ILE_API.md

ILE API

How to Use the ILE

Table of Content

• Lists
• Classes
• Options

Lists

• Agent settings
• Interactable shapes
• Materials
• Tool shapes

Classes

Some configurable ILE options use the following classes (represented as dicts in the YAML):

AgentActionConfig

Represents what animations an agent is to perform.

• id (str or list of str): The ID of the animation the agent should perform. Default: no deafult, must be set.
• step_begin (int, or list of ints, or MinMaxInt dict): The step in that this animation should start. Default: no default, must be set.
• step_end (int, or list of ints, or MinMaxInt dict): The step in that this animation should end. Default: unset - animation won't end until another animation changes it or the animation finishes.
• is_loop_animation (bool or list of bool): Determines whether the animation should loop or end when finished. Default: False

AgentConfig

Represents the template for a specific agent (with one or more possible variations) that will be added to each scene. Each template can have the following optional properties:

• num (int, or MinMaxInt dict, or list of ints and/or MinMaxInt dicts): The number of agents with this template to generate in each scene. For a list or a MinMaxInt, a new number will be randomly chosen for each scene. Default: 1
• actions (AgentActionConfig) dict, or list of AgentActionConfig dicts): Configures the agent's actions (a.k.a. animations). If configured as a list, each action will be applied, and any randomness will be reconciled within each array element for each scene. Default: idle
• agent_settings (AgentSettings dict, or list of AgentSettings dicts): Configures the agent's appearance. Default: random
• keyword_location (KeywordLocationConfig dict): A keyword location for this agent.
• movement (bool, or AgentMovementConfig) dict, or list of bools and/or AgentMovementConfig dicts): Configures this agent to move (walk/run) around the room. If true, the agent will be assigned a random movement pattern for each scene. If configured as a list, one option will be randomly chosen for each scene. Default: no movement
• pointing (AgentPointingConfig dict, or list of AgentPointingConfig dicts): Configures this agent to start pointing on a specific step. The pointing lasts indefinitely. This cancels out any other actions or movement. Use pointing.object_label to point at a specific object. Use pointing.walk_distance to turn around and walk toward the object before pointing at it. If configured as a list, one option will be randomly chosen for each scene. Default: no pointing
• position (VectorFloatConfig dict, or list of VectorFloatConfig dicts): The position of this agent in each scene. If configured as a list, a new position will be randomly chosen for each scene. Default: random
• rotation_y (float, or MinMaxFloat dict, or list of floats and/or MinMaxFloat dicts): The rotation of this agent in each scene. If configured as a list, a new rotation will be randomly chosen for each scene. Default: random
• type (string, or list of strings) The model ("type") of the agent. Please see the list in our schema doc for all available options. Default: random

Example:

num: 1
type: agent_female_02
agent_settings:
  chest: 2
  eyes: 1
position:
  x: 3
  z: 3
rotation_y: 90
actions:
  id: TPE_wave
  step_begin: [3, 5]
  step_end:
    min: 8
    max: 11
  is_loop_animation: [True, False]

AgentMovementConfig

Represents what movements the agent is to perform. If the points field is set, the bounds and num_points fields will be ignored.

• animation (str, or list of str): The animation that will be shown while the agent moves. Default: 'TPM_walk' or 'TPM_run'
• bounds (list of VectorFloatConfig dicts): A set of points that create a polygon, inside of which the movement points will be chosen randomly. If bounds has fewer than 3 points, then the entire room is used as the bounds. This option is ignored if points is configured. Default: Entire room
• num_points (int, or MinMaxInt dict, or list of ints and/or MinMaxInt dicts): The number of random movement points to generate inside of the bounds. This option is ignored if points is set. Default: between 2 and 10, inclusive.
• points (list of VectorFloatConfig): The list of points to which the agent should move. If set, points takes precedence over bounds and num_points. Default: Use bounds
• repeat (bool or list of bool): Whether the agent's movement pattern should loop indefinitely (true) or end when finished (false). Default: random
• step_begin (int, or list of ints, or MinMaxInt dict): The step at which this movement should start. Required.

AgentPointingConfig

Represents the agent pointing.

• object_label (string, or list of strings): The label of the object in the scene at which to point. Default: The agent points in whatever direction it's facing (based on its rotation_y setting). The step in which the pointing should start. Default: 1
• step_begin (int, or MinMaxInt dict, or list of ints and/or MinMaxInt dicts): The step on which the pointing should start. The agent will idle up to this step. If walk_distance is set, then the movement will begin on this step, and the pointing will happen immediately after the movement. Default: 1 walk_distance (float, or MinMaxFloat dict, or list of floats and/or MinMaxFloat dict): If set, adjusts the agent's starting position (which can still be set normally using the position option) by the configured distance in the direction away from the object corresponding to object_label; the agent will then turn around, walk two times the configured distance, and then point at the object. Will override step_begin to be after the movement ends. Use the step_begin option to configure when the movement should begin. Default: no movement

AgentSettings

Describes the appearance of the agent. Each property defaults to a random valid setting. Please see the sections in our schema doc for the available options.

• chest (int, or list of ints)
• chestMaterial (int, or list of ints)
• eyes (int, or list of ints)
• feet (int, or list of ints)
• feetMaterial (int, or list of ints)
• glasses (int, or list of ints)
• hair (int, or list of ints)
• hairMaterial (int, or list of ints)
• hatMaterial (int, or list of ints)
• hideHair (bool, or list of bools)
• isElder (bool, or list of bools)
• jacket (int, or list of ints)
• jacketMaterial (int, or list of ints)
• legs (int, or list of ints)
• legsMaterial (int, or list of ints)
• showBeard (bool, or list of bools)
• showGlasses (bool, or list of bools)
• showJacket (bool, or list of bools)
• showTie (bool, or list of bools)
• skin (int, or list of ints)
• tie (int, or list of ints)
• tieMaterial (int, or list of ints)

AgentTargetConfig

Defines all the configurable options for shortcut_agent_with_target.

• agent (AgentConfig dict, or list of AgentConfig dicts): Configures the settings for the agent.
• agent_position (VectorFloatConfig dict, or list VectorFloatConfig dicts): Deprecated. Please use agent.position
• movement (bool, or list of bools): Deprecated. Please use agent.movement
• movement_bounds (list of VectorFloatConfig dicts): Deprecated. Please use agent.bounds

BisectingPlatformConfig

Defines details of the shortcut_bisecting_platform shortcut. This shortcut creates a platform that bisects the room, where the performer will start. On default, a blocking wall is on that platform, forcing the performer to choose a side to drop off of the platform, but this can be disabled.

• has_blocking_wall (bool): Enables the blocking wall so that the performer has to stop and choose a side of the room. Default: True
• has_double_blocking_wall (bool): Enables the double blocking wall used in Spatial Reorientation tasks. Overrides both has_blocking_wall and has_long_blocking_wall. Default: False
• has_long_blocking_wall (bool): Enables the extra-long blocking wall. Overrides has_blocking_wall. Default: False
• is_short (bool): Makes the platform short (a Y scale of 0.5 rather than 1). Default: False
• is_thin (bool): Makes the platform thin (an X scale of 0.5 rather than 1). Default: False
• other_platforms (StructuralPlatformConfig dict, or list of StructuralPlatformConfig dicts): Configurations to generate other platforms that may intersect with the bisecting platform. Default: No other platforms
• position_z (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The starting Z position for the performer agent. Default: 0.5 away from the back wall

DoubleDoorConfig

Defines details of the double door shortcut. A wall with 2 doors (a.k.a. door occluder, or "door-cluder" or "doorcluder") will bisect the room in the perpendicular direction. A platform can be added to the middle back wall where the AI will be positioned at the beginning of the scene. Lava can be added down the middle of the floor splitting the room in half.

• add_freeze (bool, or list of bools): If true and 'start_drop_step is' greater than 0, the user will be frozen (forced to Pass) until the wall and doors are in position. If the 'start_drop_step' is None or less than 1, this value has no effect. Default: True
• add_lava (bool, or list of bools): If true adds lava along the z axis of the room. Default: True
• add_platform (bool, or list of bools): If true adds a platform at the back of the room where the AI will start on. Default: True
• door_material (string, or list of strings): The material or material type for the doors.
• occluder_wall_position_z (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Where the occluder wall will cross the z-axis in the room. occluder_distance_from_performer will override this value. Default: 0 (middle of the room)
• occluder_distance_from_performer (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): If there is a platform the, the performer will start on top of the platform and the occluder wall will be placed this distance (occluder_distance_from_performer) from the performer. Must be greater than 1 or null Default: 6.5
• platform_height (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The height (y scale) of the platform Default: 1.5
• platform_length (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The lenth (z scale) of the platform Default: 1
• platform_width (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The width (z scale) of the platform Default: 1
• start_drop_step (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Step number to start dropping the bisecting wall with doors. If None or less than 1, the wall will start in position. Default: None
• wall_material (string, or list of strings): The material or material type for the wall.

FloorAreaConfig

Defines an area of the floor of the room. Note: Coordinates must be integers. Areas are always size 1x1 centered on the given X/Z coordinate. Adjacent areas are combined.

• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of areas to be used with these parameters
• position_x (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): X position of the area.
• position_z (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Z position of the area.
• size (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Size of a single contiguous area. If position_x and/or position_z are set, the area will include a space matching those coordinate(s). Default: 1

FloorMaterialConfig

Defines details of a specific material on a specific location of the floor. Be careful if num is greater than 1, be sure there are possibilities such that enough floor areas can be generated. Note: Coordinates must be integers. Areas are always size 1x1 centered on the given X/Z coordinate. Adjacent areas are combined.

• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of areas to be used with these parameters
• material (string, or list of strings): The floor's material or material type.
• position_x (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): X position of the area.
• position_z (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Z position of the area.

GoalConfig

A dict with str category and optional target or targets properties that correspond to the goal and target object(s) in each scene.

The goal category for interactive scenes is either "retrieval" (for one target) or "multi retrieval" (for multiple targets). If the category is not set, it will default to "retrieval" if target is set or "multi retrieval" if targets is set.

The goal category for passive scenes is either "intuitive physics" (for passive physics scenes), "agents" (for NYU passive agent scenes), or "passive" (for all other passive scenes). Please see passive_physics_scene for making passive physics scenes. Please do not use the ILE Scene Generator for making NYU passive agent scenes.

See the MCS Python API for more information about the goal categories:

https://nextcenturycorporation.github.io/MCS/api.html#machine_common_sense.GoalCategory

The target property is only needed for "retrieval" scenes, and the targets property is only needed for "multi retrieval" scenes. The target property is either an InteractableObjectConfig dict or a list of InteractableObjectConfig dicts. For a list, one dict will be randomly chosen within the list in each new scene. The targets property is a list of InteractableObjectConfig dicts, and ALL the dicts will be used as targets in each new scene. All goal target objects will be assigned the 'target' label. The num property will only be used with multiple targets.

Please note that the "traversal" and "transferral" goal categories, along with the target_1 and target_2 properties, are no longer supported.

Example:

category: retrieval
target:
    shape: soccer_ball
    scale:
      min: 1.0
      max: 3.0

ImitationTaskConfig

Defines details of the shortcut_imitation_task shortcut.

• trigger_order (string, or list of strings): The combination the three containers must be openend in order to make the target appear. Oriented by facing the front of the containers. Must be one of the following. For opening 2 containers: left_middle, left_right, middle_left, middle_right, right_middle, right_left. For opening 1 container: left, middle, right. Default: random
• containers_on_right_side (bool, or list of bools): Whether the containers should be to the right or left of the performer. Default: random
• kidnap_option: (string, or list of strings): Dictates what teleports in the scene after the agent performs its imitation sequence to and the performer is kidnapped. Options are:

1. agent_only: The imitation agent teleports away from the containers but in view. Nothing else is teleported.
2. containers: The containers are teleported anr rotated but still in view. The imitation agent is teleported away from the containers but in view.
3. performer: The performer is teleported to a random part of the room but looks at the center of the room where the containers still are. The imitation agent is teleported away from the containers but in view. Default: random

• relative_container_rotation: (int, or list of int): Dictates what degree of rotation change the containers will rotate relative to their clockwise starting rotation. For example, if the they are facing 270 (starting on the right side) and container_rotation of 45 will make the final rotation 315. Examples of what rotations containers cannot rotate:

1. 360 and 180 for both left and right because the containers will face identical or opposite from their start.
2. 90 on right side, 270 on left which will make the containers face directly away from the performer. Direction containers can rotate for right side: [45, 135, 225, 270, 315] Direction containers can rotate for left side: [45, 90, 135, 225, 315] Default: random

• global_container_rotation: (int, or list of int): Dictates what degree of rotation change the containers will rotate on based on the absolute global rotation. Options are: [45, 135, 180, 225, 315]. Will override relative_container_rotation Default: random

InteractableObjectConfig

Represents the template for a specific object (with one or more possible variations) that will be added to each scene. Each template can have the following optional properties:

• num (int, or list of ints, or MinMaxInt dict): The number of objects with this template to generate in each scene. For a list or a MinMaxInt, a new number will be randomly chosen for each scene. Default: 1
• num_targets_minus (int, or list of ints, or MinMaxInt dict): Overrides the num option. Count the total number of targets, subtract num_targets_minus from the count, and generate that many objects. For example, in a scene with 5 targets, a num_targets_minus of 1 would generate 4 objects. Default: Use num
• dimensions (VectorFloatConfig dict, int, MinMaxInt, or a list of any of those types): Sets the dimensions of the object in meters. Overrides scale. If only one dimension is configured, then the same scale will be used for the other two dimensions. Default: Use scale
• identical_to (str): Used to match to another object with the specified label, so that this definition can share that object's exact shape, scale, and material. Overrides identical_except_color
• identical_except_color (str): Used to match to another object with the specified label, so that this definition can share that object's exact shape and scale, but not its material (color/texture).
• keyword_location: (KeywordLocationConfig): Used to identify one of the qualitative locations specified by keywords. This field should not be set when position or rotation are also set.
• labels (string, or list of strings): labels to associate with this object. Components can use this label to reference this object or a group of objects. Labels do not need to be unique and when objects share a labels, components have options to randomly choose one or choose all. See specific label options for details.
• locked (bool or list of bools): If true and the resulting object is lockable, like a container or door, the object will be locked. If the object is not lockable, this field has no affect.
• material (string, or list of strings): The material (color/texture) to use on this object in each scene. For a list, a new material will be randomly chosen for each scene. Default: random
• not_material (string): Do not use this material, or any other materials that share the same colors as this material, on this object. Default: none
• position (VectorFloatConfig dict, or list of VectorFloatConfig dicts): The position of this object in each scene. For a list, a new position will be randomly chosen for each scene. Default: random
• position_relative (RelativePositionConfig dict, or list of RelativePositionConfig dicts): Configuration options for positioning this object relative to another object, rather than using position. If configuring this as a list, then all listed options will be applied to each scene in the listed order, with later options overriding earlier options if necessary. Default: not used
• rotation (VectorFloatConfig dict, or list of VectorFloatConfig dicts): The rotation of this object in each scene. For a list, a new rotation will be randomly chosen for each scene. Default: random
• scale (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts, or VectorFloatConfig dict, or list of VectorFloatConfig dicts): The scale of this object in each scene. A single float will be used as the scale for all object dimensions (X/Y/Z). For a list or a MinMaxFloat, a new scale will be randomly chosen for each scene. This field can be overriden by 'dimensions'. Default: 1
• separate_lid (int, or list of ints, or MinMaxInt dict): Only applies to objects with a 'separate_container' shape. If a negative number or None the container will not have a separate lid. If 0 the container will have a separate lid already attatched at the start of the scene. If greater than 0 the container's separate lid will be placed by a placer. The number given will be the step the placer starts its placement. 15 steps later the lid will be attached to the object. Note that if the container has any moves configured, the placer and separate lid will automatically calculate the x,z position of the container until the step it starts it decent. So ensure the container is not moving from the step the placer begins its decent to when the lid attachment occurs 15 steps later, otherwise the placer will place the lid where the container was at the step it started its decent while the container moves away from that position.
• separate_lid_after: (str, or list of strs): Overrides the separate_lid (overriding the manual config) based on the movement of other object(s) in the scene. Should be set to one or more labels for mechanical objects that may move or rotate, like placers. The separate_lid of this object will be set to the step immediately after ALL of the objects finish moving and rotating. If multiple labels are configured, all labels will be used. Please note that if this container is moved by a "placer" before its lid is attached, then its lid will be correctly repositioned to use the new location; but if this container changes position before its lid is attached due to a turntable rotating underneath it, then its lid will NOT be correctly repositioned. Default: Use separate_lid
• shape (string, or list of strings): The shape (object type) of this object in each scene. For a list, a new shape will be randomly chosen for each scene. Default: random
• surrounded_by_lava (bool, or list of bools): Whether or not this object will be surrounded by lava. Default: False
• surrounding_lava_size: (int, or list of ints): The width of the surrounding lava. Only used if surrounded_by_lava is set to True. In that case, default is 1, otherwise the value will be None. Note that this value will need to fit within room bounds, otherwise may error.
• surrounding_safe_zone: (bool, or list of bools): If set to True and surrounded_by_lava is True, a buffer area (at least 0.5) will be set around the surrounding lava to ensure that it does not touch any other sections of lava or room walls/corners, and therefore allowing the performer to move around the surrounding lava's perimeter.
• rotate_cylinders (bool): Whether or not to rotate cylindrical shapes along their x axis so that they are placed on their round sides (needed for collision scenes). This would only apply to these shapes: 'cylinder', 'double_cone', 'dumbbell_1', 'dumbbell_2', 'tie_fighter', 'tube_narrow', 'tube_wide'. Note that this will override any x rotation previously specified by 'rotation'. Default: False

Example:

num: 1
material: "Custom/Materials/Blue"
position:
    x: 3
    z: 3
rotation:
    y: 90
scale: 1.5
shape: ball

Labels Example:

specific_interactable_objects:
  -
    num: 1
    position:
      x: 3
      z: 3
    rotation:
      y: 90
    shape: chest_3
    scale: 1.5
    labels: my_container
  -
    num: 1
    material: "Custom/Materials/Blue"
    scale: 0.5
    shape: ball
    keyword_location:
      keyword: in
      container_label: my_container
  -
    num: 1
    material: "Custom/Materials/Blue"
    scale: 0.5
    shape: ball
    keyword_location:
      keyword: adjacent
      relative_object_label: my_container
  -
    num: 2
    identical_to: my_container

KeywordLocationConfig

Describes an object's keyword location. Can have the following properties:

• keyword (string, or list of strings): The keyword location, which can be one of the following:
  • adjacent - The object will be placed near another object. The other object must be referenced by the 'relative_object_label' field, and the distance away from the relative object must be set by the adjacent_distance field.
  • adjacent_corner - The object will be placed near the corner referenced by the 'relative_object_label' field. The corner labels are 'front_left' (-X, +Z), 'front_right' (+X, +Z), 'back_left' (-X, -Z), and 'back_right' (+X, -Z). The distance away from the corner will be determined by the adjacent_distance field.
  • adjacent_performer - The object will be placed next to the performer. The object can be placed in 'front', 'back', left, or 'right' of the performer using the 'direction'. The object can also be specified to be 'in_reach' or 'out_of_reach' via the 'distance'. By default, the object will be placed in a random direction, but 'in_reach'. other object must be referenced by the 'relative_object_label' field. If multiple objects have this label, one will be randomly chosen.
  • back - The object will be placed in the 180 degree arc behind the performer's start position.
  • behind - The object will be placed behind another object, relative to the performer's start position. The other object must be referenced by the 'relative_object_label' field. If multiple objects have this label, one will be randomly chosen.
  • between - The object will be placed between the performer's start position and another object. The other object must be referenced by the 'relative_object_label' field. If multiple objects have this label, one will be randomly chosen.
  • front - The object will be placed in a line in front of the performer's start position.
  • in - The object will be placed inside a container. The container must be referenced by the 'container_label' field. If multiple objects have this label, one will be randomly chosen.
  • in_with - The object will be placed inside a container along with another object. The container must be referenced by the 'container_label' field. The other object must be referenced by the 'relative_object_label' field. If multiple objects have these label, one will be randomly chosen for each field.
  • occlude - The object will be placed between the performer's start position and another object so that this object completely occludes the view of the other object. The other object must be referenced by the 'relative_object_label' field. If multiple objects have this label, one will be randomly chosen.
  • on_center - The object will be placed on top of another object in the center of the bounds. This option is best for objects the are similar in size or for use cases where objects are desired to be centered. The object must be referenced by the 'relative_object_label' field. If multiple objects have this label, one will be randomly chosen.
  • on_top - The object will be placed on top of another object in a random location. This option is best for when the object is significantly smaller than the object it is placed on (I.E. a small ball on a large platform). If the objects are similar in size (I.E. two bowls), use 'on_center'. The object must be referenced by the 'relative_object_label' field. If multiple objects have this label, one will be randomly chosen.
  • opposite_x - The object will be placed in the exact same location as the object referenced by relative_object_label except that its x location will be on the opposite side of the room. Its rotation will also be mirrored, though it can be adjusted using the rotation property within this keyword location option. There are no adjustments to find a valid location if another object already exists in the final location.
  • opposite_z - The object will be placed in the exact same location as the object referenced by relative_object_label except that its z location will be on the opposite side of the room. There are no adjustments to find a valid location if another object already exists in final location.
  • random - The object will be positioned in a random location, as if it did not have a keyword location.
  • associated_with_agent - This object will be held by an agent referenced by the 'relative_object_label' field.
  • along_wall - This object will be placed along a wall referenced by the 'relative_object_label' field. The wall labels are 'front_wall', 'back_wall', 'left_wall, and 'right_wall'. If no wall is provided, a wall will be chosen at random.
• adjacent_distance (VectorFloatConfig, or list of VectorFloatConfigs): The X/Z distance in global coordinates between this object's position and the relative object's (or corner's) position. Only usable with the adjacent or adjacent_corner keyword. By default, this object will be positioned 0.1 away from the relative object (or corner) in a random, valid direction. Note that if using this for a corner with the surrounded_by_lava property, you'll need to use the same dimensions for x/z and that they're divisible by 0.5 to make sure they line up correctly with the lava island.
• container_label (string, or list of strings): The label of a container object that already exists in your configuration. Only required by some keyword locations.
• relative_object_label (string, or list of strings): The label of a second object that already exists in your configuration. Only required by some keyword locations.
• position_relative_to_start (VectorFloatConfig, or list of VectorFloatConfigs): Currently only supported with the on_center keyword: How much to translate object from the center position of the relative object along the x and z axis. This works like a percentage, represented as x/z values that range from -1.0 to 1.0 (with both of those values being furthest from the center in either direction). Note that this assumes the relative object has a rectangular boundary.

