikc