Skip to content

Latest commit

 

History

History
40 lines (24 loc) · 1.27 KB

setup.md

File metadata and controls

40 lines (24 loc) · 1.27 KB

gym-mosuemaze

The MouseMaze environment is a domain featuring discrete state and action spaces.

The setup

The mousemaze task initializes a single mouse agent, 2 shock-wire/traps and 3 cheese pieces/pizza slices in the maze.

Mouse

The mouse agent is rewarded +25 for every cheese piece/pizza slice it lands on and eats. The mouse agent is rewarded -8 for every shock-wire/trap it lands on. The mouse agent is rewarded -1 for every move it takes that does not complete the episode or if the mouse does not land any anything in that step. The mouse agent is rewarded -7 for every time it runs into a wall or tries leaving the map.

Maze

The maze itself will be a 4x4 block-layout which will have walls between certain blocks. It's starting state might look like this:

Image of Maze

but it will render for console output.

Action Space

The mouse agent can take the following actions:

  • Go North
  • Go South
  • Go West
  • Go East

Episode

The episode starts when the environment is initialized or reset. The episode ends when the mouse has collected 2 cheese pieces/pizza slices.

State Space

Installation

cd gym-mousemaze
pip install -e .