KeywordObjectsConfig

Describes a single group of a keyword objects. Keyword objects has the following properties:

• keyword (string, or list of strings): The type of object, like confusor, container, obstacle, or occluder. If an array is given, one will be chosen for each creation of this group.
  • "asymmetric_containers": Asymmetric open-topped containers. Subset of "open_topped_containers". Used in interactive support relations tasks.
  • "bins": Bin-like containers. Subset of "open_topped_containers". Used in interactive spatial reorientation tasks.
  • "confusors": Objects that are similar to the target in two of the following: color, shape, size.
  • "containers": Receptacles that can contain other objects and can open and close, like chests. These containers can be of any size.
  • "containers_can_contain_target": Receptacles that can contain other objects and can open and close, like chests. These containers will be large enough to contain the target. If a goal target does not exist, generation will fail.
  • "containers_cannot_contain_target": Receptacles that can contain other objects and can open and close, like chests. These containers will not be large enough to contain the target. If a goal target does not exist, generation will fail.
  • "context": Objects (also called "distractors") that are small, pickupable, and serve to clutter the room to possibly distract the AI. They are never the same shape as the target object, if the scene has a goal with a target object.
  • "obstacles": Objects that prevent movement through/under them but allow sight through them, like chairs or tables with four individual legs.
  • "occluders": Objects that prevent both movement through/under them AND sight through them, like sofas or bookcases with solid back panels. This is completely different from the occluders in passive intuitive physics scenes.
  • "open_topped_containers": Objects that can contain other objects but have no lid and open tops instead.
  • "symmetric_containers": Symmetric open-topped containers. Subset of "open_topped_containers". Used in interactive support relations tasks.
• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): The number of this object the user wants to create.
• keyword_location: (KeywordLocationConfig): One of the keyword locations for this object or set of objects. Any choices in keyword_location are made for each object inside the group, not the group as a whole. Overrides any configured position and rotation.
• labels (string, or list of strings): labels to associate with this object. Components can use this label to reference this object or a group of objects. Labels do not need to be unique and when objects share a labels, components have options to randomly choose one or choose all. See specific label options for details.
• position (VectorFloatConfig dict, or list of VectorFloatConfig dicts): The position of these objects in each scene. If given a list, a position will be randomly chosen for each object and each scene. Is overridden by the keyword_location. Default: random
• position_relative (RelativePositionConfig dict, or list of RelativePositionConfig dicts): Configuration options for positioning this object relative to another object, rather than using position. If configuring this as a list, then all listed options will be applied to each scene in the listed order, with later options overriding earlier options if necessary. Default: not used
• rotation (VectorFloatConfig dict, or list of VectorFloatConfig dicts): The rotation of these objects in each scene. If given a list, a rotation will be randomly chosen for each object and each scene. Is overridden by the keyword_location. Default: random

KnowledgeableAgentPairConfig

Defines all of the configurable options for [knowledgeable_agent_pair] (#knowledgeable_agent_pair).

• position_1 (Vector3): One possible location for an agent to be.
• position_2 (Vector3): One possible location for an agent to be.
• pointing_step (int): Starting step of pointing action.
• target_labels (str): Label for the knowledgeable agent to use to find the goal.
• non_target_labels (str): Label so the non-knowledgeable agent to use to find what it should be pointing to.

LavaTargetToolConfig

Defines details of the shortcut_lava_target_tool shortcut. This shortcut creates a room with a target object on an island surrounded by lava. There will also be a block tool to facilitate acquiring the goal object.

• front_lava_width (int, or list of ints, or MinMaxInt: The number of tiles of lava in front of the island. Must produce value between 2 and 6 for rectangular tools, 1 to 3 for hooked and isosceles tools. Default: Random based on room size and island size
• guide_rails (bool, or list of bools): If True, guide rails will be generated to guide the tool in the direction it is oriented. If a target exists, the guide rails will extend to the target. This option cannot be used with tool_rotation. Default: False
• island_size (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): The width and length of the island inside the lava. Must produce value from 1 to 5 for rectangular tools, 1 to 3 for hooked and isosceles tools. Default: Random based on room size
• left_lava_width (int, or list of ints, or MinMaxInt: The number of tiles of lava left of the island. Must produce value between 2 and 6 for rectangular tools, but will be ignored for hooked and isosceles tools, since the lava should extend to the wall in that case. Default: Random based on room size and island size
• random_performer_position (bool, or list of bools): If True, the performer will be randomly placed in the room. They will not be placed in the lava or the island Default: False
• random_target_position (bool, or list of bools): If True, the target object will be positioned randomly in the room, rather than being positioned on the island surrounded by lava. Default: False
• rear_lava_width (int, or list of ints, or MinMaxInt: The number of tiles of lava behind of the island. Must produce value between 2 and 6 for rectangular tools, 1 to 3 for hooked and isosceles tools. Default: Random based on room size, island size, and other lava widths.
• right_lava_width (int, or list of ints, or MinMaxInt: The number of tiles right of the island. Must produce value between 2 and 6 for rectangular tools, but will be ignored for hooked and isosceles tools, since the lava should extend to the wall in that case. Default: Random based on room size and island size
• random_performer_position (bool, or list of bools): If True, the performer will be randomly placed in the room. They will not be placed in the lava or the island. If the tool_type is inaccessible the performer will randomly start on the side of the room where the target is where they cannot access the tool. Default: False
• tool_rotation (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Angle that tool should be rotated out of alignment with target. This option cannot be used with guide_rails, or the broken and inaccessible tool_type choices. For hooked and isosceles tools, we advice against setting a rotation higher than 315. Defaults to one of the following: 0, 45, 90, 135, 180, 225, 270, 315
• distance_between_performer_and_tool (float, or MinMaxFloat dict, or list of floats and/or MinMaxFloat dicts): Distance between the performer agent and the tool. If set to 0, will try all distances between 0.5 and the bounds of the room. If set to a MinMaxFloat with a max of 0 or greater than the room bounds, will try all distances between the configured min and the bounds of the room. If set to a MinMaxFloat with a min of 0, will try all distances between the configured max and 0.5. This cannot be used in combination with random_performer_position. Default: None
• tool_offset_backward_from_lava (float, or MinMaxFloat dict, or list of floats and/or MinMaxFloat dicts): The vertical offset of the tool: either further into the lava, for hooked and isosceles tools, or further from the lava, for other tools. Must be 0 or greater. Cannot be used with hooked and isosceles tools when tool_rotation is non-zero. Default: 0
• tool_horizontal_offset (float, or MinMaxFloat dict, or list of floats and/or MinMaxFloat dicts): The horizontal offset of tool either left or right from being aligned with the target. If tool_type is inaccessible this has alternate behavior. See inaccessible_tool_blocking_wall_horizontal_offset for description. Default: 0
• inaccessible_tool_blocking_wall_horizontal_offset ( RandomizableFloat): The horizontal offset of the blocking wall away from the target's horizontal position. Must be less than or equal to 0.5 (right side) or greater than or equal to 0.5 (left side). The performer will spawn on the side with the target. The tool will spawn on the opposite side of the wall. Setting tool_horizontal_offset has alternate behavior when combined with this property. tool_horizontal_offset will offset the tool from the the blocking wall and take the absolute value of the offset. The offset is based on the closest edges of the tool and the wall. For example, if the tool is rotated 90 degrees and the tool_horizontal_offset is 3 while inaccessible_tool_blocking_wall_horizontal_offset is -2 the the tool's closest edge to the wall will be a distance of 3 to the left of the wall since the wall has a negative offset to the left. Default: None
• tool_type (str, or list of strs): The type of tool to generate, either rectangular, hooked, isosceles, small, broken, or inaccessible. Both hooked and isoscelestools are L-shaped;hookedtools always have width 3, andisoscelestools always have width equal to their length. Ifhookedandisoscelestools are chosen and lava widths are not specified, the room will default to having an island size of 1, with lava extending all the way to the walls in both the left and right directions. The front and rear lava in the default hooked/isosceles tool case will each have a size of 1. Ifsmallis chosen the tool will always be a length of 1. Ifbrokenis chosen the tool will be the correct length but have scattered positions amd rotations for each individual broken piece. The tool will still have an overall rotation iftool_rotationis set. Ifinaccessiblethe room will be divided vertically by a blocking wall with a short height. Default:rectangular`

MinMaxFloat

A dict with float min and float max properties that represents a random inclusive numerical range. A float will be randomly chosen from within the range in each new scene.

Example:

min: 0.5
max: 1.5

MinMaxInt

A dict with int min and int max properties that represents a random inclusive numerical range. An int will be randomly chosen from within the range in each new scene.

Example:

min: 1
max: 10

PerformerStartsNearConfig

Defines details of performer_starts_near which places the performer near an object of a given label at a specified distance away.

• label (string or list of strings): Label of the object the performer will be placed near. Required.
• distance (float, or MinMaxFloat dict, or list of floats and/or MinMaxFloat dicts): Distance between the performer agent and the object. If set to 0, will try all distances between 0.5 and the bounds of the room. If set to a MinMaxFloat with a max of 0 or greater than the room bounds, will try all distances between the configured min and the bounds of the room. If set to a MinMaxFloat with a min of 0, will try all distances between the configured max and 0.5. Default: 0.5

Example:

label: container
distance: 0.5

PlacersWithDecoyConfig

Defines the configureable options for [placers_with_decoy] (#placers_with_decoy)

• activation_step List[int]: A list of steps that the placer/decoy can be activated.
• object_end_position_x List[float]: A list of x positions that the placer/decoy can move to.
• decoy_y_modifier float: The distance that the decoy should stop short so it doesn't go all the way to the ground. Should be the height of the target object.
• placed_object_position RandomizableVectorFloat3d: Starting position of the placer/decoy.
• labels RandomizableString: The labels that should be applied the to placer/decoy.
• move_object_y RandomizableFloat: A modifier to the release height of the target object.
• move_object_z RandomizableFloat: A modifier to the release depth of the target object.
• placed_object_labels RandomizableString: Labels to be applied to the target object.
• placed_object_rotation RandomizableFloat: Rotation to be applied to the target object during release.

