Skip to content

Latest commit

 

History

History
177 lines (124 loc) · 6.65 KB

torille.md

File metadata and controls

177 lines (124 loc) · 6.65 KB

torille

ToribashControl

Main class which controls Toribash process and handles communication.

__init__(settings=None, draw_game=False, executable=ToribashConstants.TORIBASH_EXE, port=ToribashConstants.PORT)

Creates a new Toribash controller, but does not launch Toribash instance. ToribashControl instance can be serialized (pickled) at this point.

Parameters:

  • settings: ToribashSettings used to define initial settings of the game.
  • draw_game: Boolean specifying if game should be human-"enjoyable", i.e. run at convenient frame-rate and display characters
  • executable: String which specifies path to the game executable (toribash.exe). By default this assumes Toribash binary is in toribash directory next to torille.py, which is the case is installed via pip.
  • port: Integer which specifies port where Toribash Lua attempts to connect to.

init()

Launches Toribash instance and begins to control it. Uses FileLock to prevent multiple parallel instances from mixing Toribash instances with each other.

ToribashControl object is not serializable after calling this.

close()

Closes the connection with Toribash and kills the process. Can be called at any point after init()

get_state()

Returns current state of the game as a ToribashState object.

Throws exceptions if game has not been initialized, connection is broken or connection times out.

**Note: ** This function can't be called twice in row: It has to be followed by make_actions() or reset()

Returns:

  • state: ToribashState of the current state of the game
  • terminal: Boolean indicating if state was terminal

make_actions(actions)

Sends given actions to Toribash and proceeds the game by set amount of frames (given in settings).

Throws exceptions if game has not been initialized, connection is broken/times out or actions are wrong.

**Note: ** This function can't be called twice in row: It has to be followed by get_state() or reset()

Parameters:

  • actions: List of two lists with ToribashConstants.NUM_CONTROLLABLES elements. Each element is integer from range {1,2,3,4}. These represent joint states for both Toribash characters. **Note: ** Toribash character hands only take {0,1}, but for simplicity controller maps {1,2} => 0 and {3,4} => 4.

reset(settings=None)

Resets the game to initial state, sets new settings (if given) and returns initial state.

**Note: ** Must be followed by get_state() which returned terminal=True.

Parameters:

  • settings: ToribashSettings used to define initial settings of the game.

Returns

  • state: ToribashState of the initial state of the game

validate_actions(actions)

Used to validate actions internally to avoid Toribash crashing without errors.

Throws an exception if action is invalid.

Parameters:

  • actions: List of two lists with ToribashConstants.NUM_CONTROLLABLES integers.

get_state_dim()

Returns number of variables in state per player

Returns:

  • state_dim: Number of elements in state vector per player

get_num_joints()

Returns number of joints per player (number of controllables)

Returns:

  • action_dim: Number of joints per player

get_num_joint_states()

Returns number of states a joint can be in

Returns:

  • action_categories: Number of states joint can be in

ToribashState

Class used to refine and represent state from Toribash. Returned by ToribashControl.get_state() and ToribashControl.reset().

limb_positions

A Numpy array of shape (2, ToribashConstants.NUM_LIMBS, 3) containing positions of body parts of both players.

joint_states

A Numpy array of shape (2, ToribashConstants.NUM_CONTROLLABLES) containing current states of the joints of both players

injuries

A Numpy array of shape (2,) containing current injury of both players (value seen in game at the top-left and top-right.)

ToribashSettings

Class used to represent and contain rules for Toribash. These can be used to modify the game's mechanics.

**Note: ** New settings can only be applied at ToribashControl.reset(settings).

DEFAULT_SETTINGS (class-variable)

OrderedDict defining the default settings of Toribash game:

Possible settings

Name Description Type Default
matchframes  Length of an episode in frames Integer 500
turnframes Number of frames per turn (aka frame-skip) Integer (interval [2,matchframes]) 10
engagement_distance Starting distance between characters  Integer 100
engagement_height  Starting height of characters Integer 0
engagement_rotation Starting rotation (degrees, anti-clockwise) of characters Integer 0
gravity_x Strength of gravity in X axis  Float 0.0
gravity_y Strength of gravity in Y axis  Float 0.0
gravity_z Strength of gravity in Z axis  Float -9.81
damage Is damaging enabled {0,1} 0
dismemberment_enable Is dismemberment enabled {0,1} 1
dismemberment_threshold Force required to dismemberment to happen Integer 100
fractures_enable Enable fractures (disables joints) {0,1} 0
fractures_threshold Force required for fractures Integer 0
disqualification_enabled Enable disqualification (see link below for more info) {0,1} 0
disqualification_flags Settings for disqualification (see link below for more info) Integer 0
disqualification_timeout How long one can touch ground before disqualified Integer 0
dojo_type How battle arena works (see link below for more info) Integer 0
dojo_size How large the arena is  Integer 0
replay_file If not equal to "None", Toribash will save replay of the played episode in replay/[replay_file] at end of the episode String "None"

See this Toribash topic for more info on settings.

__init__(**kwargs)

Creates new ToribashSettings object using ToribashSettings.DEFAULT_SETTINGS as a base and overriding values based on kwargs.

Parameters:

  • kwargs: Keyword arguments used to override default settings. See keys in DEFAULT_SETTINGS. E.g. ToribashSettings(matchframes=1000) creates new settings object where match length is set to 1000 frames, rather than default 500.

get(key)

Return the value of given setting name.

Parameters:

  • key: String of the setting name.

Returns:

  • value: Value of the setting given

set(key,value)

Set given setting to a given value.

Parameters:

  • key: String of the setting name.
  • value: Value of the setting

validate_settings()

Used internally to validate settings (correct type, no illegal characters (",") in strings).

Throws errors and warnings accordingly.