RandomStructuralObjectConfig

A dict that configures the number of structural objects that will be added to each scene. Each dict can have the following optional properties:

• type (string, or list of strings): A type or a list of types that are options for this group. The options include the following:
  • doors: A random door.
  • droppers: A random dropper that drops a random object between step 0 and step 10.
  • floor_materials: A random floor area's texture is changed to a different texture.
  • holes: A hole in the room's floor in a random location. A hole is just where the floor has dropped to where the performer agent cannot get back up if it falls down. Random holes will not appear below the performer agent's start position. Note: Holes can end up appearing below objects which can cause the objects to fall into the hole.
  • l_occluders: A random L-shaped occluder.
  • lava: A random area in the floor is changed to a pool of lava.
  • moving_occluders: A random occluder on a pole that moves up and rotates to reveal anything behind it and then back down. It may repeat the movement indefinitely or stop after the first iteration.
  • occluding_walls: A random occluding wall.
  • placers: A random placer putting down a random object.
  • platforms: A random platform.
  • ramps: A random ramp.
  • throwers: A random thrower that throws a random object between step 0 and step 10. Throw force is 500 to 1000 times the mass of the thrown object.
  • tube_occluders: A random tube occluder.
  • turntables: A random turntable.
  • walls: A random interior room wall. Default: All types
• num (int, or list of ints, or MinMaxInt): The number of structural objects that should be generated in this group. Each object is one of the type choices.
• keyword_location: (KeywordLocationConfig): Used to identify one of the qualitative locations specified by keywords. This field should not be set when position or rotation are also set. Currently only used by occluding_walls.
• labels (string, or list of strings): One or more labels that are automatically assigned to the random object.
• relative_object_label (string, or list of strings): One or more labels that are automatically used by random droppers, placers, and throwers as the placed/projectile object labels. Currently ignored by all other structural object types.

RelativePathConfig

Defines details of an object's movement path as it relates to another object, for the moving object to avoid or collide with the other object.

• labels (string, or list of strings): The label of the object that the moving object should avoid or collide with. Required.
• option (string, or list of strings): Whether the moving object should avoid or collide with the other object. Options: "avoid", "collide". Default: null

RelativePositionConfig

Configure this object's position relative to an existing object in your scene. All options require the relative object's label to be configured.

• add_x (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The value added to this object's X position after being positioned at the relative object's X position. Default: 0
• add_z (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The value added to this object's Z position after being positioned at the relative object's Z position. Default: 0
• label (str, or list of strs): The label for an existing object to use as the "relative object" for positioning this object. Labels are not unique, so if multiple objects share the same label, the ILE will choose one of the available objects for each scene it generates.
• use_x (bool, or list of bools): Whether to use the relative object's X position for this object's X position. Default: if use_z is not set, then true; otherwise, false
• use_z (bool, or list of bools): Whether to use the relative object's Z position for this object's Z position. Default: if use_x is not set, then true; otherwise, false
• view_angle_x (bool, or list of bools): Whether to adjust this object's X position based on the angle of view from the performer agent's starting position and the relative object's position. Useful for positioning objects behind occluders (especially in the passive physics tasks).

SeeingLeadsToKnowingConfig

Defines details of the shortcut_seeing_leads_to_knowing shortcut. In this shortcut, the performer starts with four open containers in view. An agent will then walk into view. Placers will descend into each container, with only one having the soccer ball. After the placers are finished moving, the agent will attempt to go to the container that has the target object. The performer will then have to pick whether the scene is plausible or implausible based on the agent's behavior (note that for training, all generated scenes are plausible).

• target_behind_agent (bool, or list of bools): Determines whether or not the target is behind the agent. If target is behind the agent, the agent will guess and go towards one of the buckets behind itself.

SidestepsConfig

Contains data to describe the performer's sidesteps.

• begin (int, or MinMaxInt dict, or list of ints and/or MinMaxInt dicts): The step where the performer agent starts being frozen and can only use the "Pass" action. For example, if 1, the performer agent must pass immediately at the start of the scene. This is an inclusive limit.
• object_label (string, or list of strings): The label of the object the performer will sidestep around. The performer must be 3 distance away from the object's center for sidesteps to work.
• degrees (int, or MinMaxInt dict, or list of ints and/or MinMaxInt dicts): The positive or negative degrees the performer will sidestep around the object. Positive forces the performer to sidestep right while negative sidesteps left. The degree value must always be in 90 or -90 degree increments: [90, 180, 270, 360, -90, -180, -270, -360] Default: [90, 180, 270, 360, -90, -180, -270, -360]

StepBeginEnd

Contains a step range for a specific event.

• begin (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): The step where the performer agent starts being frozen and can only use the "Pass" action. For example, if 1, the performer agent must pass immediately at the start of the scene. This is an inclusive limit.
• end (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): The step where the performer agent ends being frozen and can resume using actions besides "Pass". Therefore, this is an exclusive limit.

StopPositionConfig

Set a stop position for a thrown object. Works best with non-spherical, non-cylindrical shapes, like toys with wheels.

• x (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The stop X position. Must also configure z. Default: null
• z (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The stop Z position. Must also configure x. Default: null
• behind (str, or list of strs): The label for an existing object behind which this object should stop. Only works if passive_physics_scene is true. Overrides the x and z options. Useful for stopping objects behind occluders. Default: null
• offscreen (bool, or list of bools): Sets the stop position offscreen, out of view of the performer agent. Only works if passive_physics_scene is true. Overrides the x and z options. Default: false

StructuralDoorConfig

Defines details of a structural door that can be opened and closed.

• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of structures to be created with these parameters
• labels (string, or list of strings): A label or labels to be assigned to this object. Always automatically assigned "doors"
• material (string, or list of strings): The structure's material or material type.
• position (VectorFloatConfig dict, or list of VectorFloatConfig dicts): The structure's position in the scene
• rotation_y (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The structure's rotation in the scene. For doors, must be 0, 90, 180, or 270
• wall_material (string, or list of strings): The material for the wall around the door.
• wall_scale_x (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Scale of the walls around the door in the x direction. Default: A random value between 2 and the size of the room in the direction parallel with the door and wall.
• wall_scale_y (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Scale of the walls around the door in the y direction. The door will be 2 units high, so this scale must be greater than 2 for the top wall to appear. Default: A random value between 2 and the height of the room.

StructuralDropperConfig

Defines details of a structural dropper and its dropped projectile.

• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of structures to be created with these parameters
• drop_step (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): The step of the simulation in which the projectile should be dropped.
• labels (string, or list of strings): A label or labels to be assigned to this object. Always automatically assigned "droppers"
• position_relative (RelativePositionConfig dict, or list of RelativePositionConfig dicts): Configuration options for positioning this object relative to another object, rather than using position_x or position_z. If configuring this as a list, then all listed options will be applied to each scene in the listed order, with later options overriding earlier options if necessary. Default: not used
• position_x (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Position in the x direction of the of the ceiling where the dropper should be placed.
• position_z (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Position in the z direction of the of the ceiling where the dropper should be placed.
• projectile_dimensions (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Dimensions of the projectile. Overrides the projectile_scale if configured. Default: use projectile_scale
• projectile_labels (string, or list of strings): A label for an existing object in your ILE configuration that will be used as this device's projectile, or new label(s) to associate with a new projectile object. Other configuration options may use this label to reference this object or a group of objects. Labels are not unique, and when multiple objects share labels, the ILE may choose one available object or all of them, depending on the specific option. The ILE will ignore any objects that have keyword locations or are used by other droppers/placers/throwers.
• projectile_material (string, or list of strings): The projectiles's material or material type.
• projectile_scale (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Scale of the projectile. Default is based on the shape.
• projectile_shape (string, or list of strings): The shape or type of the projectile.
• no_projectile (bool, or list of bools): If true, this device will not be holding a projectile; other "projectile" options will be ignored. Default: false

StructuralLOccluderConfig

Defines details of a structural L-shaped occluder.

• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of structures to be created with these parameters
• backwards (bool, or list of bools): Whether to create a backwards L. Default: [true, false]
• labels (string, or list of strings): A label or labels to be assigned to this object. Always automatically assigned "l_occluders"
• material (string, or list of strings): The structure's material or material type.
• position (VectorFloatConfig dict, or list of VectorFloatConfig dicts): The structure's position in the scene
• rotation_y (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The structure's rotation in the scene
• scale_front_x (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Scale in the x direction of the front part of the occluder
• scale_front_z (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Scale in the z direction of the front part of the occluder
• scale_side_x (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Scale in the x direction of the side part of the occluder
• scale_side_z (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Scale in the z direction of the side part of the occluder
• scale_y (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Scale in the y direction for the entire occluder

StructuralMovingOccluderConfig

Defines details of a structural moving occluder.

• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of structures to be created with these parameters
• labels (string, or list of strings): A label or labels to be assigned to this object. Always automatically assigned "moving_occluders"
• occluder_height (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Height (Y scale) of the occluder wall. Default is between .25 and 2.5.
• occluder_thickness (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Thickness (Z scale) of the occluder wall. Default is between .02 and 0.5.
• occluder_width (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Width (X scale) of the occluder wall. Default is between .25 and 4.
• origin (string, or list of strings): Location that the occluder's pole will originate from. Options are top, front, back, left, right. Default is weighted such that top occurs 50% of the time and the sides are each 12.5%. Users can weight options by included them more than once in an array. For example, the default can be represented as:

['top', 'top', 'top', 'top', 'front', 'back', 'left', 'right']

• pole_material (string, or list of strings): Material of the occluder pole (cylinder)
• position_x (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): X position of the center of the occluder
• position_z (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Z position of the center of the occluder
• move_down_only (bool, or list of bools): If true, occluder will start near the ceiling, moving downwards until it touches the floor with no rotation. Note that if this is true, the following settings are ignored: move_up_before_last_step, repeat_movement, and reverse_direction. Default: false
• move_up_before_last_step (bool, or list of bools): If true, repeat the occluder's full movement and rotation before the scene's last step. Ignored if last_step isn't configured, or if repeat_movement is true. Default: false
• repeat_interval (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): if repeat_movement is true, the number of steps to wait before repeating the full movement. Default is between 1 and 20.
• repeat_movement (bool, or list of bools): If true, repeat the occluder's full movement and rotation indefinitely, using repeat_interval as the number of steps to wait. Default: false
• reverse_direction (bool, or list of bools): Reverse the rotation direction of a sideways wall by rotating the wall 180 degrees. Only used if origin is set to a wall and not top. Default: [true, false]
• rotation_y (float, or list of floats, or MinMaxInt dict, or list of MinMaxFloat dicts): Y rotation of a non-sideways occluder wall; only used if any origin is set to top. Default is 0 to 359.
• wall_material (string, or list of strings): Material of the occluder wall (cube)

StructuralNotchedOccluderConfig

• height (float): The overall height of the occluder.
• material(string): The material the occluder is made of.
• position_z (float): The occulders z position within the room.
• down_step (int): The step the occluder will start to move down.
• up_step (int): The step the occluder will start to return up.
• notch_height (float): The height of the occluder notch.
• notch_width (float): The width of the occluder notch.

StructuralObjectMovementConfig

Represents what movements the structural object will make. Currently only used for configuring turntables.

• end_after_rotation (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): The amount the structure will rotate in full, beginning on step_begin and rotating rotation_y each step. All values will be % 360 (set 360 for a complete rotation). If end_after_rotation does not divide evenly by rotation_y, then the number of rotation steps will round down. Default is either 90, 180, 270, or 360.
• step_begin (int, or list of ints, or MinMaxInt dict): The step at which this movement should start. Default is a value between 0 and 10.
• step_end (int, or list of ints, or MinMaxInt dict): The step at which this movement should end. Overrides end_after_rotation. Default will use end_after_rotation.
• rotation_y (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): The amount that the structure will rotate each step. Default is randomly either 5 or -5.

StructuralOccludingWallConfig

• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of walls to be created with these parameters
• keyword_location: (KeywordLocationConfig): Used to identify one of the qualitative locations specified by keywords. This field should not be set when position or rotation are also set.
• labels (string, or list of strings): A label or labels to be assigned to this object. Always automatically assigned "occluding_walls"
• material (string, or list of strings): The wall's material or material type.
• position (VectorFloatConfig dict, or list of VectorFloatConfig dicts): The wall's position in the scene. Will be overrided by keyword location.
• rotation_y (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The wall's rotation in the scene. Will be overrided by keyword location.
• scale (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Scale of the wall. Default is scaled to target size. This will override the scale provided by the type field.
• type (string, or list of strings): describes the type of occluding wall. Types include: occludes - occludes the target or object. short - wide enough, but not tall enough to occlude the target. thin - tall enough, but not wide enough to occlude the target. hole - wall with a hole that reveals the target. Default: ['occludes', 'occludes', 'occludes', 'short', 'thin', 'hole']

StructuralPlacerConfig

Defines details for an instance of a placer (cylinder) descending from the ceiling on the given activation step to place an object with the given position. For some object shapes (specifically container_symmetric_* and container_asymmetric_*), two placers will be made and attached to the object.

• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of areas to be used with these parameters
• activate_after: (str, or list of strs): Overrides the activation_step (overriding the manual config) based on the movement of other object(s) in the scene. Should be set to one or more labels for mechanical objects that may move or rotate, like placers or turntables, as well as agents. The activation_step of this object will be set to the step immediately after ALL of the objects and agents finish moving and rotating. If multiple labels are configured, all labels will be used. Default: Use activation_step
• activate_on_start_or_after: (str, or list of strs, or bool): Overrides the activation_step (overriding the manual config) based on the movement of other object(s) in the scene. Should be set to one or more labels for mechanical objects that can move or rotate, like placers or turntables, as well as agents. If ANY of the objects or agents begin moving or rotating immediately at the start of the scene (step 1), then the activation_step of this object will be set to the step immediately after ALL of the objects and agents finish moving and rotating; otherwise, if ALL of the objects and agents begin moving and rotating after step 1, then the activation_step of this object will be set to 1. If multiple labels are configured, all labels will be used. Default: Use activation_step
• activation_step: (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Step on which the placer should begin its downward movement. Default: between 0 and 10
• deactivation_step: (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Step on which the placer should release its held object. This number must be a step after the end of the placer's downward movement. Default: At the end of the placer's downward movement
• empty_placer (bool, or list of bools): If True, the placer will not hold, drop, or move an object. Use 'placed_object_position' to set the placer's starting X/Z position if needed. Cannot be used in combination with any of the following config options: 'placed_object_labels', 'placed_object_material', 'placed_object_rotation', 'placed_object_scale', 'placed_object_shape'. Default: False
• end_height: (float, or list of floats, or MinMaxFloat dict): Height at which the placer should release its held object. Alternatively, one can use the end_height_relative_object_label. Default: 0 (so the held object is in contact with the floor)
• end_height_relative_object_label (string): Label used to match the bottom of the object held by the placer to the height of another (for example, the support platform in gravity scenes). This will override end_height if both are set.
• existing_object_required (bool, or list of bools): If this is true and placed_object_labels is set, this placer will be required to use an existing object with the configured label. If this is false, and no objects exist with the placed_object_labels, then the placer will automatically generate a new object and assign it all the placed_object_labels. Default: False
• labels (string, or list of strings): A label or labels to be assigned to this object. Always automatically assigned "placers"
• move_object (bool): If True, the placer will move the object on the X axis from its placed_object_position to move_object_end_position. Default: False
• move_object_end_position: (VectorFloatConfig dict, or list of VectorFloatConfig dicts): If move_object is True, the placer will move the object from its placed_object_position to the X position in move_object_end_position. Note that the Y and Z positions in move_object_end_position are ignored; the Y and Z positions set in placed_object_position are kept the same. (This option remains a Vector for backwards compatibility.)
• move_object_y: If move_object is True, the placer will raise the object by this amount before moving it on the X/Z axes, and lower it by the same amount afterward. Default: 0
• move_object_z: If move_object is True, the placer will move the object by this amount in the -Z direction after raising it on the Y axis but before moving it on the X axis, and move it by the same amount again in the +Z direction after moving it on the X axis and before lowering it on the Y axis. Default: 1.5
• pickup_object (bool): If True, the placer will pickup the object. Default: False
• placed_object_above (string, or list of strings): A label for an existing object in your configuration whose X/Z position will be used for this placer's (and the placed object's) starting position. Overrides placed_object_position. Please use end_height_relative_object_label if you need to set the held object's ending Y position.
• placed_object_labels (string, or list of strings): A label for an existing object in your configuration that will be used as this placer's placed object, or new label(s) to associate with a new placed object. Other configuration options may use this label to reference this object or a group of objects. Labels are not unique, and when multiple objects share labels, the ILE may choose one available object or all of them, depending on the specific option. The ILE will ignore any objects that have keyword locations or are used by other droppers/placers/throwers.
• placed_object_material (string, or list of strings): The material (color/texture) to use on the placed object in each scene. For a list, a new material will be randomly chosen for each scene. Default: random
• placed_object_position: (VectorFloatConfig dict, or list of VectorFloatConfig dicts): The placed object's position in the scene
• placed_object_rotation: (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The placed object's rotation on the y axis.
• placed_object_scale: (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Placed object's scale. Default is a value between 0.2 and 2.
• placed_object_shape (string, or list of strings): The shape (object type) of the placed object. For a list, a new shape will be randomly chosen for each scene. Default: random
• position_relative (RelativePositionConfig dict, or list of RelativePositionConfig dicts): Configuration options for positioning this object relative to another object, rather than using placed_object_position. If configuring this as a list, then all listed options will be applied to each scene in the listed order, with later options overriding earlier options if necessary. Default: not used
• randomize_once (StructuralPlacerConfig dict) Placer configuration options that are only randomized once. If this configuration template would generate more than one placer (has a num greater than 1), then each placer generated by this template will use the same randomized values for all the options in randomize_once. Default: not used
• retain_position (bool, or list of bools): The placed object will retain its current X/Z position. Overrides placed_object_position and `placed_object_above'. Default: False

StructuralPlatformConfig

Defines details of a structural platform. The top of a platform should never exceed room_dimension_y - 1.25 if a target or performer are to be placed on top of the platform to ensure the performer can reach the target.

• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of structures to be created with these parameters
• attached_ramps (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of ramps that should be attached to this platform to allow the performer to climb up to this platform. Default: 0
• labels (string, or list of strings): A label or labels to be assigned to this object. Always automatically assigned "platforms"
• lips ([StructuralPlatformLipsConfig] (#StructuralPlatformLipsConfig), or list of StructuralPlatformLipsConfig): The platform's lips. If set, the platform's X/Z scales will be increased to 0.8 if lower than 0.8. Default: no lips
• material (string, or list of strings): The structure's material or material type.
• platform_underneath (bool or list of bools): If true, add a platform below this platform that touches the floor on the bottom and this platform on the top. This platform will fully be encased in the x/z directions by the platform created underneath. Default: False
• platform_underneath_attached_ramps (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of ramps that should be attached to the platform created below this platform to allow the performer to climb onto that platform. Default: 0
• position (VectorFloatConfig dict, or list of VectorFloatConfig dicts): The structure's position in the scene
• rotation_y (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The structure's rotation in the scene
• scale (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Scale of the platform
• auto_adjust_platforms (bool or list of bools): If true, makes sure all platform heights do not exceed 1.25 units below the y room dimension allowing the performer to always stand on top of a platform. For example, a room with a room_dimension_y = 4 and auto_adjusted_platforms = True will ensure that all platform heights do not exceed 2.75. Default: False
• long_with_two_ramps (bool or list of bools): If true makes the platform extend across the entire length of the room in either the x or z dimension. The platform will have lips automatically added on all sides. A single ramp will be added on opposite sides of the platform totalling to two connected ramps. With larger room sizes the max number of these types of platforms is 3 though 1 is recommended for smaller room sizes. When combined with platform_underneath there will be a smaller platform of the same type stacked on top totalling 2 platforms and 4 ramps in the scene. Only one instance of a platform combined with platform_underneath can be added to the scene instead of the 3 allowed without a platform_underneath. This property will override attached_ramps, and platform_underneath_attached_ramps Default: False
• adjacent_to_wall (string or list of strings): Will place the platform directly adjacent to a wall or in the corner of the room. Can be combined with platform_underneath to place a stacked platform adjacent to a wall. Multiple platforms can be placed against the same wall or different walls but only one platform can be placed in each corner. Setting this will override rotation_y. Options available: 'left', 'right', 'front', 'back', 'front_left_corner', 'front_right_corner', 'back_left_corner', 'back_right_corner'. Default: None

StructuralPlatformLipsConfig

Defines the platform's lips with front, back, left, and right configurations.

• front (bool) : Positive Z axis
• back (bool) : Negative Z axis
• left (bool) : Negative X axis
• right (bool) : Positive X axis

StructuralRampConfig

Defines details of a structural ramp.

• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of structures to be created with these parameters
• angle (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Angle of the ramp upward from the floor
• labels (string, or list of strings): A label or labels to be assigned to this object. Always automatically assigned "ramps"
• length (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Length of the ramp along the floor. This is the 'adjacent' side and not the hypotenuse.
• material (string, or list of strings): The structure's material or material type.
• platform_underneath (bool or list of bools): If true, add a platform below this ramp that touches the floor on the bottom and the bottom of this ramp on the top. This ramp will fully be encased in the x/z directions by the platform created underneath. Default: False
• platform_underneath_attached_ramps (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of ramps that should be attached to the platform created below this ramp to allow the performer to climb onto that platform. Default: 0
• position (VectorFloatConfig dict, or list of VectorFloatConfig dicts): The structure's position in the scene
• rotation_y (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The structure's rotation in the scene
• width (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Width of the ramp

StructuralThrowerConfig

Defines details of a structural dropper and its thrown projectile.

• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of structures to be created with these parameters
• height (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The height on the wall that the thrower will be placed.
• impulse (bool, or list of bools): Whether to use "impulse" force mode. We recommend using impulse force mode moving forward. Please note that the default throw_force is different for impulse and non-impulse force modes. Default: true
• labels (string, or list of strings): A label or labels to be assigned to this object. Always automatically assigned "throwers"
• passive_physics_collision_force (bool, or list of bools): Automatically set the throw_force to a speed normally used in a passive physics collision scene. If set, overrides the throw_force. Default: false
• passive_physics_setup (string, or list of strings): Automatically set the wall, position_wall, rotation_y, rotation_z, and heightto values normally used in passive physics non-collision scenes. If set, this will override other config options (see below). Possible settings:"random", "roll_angled", "roll_straight", "toss_straight"`. Default: null
  • wall: ['left', 'right']
  • If "roll_angled": a height of 0 and a rotation_y of either -20 or 20
  • If "roll_straight": a height of 0 and a rotation_y of 0
  • If "toss_straight": a height of 1 and a rotation_y of 0
  • If "roll_angled": a position_wall of 1.6 if rotation_y is -20 or 5.6 if rotation_y is 20
  • If "roll_straight" or "toss_straight": a position_wall between 1.6 and 4.4
• passive_physics_throw_force (bool, or list of bools): Automatically set the throw_force to a speed normally used in passive physics non-collision scene. If set, overrides the throw_force. Default: false
• path_relative (RelativePathConfig dict): Configuration options for rotating this thrower so the thrown object will either collide with, or avoid, another object. Overrides the rotation_y. Default: not used
• position_relative (RelativePositionConfig dict, or list of RelativePositionConfig dicts): Configuration options for positioning this object relative to another object, rather than using position_wall. If configuring this as a list, then all listed options will be applied to each scene in the listed order, with later options overriding earlier options if necessary. Default: not used
• position_wall (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The position along the wall that the thrower will be placed.
• projectile_dimensions (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Dimensions of the projectile. Overrides the projectile_scale if configured. Default: use projectile_scale
• projectile_labels (string, or list of strings): A label for an existing object in your ILE configuration that will be used as this device's projectile, or new label(s) to associate with a new projectile object. Other configuration options may use this label to reference this object or a group of objects. Labels are not unique, and when multiple objects share labels, the ILE may choose one available object or all of them, depending on the specific option. The ILE will ignore any objects that have keyword locations or are used by other droppers/placers/throwers.
• projectile_material (string, or list of strings): The projectiles's material or material type.
• projectile_scale (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Scale of the projectile. Default is based on the shape.
• projectile_shape (string, or list of strings): The shape or type of the projectile.
• rotation_y (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The angle in which the thrower will be rotated from its original position on the wall to point sideways (either left or right), with 0 being the center. This value should be between -45 and 45. Default: random value between -45 and 45.
• rotation_z (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The angle in which the thrower will be rotated to point upwards. This value should be between 0 and 15. Default: random value between 0 and 15.
• stop_position (StopPositionConfig dict, or list of StopPositionConfig dicts): Sets a stop position for the thrown object. If set, overrides all other "throw force" options.
• throw_force (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Force of the throw put on the projectile. This value will be multiplied by the mass of the projectile. Default: between 5 and 20 for impulse force mode, or between 500 and 2000 for non-impulse force mode
• throw_force_multiplier (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Force of the throw put on the projectile, that will be multiplied by the appropriate room dimension for the thrower's wall position (X for left/right, Z for front/back). If set, overrides the throw_force.
• throw_step (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): The step of the simulation in which the projectile should be thrown. Please note that using a value of less than 5 may cause unexpected behavior, so we recommend using values of 5 or more in your custom config files.
• wall (string, or list of strings): Which wall the thrower should be placed on. Options are: left, right, front, back.
• no_projectile (bool, or list of bools): If true, this device will not be holding a projectile; other "projectile" options will be ignored. Default: false

StructuralTubeOccluderConfig

Defines details of a structural tube occluder (or "tube-cluder").

• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of structures to generate with these options.
• labels (string, or list of strings): A label or labels to assign to these object(s). Always automatically assigned "tube_occluders"
• down_after: (string, or list of strings): Overrides the down_step (overriding the manual config) based on the movement of other object(s) in the scene. Should be set to one or more labels for mechanical objects that may move or rotate, like placers or turntables, as well as agents. The down_step of this object will be set to the step immediately after ALL of the objects and agents finish moving and rotating. If multiple labels are configured, all labels will be used. Default: Use down_step
• down_step (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): The step on which the tube occluder should begin to move down from the ceiling to the floor. Note that this should happen before up_step. Default: between 1 and 10
• material (string, or list of strings): The structure's material or material type. Default: random
• position_x (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The structure's X position. Default: random
• position_z (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The structure's Z position. Default: random
• radius (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The structure's radius. Default: between 0.5 and 5
• up_after: (string, or list of strings): Overrides the up_step (overriding the manual config) based on the movement of other object(s) in the scene. Should be set to one or more labels for mechanical objects that may move or rotate, like placers or turntables, as well as agents. The up_step of this object will be set to the step immediately after ALL of the objects and agents finish moving and rotating. If multiple labels are configured, all labels will be used. Default: Use up_step
• up_step (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): The step on which the tube occluder should begin to move up from the floor to the ceiling. Note that this should happen after down_step. Default: between 51 and 60

StructuralTurntableConfig

Defines details of a structural turntable (also sometimes referred to as a rotating cog).

• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of structures to be created with these parameters
• labels (string, or list of strings): A label or labels to be assigned to this object. Always automatically assigned "turntables"
• material (string, or list of strings): The structure's material or material type. Default: "Custom/Materials/GreyWoodMCS"
• position (VectorFloatConfig dict, or list of VectorFloatConfig dicts): The structure's position in the scene
• rotation_y (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The structure's rotation in the scene. Default is 0.
• turntable_height (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Height of the turntable/its y-axis scale. Default is 0.1.
• turntable_radius (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Radius of the turntable. Will be used to scale the turntable in both x and z directions. Default is a value between 0.5 and 1.5.
• turntable_movement (StructuralObjectMovementConfig) or list of (StructuralObjectMovementConfig): The config for turntable movement.

StructuralWallConfig

Defines details of a structural interior wall. The wall will be the height of the room.

• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of structures to be created with these parameters
• labels (string, or list of strings): A label or labels to be assigned to this object. Always automatically assigned "walls"
• material (string, or list of strings): The structure's material or material type.
• position (VectorFloatConfig dict, or list of VectorFloatConfig dicts): The structure's position in the scene
• rotation_y (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The structure's rotation in the scene
• width (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The width of the wall. Default: Between 5% and 50% of the room's maximum dimensions.
• thickness (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The thickness of the wall. Default: 0.1
• height (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The height of the wall. Default: The room's height.
• same_material_as_room (bool, or list of bools): Whether to use the same material as the room's walls. If true, will ignore any material configured on this wall. Default: false
• ignore_bounds (bool, or list of bools): Whether to ignore the bounds of all other objects that have already been generated at the moment this wall is being generated. Default: false

TeleportConfig

Contains data to describe when and where a teleport occurs.

• step (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): The step when the performer agent is teleported. This field is required for teleport action restrictions.
• look_at_center (bool): Dynamically set the teleport rotation_y using the position_x and position_z so the performer agent is facing the center of the room. Requires both position_x and position_z to be set. Overrides rotation_y if it is also set. Default: false
• position_x (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Position in X direction where the performer agent is teleported. This field along with position_z are required if rotation_y is not set.
• position_z (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Position in Z direction where the performer agent is teleported. This field along with position_x are required if rotation_y is not set.
• rotation_y (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): Rotation in Y direction where the performer agent is teleported. This field is required for teleport action restrictions if position_x and position_z are not both set.

ToolChoiceConfig

Defines details of the shortcut_tool_choice shortcut. This shortcut creates a room with bisecting platform, with two identical lava islands on either side. The performer has to choose the side with the tool that can be used to obtain the target.

• improbable_option (str, or list of strs): Determines the tool (if any) that will be placed on the side where it is not possible to obtain the ball without stepping in lava. Possible values: 'no_tool', 'too_short'.

ToolConfig

Defines details of a tool object.

• num (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Number of tools to generate with these parameters
• align_distance (float, or MinMaxFloat dict, or list of floats and/or MinMaxFloat dicts): The distance separating the edges of two objects. Only used if align_with is also set. Default: between 1 and 2
• align_with (string, or list of strings): The label of an object (that already exists in your configuration) with which this tool should be aligned (this tool will be positioned behind the object, so you can push this tool forward into the object). Overrides any configured position and rotation_y. Use align_distance to set the distance separating the edges of the two objects. Default: Use position
• labels (string, or list of strings): A label or labels to be assigned to this object. Always automatically assigned "tools"
• length (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): The length of the tool. Tools only have specific sizes and the values much match exactly. Valid lengths are integers 4 to 9. If shape is set, this value is ignored. Default: random
• position (VectorFloatConfig dict, or list of VectorFloatConfig dicts): The tool's position in the scene
• rotation_y (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The tool's rotation in the scene
• shape (string, or list of strings): The shape (object type) of this object in each scene. For a list, a new shape will be randomly chosen for each scene. Must be a valid tool shape. If set, ignores length, width, and tool_type. Default: Use tool_type
• tool_type (str, or list of strs): The type of tool to generate, either rectangular, hooked, isosceles, or small. Both hooked and isoscelestools are L-shaped;hookedtools always have width 3, andisoscelestools always have width equal to their length. Default:rectangularorhooked`
• width (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The width of the tool. Tools only have specific sizes and the values much match exactly. Valid widths are 0.5, 0.75, 1.0. If shape is set, this value is ignored. Default: random
• material (string, or list of strings): The material (color/texture) or material category to use on this tool in each scene. For lists, a new material will be randomly chosen for each scene. Default: TOOL_MATERIALS
• surrounded_by_lava (bool, or list of bools): Whether or not this tool will be surrounded by lava. If True, width of lava will be 1. Default: False
• surrounding_lava_size: (int, or list of ints): The width of the surrounding lava. Only used if surrounded_by_lava is set to True. In that case, default is 1, otherwise the value will be None. Note that this value will need to fit within room bounds, otherwise may error.
• surrounding_safe_zone: (bool, or list of bools): If set to True and surrounded_by_lava is True, a buffer area (at least 0.5) will be set around the surrounding lava to ensure that it does not touch any other sections of lava or room walls/corners, and therefore allowing the performer to move around the surrounding lava's perimeter.

TripleDoorConfig

Defines details of the triple door shortcut. This short cut contains a platform that bisects the room. A wall with 3 doors (a.k.a. door occluder, or "door-cluder" or "doorcluder") will bisect the room in the perpendicular direction.

• add_extension (bool, or list of bools): If true, add an extension to one side of the far end (+Z) of the platform with the label "platform_extension". Default: False
• add_freeze (bool, or list of bools): If true and 'start_drop_step is' greater than 0, the user will be frozen (forced to Pass) until the wall and doors are in position. If the 'start_drop_step' is None or less than 1, this value has no effect. Default: True
• add_lips (bool, or list of bools): If true, lips will be added on the platform beyond the doors and wall such that the performer will be forced to go through the corresponding door to enter the area behind it. Default: True
• bigger_far_end (bool, or list of bools): If true, increases the height and the width of the far end of the platform so it's too tall for the performer to walk onto and it's twice as wide. Default: false
• door_material (string, or list of strings): The material or material type for the doors.
• extension_length (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): If add_extension is true, set the length of the extension (on the X axis). Default: Up to half of the room's X dimension. otherwise, don't add a platform extension.
• extension_position (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): If add_extension is true, set the position of the extension (on the Z axis). If the position is positive, the extension will be on the right side (+X). If the position is negative, the extension will be on the left side (-X). Default: Up to half of the room's Z dimension.
• restrict_open_doors (bool, or list of bools): If true, the performer will only be able to open one door. Using this feature and 'add_lips' will result in a forced choice by the performer. Default: True
• start_drop_step (int, or list of ints, or MinMaxInt dict, or list of MinMaxInt dicts): Step number to start dropping the bisecting wall with doors. If None or less than 1, the wall will start in position. Default: None
• wall_height (float, or list of floats, or MinMaxFloat dict, or list of MinMaxFloat dicts): The height for the wall. The height for the center door will be -2 to account for being on top of the platform.
• wall_material (string, or list of strings): The material or material type for the wall.

TurntablesAgentNonAgentConfig

Defines all of the configurable options for [turntables_with_agent_and_non_agent] (#turntables_with_agent_and_non_agent).

• agent_label (string): The label for an existing agent on top of one of the turntables. The turntable underneath this agent will NOT rotate.
• non_agent_label (string): The label for an existing non-agent object on top of one of the turntables. The turntable underneath this object will rotate so the object faces one of the objects corresponding to the direction_labels. The turntable may rotate either clockwise or counter-clockwise, but always in 5-degree-per-step increments. The rotation will begin after the agent's movement, unless the agent's movement begins after step 45, in which case the rotation will begin on step 1.
• turntable_labels (list of strings): The labels for all turntables to be affected.
• direction_labels (list of strings): One or more labels for the objects in the scene toward which the non-agent object may face. If multiple labels are configured, then one object is randomly chosen for each scene.

VectorFloatConfig

A dict with x, y, and z properties (each of which are either a float, list of floats, or MinMaxFloat) that represents a global coordinate in the scene. For each list, one float will be randomly chosen from the list in each new scene; for each MinMaxFloat, one float will be randomly chosen from within the corresponding range in each new scene.

Example:

x:
    - -0.5
    - -0.25
    - 0
    - 0.25
    - 0.5
y: 0.1
z:
    min: -0.5
    max: 0.5

VectorIntConfig

A dict with x, y, and z properties (each of which are either an int, list of ints, or MinMaxInt) that represents a global coordinate in the scene. For each list, one int will be randomly chosen from the list in each new scene; for each MinMaxInt, one int will be randomly chosen from within the corresponding range in each new scene.

Example:

x:
    - -1
    - 0
    - 1
y: 0
z:
    min: -1
    max: 1

Options

You may set the following options in your ILE (YAML) config file:

auto_last_step

(bool, or list of bools): Determines if the last step should automatically be determined by room size. The last step is calculated such that the performer can walk in a circle along the walls of the room approximately 5 times. If 'last_step' is set, that field takes precedence. Default: False

Simple Example:

auto_last_step: False

Advanced Example:

auto_last_step: True

ceiling_material

(string, or list of strings): A single material for the ceiling, or a list of materials for the ceiling, from which one is chosen at random for each scene. Default: random

Simple Example:

ceiling_material: null

Advanced Example:

ceiling_material: "Custom/Materials/GreyDrywallMCS"

check_valid_path

(bool or str): If true, checks for a valid path between the performer agent's starting position and the target's position and retries generating the current scene if one cannot be found. Considers all objects and structures that would block the performer when their position is y = 0 or are light enough to be pushed. The check considers all holes and areas of lava in the scene. It also considers moving up and/or down ramps that are attached to platforms (via the attached_ramps option in structural_platforms), as well as across those platforms. Pathfinding is otherwise only done in two dimensions. This check is skipped if false.

Can also be set as a string corresponding to the label of one or more objects that already exist in the scene.

Default: False

Simple Example:

check_valid_path: false

Advanced Example:

check_valid_path: true

circles

(list of either ints, or lists of ints, or MinMaxInt dicts, or lists of MinMaxInt dicts): When the AI should be forced to rotate in a complete circle counterclockwise (by only allowing RotateRight actions for 36 consecutive actions). This field must be blank or an empty array if passive_scene is true.

Simple Example:

circles: null

Advanced Example:

circles:
  - 7
  - [107, 207]

doors

(StructuralDoorConfig, or list of StructuralDoorConfig dict) -- Groups of door configurations and how many should be generated from the given options. Note: Doors do not contain any frame or wall support. Default: 0

Simple Example:

doors:
    - num: 0

Advanced Example:

doors:
    - num:
        min: 1
        max: 3
    - num: 1
      position:
        x: [1,2]
        y: 0
        z:
          min: -3
          max: 3
      scale: [1, 0.5]
      material: PLASTIC_MATERIALS
    - num: [1, 3]
      position:
        x: [4, 5]
        y: 0
        z:
          min: -5
          max: -4
      scale:
        x: 1
        y:
          min: .5
          max: 2
        z: [.75, 1.25]
      material: AI2-THOR/Materials/Metals/BrushedAluminum_Blue

excluded_colors

(string, or list of strings): Zero or more color words to exclude from being randomly generated as object or room materials. Materials with the listed colors can still be generated using specifically set configuration options, like the floor_material and wall_material options, or the material property in the specific_interactable_objects option. Useful if you want to avoid generating random objects with the same color as a configured object. Default: no excluded colors

Simple Example:

excluded_colors: null

Advanced Example:

excluded_colors: "red"

excluded_shapes

(string, or list of strings): Zero or more object shapes (types) to exclude from being randomly generated. Objects with the listed shapes can still be generated using specifically set configuration options, like the type property in the goal.target and specific_interactable_objects options. Useful if you want to avoid randomly generating additional objects of the same shape as a configured goal target. Default: no excluded shapes

Simple Example:

excluded_shapes: null

Advanced Example:

excluded_shapes: "soccer_ball"

floor_material

(string, or list of strings): A single material for the floor, or a list of materials for the floor, from which one is chosen at random for each scene. Default: random

Simple Example:

floor_material: null

Advanced Example:

floor_material: "Custom/Materials/GreyCarpetMCS"

floor_material_override

(FloorMaterialConfig, or list of FloorMaterialConfig dict) -- Groups of floor material configurations and how many should be generated from the given options. Default: 0

Simple Example:

floor_material_override:
    - num: 0

Advanced Example:

floor_material_override:
    - num:
        min: 0
        max: 2
    - num: 1
      position_x: 2
      position_z: 2
      material: PLASTIC_MATERIALS
    - num: [1, 3]
      position_x: [4, 5]
      position_z:
        min: -5
        max: -4
      material: AI2-THOR/Materials/Metals/BrushedAluminum_Blue

forced_choice_multi_retrieval_target

(str): Whether to set a new "multi retrieval" goal using objects of a specific type that already exist in the scene. Set this option to an object shape like "soccer_ball". This option splits all matching objects into "left" and "right" groups based on their X positions (ignoring objects that are picked up by placers). All objects in the larger group will be used as the new goal's target(s). Overrides the configured goal. Default: not used

Simple Example:

forced_choice_multi_retrieval_target: null

Advanced Example:

forced_choice_multi_retrieval_target: 'soccer_ball'

freeze_while_moving

(string or list of strings): Forces the performer to freeze ("Pass") until all objects with the given labels complete their last movements and rotations. This field must be blank or an empty array if passive_scene is true.

Simple Example:

freeze_while_moving: null

Advanced Example:

freezes_while_moving:
  - [placers, turntables]
  - [turntables]

freezes

(list of StepBeginEnd dicts): When a freeze should occur. A freeze forces the performer agent to only "Pass" for a range of steps. User should try to avoid freeze overlaps, but if using ranges and choices, the ILE will retry on overlaps. This field must be blank or an empty array if passive_scene is true.

Simple Example:

freezes: null

Advanced Example:

freezes:
  -
    begin: 1
    end: 3
  -
    begin: [11, 13 ,15]
    end:
      min: 16
      max: 26

goal

(GoalConfig dict): The goal category and target(s) in each scene, if any. Default: no goal

Simple Example:

goal: null

Advanced Example:

goal:
    category: retrieval
    target:
        shape: soccer_ball
        scale:
          min: 1.0
          max: 3.0

holes

(FloorAreaConfig, or list of FloorAreaConfig dict) -- Groups of hole configurations and how many should be generated from the given options. Note: Holes can end up appearing below objects which can cause the objects to fall into the hole. Default: 0

Simple Example:

holes:
    - num: 0

Advanced Example:

holes:
    - num:
        min: 0
        max: 2
    - num: 1
      position_x: 2
      position_z: 2
    - num: [1, 3]
      position_x: [4, 5]
      position_z:
        min: -5
        max: -4

keyword_objects

(KeywordObjectsConfig dict, or list of KeywordObjectsConfig dicts): Creates interactable objects of specific categories like confusors, containers, obstacles, or occluders. Default:

• 0 to 10 context objects
• 2 to 4 of container, obstacles, and occluders

keyword_objects:
    - keyword: ["containers", "obstacles", "occluders"]
    num:
        min: 2
        max: 4
    - keyword: "context"
        min: 0
        max: 10

All objects created here will be given the following labels. For more information see (InteractableObjectConfig) for more information on labels.

• asymmetric_containers: keywords_asymmetric_containers
• bins: keywords_bins
• confusors: keywords_confusors
• containers: keywords_containers
• containers_can_contain_target: keywords_containers, keywords_containers_can_contain_target
• containers_cannot_contain_target: keywords_containers, keywords_containers_cannot_contain_target
• context: keywords_context
• obstacles: keywords_obstacles
• occluders: keywords_occluders
• open_topped_containers: keywords_open_topped_containers
• symmetric_containers: keywords_symmetric_containers

Simple Example:

keyword_objects: null

Advanced Example:

keyword_objects:
  -
    keyword: confusors
    num: 1
  -
    keyword: containers
    num:
      min: 0
      max: 3
  -
    keyword: containers_can_contain_target
    num: 1
    keyword_location:
      keyword: front
  -
    keyword: containers_cannot_contain_target
    num: 1
  -
    keyword: obstacles
    num: [2, 4]
    keyword_location:
      keyword: adjacent
      relative_object_label: other_object
  -
    keyword: occluders
    num: 1
  -
    keyword: context
    num: 20

Note: Creating too many random objects can increase the chance of failure.

knowledgeable_agent_pair

(bool, KnowledgeableAgentPairConfig): Config to create a pair of agents. One agent will be knowledgeable (will know where the goal is), the other won't know and will point into space.

Simple Example:

knowledgeable_agent_pair:
        position_1:
            x: 1.0
            y: 0
            z: 0
        position_2:
            x: -1.0
            y: 0
            z: 0

Advanced Example Simple Example:

knowledgeable_agent_pair:
    position_1:
        x: 1.0
        y: 0
        z: 0
    position_2:
        x: -1.0
        y: 0
        z: 0
    pointing_step: 10,
    target_labels: target

#### last_step

(int, or list of ints): The last possible action step, or list of last
steps, from which one is chosen at random for each scene. This field will
overwrite 'auto_last_step' if set.  Default: no last step
(unlimited)

Simple Example:

last_step: null

Advanced Example:

last_step: 1000

#### lava

([FloorAreaConfig](#FloorAreaConfig), or list of
[FloorAreaConfig](#FloorAreaConfig) dict) --
Groups of lava configurations. Default: 0


Simple Example:

lava: - num: 0

Advanced Example:

lava: - num: min: 0 max: 2 - num: 1 position_x: 1 position_z: 1 - num: [1, 3] position_x: [-5, -4] position_z: min: -5 max: -4

#### num_random_agents

(int, or [MinMaxInt](#MinMaxInt) dict, or list of ints or
[MinMaxInt](#MinMaxInt) dicts): The number of random agents to add to the
scene.  Default: `0`

Simple Example:

num_random_agents: 0

Advanced Example:

num_random_agents: min: 2 max: 4

#### num_random_interactable_objects

(int, or [MinMaxInt](#MinMaxInt) dict): The number of random interactable
objects that will be added to each scene. Default: 0

Simple Example:

num_random_interactable_objects: null

Advanced Example:

num_random_interactable_objects: min: 1 max: 10

#### occluder_gap

(float, or list of floats, or [MinMaxFloat](#MinMaxFloat) dict: Gap (X
distance) between moving structual occluders. Will override
`position_x`. Only applies when `passive_physics_scene` is True.
Default: no restrictions

All scenes are generated with a .5 gap between occluders.

Simple Example:

occluder_gap: .5

All scenes are generated with .5 or 1.0 gap between occluders.

Advanced Example:

occluder_gap: [.5, 1.0]

All scenes are generated with a gap between (inclusive) .5 and 1.0 between
occluders.

Advanced Example:

occluder_gap: min: .5 max: 1.0

#### occluder_gap_viewport

(float, or list of floats, or [MinMaxFloat](#MinMaxFloat) dict: Gap (X
distance) between occluders and edge of viewport. If both are included,
`occluder_gap` will take precedence over `occluder_gap_viewport`.
Will override `position_x`.
Only applies when `passive_physics_scene` is True.
Default: no restrictions

All scenes are generated with a .5 gap between the occluder and the
viewport.

Simple Example:

occluder_gap_viewport: .5

All scenes are generated with .5 or 1.0 gap between occluder and the
viewport.

Advanced Example:

occluder_gap_viewport: [.5, 1.0]

All scenes are generated with a gap between (inclusive) .5 and 1.0 between
the occluder nad the edge of the viewport.

Advanced Example:

occluder_gap_viewport: min: .5 max: 1.0

#### passive_physics_floor

**Deprecated.** Please use `passive_physics_scene: true`

(bool): Lowers the friction of the floor (making it more "slippery").
Used in passive physics evaluation scenes. Default: False

Simple Example:

passive_physics_floor: False

Advanced Example:

passive_physics_floor: True

#### passive_physics_scene

(bool): Setup each scene in exactly the same way as the "passive physics"
scenes used in the MCS evaluations, including the camera position, room
dimensions, floor friction, and restrictions on only "Pass" actions.
Default: False

Simple Example:

passive_physics_scene: False

Advanced Example:

passive_physics_scene: True

#### passive_scene

(bool): Determine if scene should be considered passive and the
performer agent should be restricted to only use the `"Pass"` action.
If true, ILE will raise an exception if last_step is not set or either
`circles`, `freezes`, `swivels` or `teleports` has any entries.
Redundant if the `passive_physics_scene` config option is `true`.
Default: `false`

#### performer_look_at

(string or list of strings): If set, configures the performer to start
looking at an object found by the label matching the string given.
Overrides `performer_start_rotation`.
Default: Use `performer_start_rotation`

Simple Example:

performer_look_at: null

Advanced Example:

performer_look_at: [target, agent]

#### performer_start_position

([VectorFloatConfig](#VectorFloatConfig) dict, or list of VectorFloatConfig
dicts or [KeywordLocationConfig](#KeywordLocationConfig)): The starting
position of the performer agent, or a list of
positions, from which one is chosen at random for each scene. The
(optional) `y` is used to position on top of structural objects like
platforms. Valid parameters are constrained by room dimensions.
`x` and `z` positions must be positioned within half of the room
dimension bounds minus an additional 0.25 to account for the performer
width. For example: in a room where 'dimension x = 5', valid
`x` parameters would be '4.75, MinMaxFloat(-4.75, 4.75) and
[-4.75, 0, 4.75].' In the case of variable room dimensions that use a
MinMaxInt or list, valid parameters are bound by the maximum room
dimension. For example: with 'dimension `x` = MinMax(5, 7) or [5, 6, 7]
valid x parameters would be '3.25, MinMaxFloat(-3.25, 3.25), and
[-3.25, 0, 3.25].' For `y` start and room dimensions, the min y position
must always be greater than 0 and the max must always be less than or equal
to room dimension y - 1.25.' This ensures the performer does not clip
into the ceiling. Alternatively, the performer start can be a
[KeywordLocationConfig](#KeywordLocationConfig).  Performer only supports
the following keyword locations: `along_wall`
Default: random within the room

Simple Example:

performer_start_position: null

Advanced Example:

performer_start_position: x: - -1 - -0.5 - 0 - 0.5 - 1 y: 0 z: min: -1 max: 1

#### performer_start_rotation

([VectorFloatConfig](#VectorFloatConfig) dict, or list of VectorFloatConfig
dicts): The starting rotation of the performer agent, or a list of
rotations, from which one is chosen at random for each scene. The
(required) `y` is left/right and (optional) `x` is up/down. Default: random

Simple Example:

performer_start_rotation: null

Advanced Example:

performer_start_rotation: x: 0 y: - 0 - 90 - 180 - 270

#### performer_starts_near

([PerformerStartsNearConfig](#PerformerStartsNearConfig) dict, or list
of PerformerStartsNearConfig dicts:
Dictates if the starting position of the performer will be near an
object of a given `label` at a specified `distance` away.
Use in combination with `performer_look_at` to look at the object at start.
Overrides `performer_start_position`.
Default: Use `performer_start_position`

Simple Example:

performer_starts_near: null

Advanced Example:

performer_starts_near: label: container distance: 0.5

#### placers

([StructuralPlacerConfig](#StructuralPlacerConfig), or list of
[StructuralPlacerConfig](#StructuralPlacerConfig) dict) --
Groups of Placer configurations and how many should be generated
from the given options.
Default: 0


Simple Example:

placers: - num: 0

Advanced Example:

placers: - num: min: 0 max: 2 - num: 1 placed_object_material: PLASTIC_MATERIALS placed_object_position: x: 2 y: [3, 4] z: min: -2 max: 0 placed_object_rotation: 45 placed_object_scale: 1.2 placed_object_shape: case_1 activation_step: min: 3 max: 8 labels: [placed_object, placed_case] - num: [5, 10] placed_object_material: AI2-THOR/Materials/Plastics/BlueRubber placed_object_shape: ball

#### placers_with_decoy

(bool): Creates one placer to move and the target object and another that
is a decoy that pretends to move something. The placers have the order
and location they will move to randomized.
Default: False

Simple Example:

placers_with_decoy: False

Advanced Example 1:

'placers_with_decoy': { 'activation_step': [13, 59], 'object_end_position_x': [2.0, -2.0], 'placed_object_position': VectorFloatConfig(x=1, y=0.501, z=2), 'decoy_y_modifier': 0.22 }

Advanced Example 2:

placers_with_decoy: 'placers_with_decoy': { 'activation_step': [13, 59], 'object_end_position_x': [2.0, -2.0], 'labels': 'target_placers', 'move_object_y': 0, 'move_object_z': 0, 'placed_object_labels': 'target', 'placed_object_position': VectorFloatConfig(x=1, y=0.501, z=2), 'placed_object_rotation': 0.0, 'decoy_y_modifier': 0.22 }

#### random_structural_objects

([RandomStructuralObjectConfig](#RandomStructuralObjectConfig), or list of
[RandomStructuralObjectConfig](#RandomStructuralObjectConfig) dict) --
Groups of random object types and how many should be generated from the
type options.
Default: 2 to 4 of all types

random_structural_objects:

• type:
  • doors
  • droppers
  • floor_materials
  • holes
  • l_occluders
  • lava
  • moving_occluders
  • placers
  • platforms
  • ramps
  • throwers
  • tools
  • turntables
  • walls num: min: 2 max: 4

Simple Example:

random_structural_objects: null

Advanced Example:

random_structural_objects:

• type:
  • doors
  • droppers
  • floor_materials
  • holes
  • l_occluders
  • lava
  • moving_occluders
  • placers
  • platforms
  • ramps
  • throwers
  • turntables
  • walls num: min: 0 max: 2
• type: ['walls', 'l_occluders'] num: [3, 5, 7]
• type: 'walls' num: 2

#### restrict_open_doors

(bool): If there are multiple doors in a scene, only allow for one door to
ever be opened.
Default: False

Simple Example:

restrict_open_doors: False

Advanced Example:

restrict_open_doors: True

#### restrict_open_objects

(bool): If there are multiple openable objects in a scene, including
containers and doors, only allow for one to ever be opened.
Default: False

Simple Example:

restrict_open_objects: False

Advanced Example:

restrict_open_objects: True

#### room_dimensions

([VectorIntConfig](#VectorIntConfig) dict, or list of VectorIntConfig
dicts): The total dimensions for the room, or list of dimensions, from
which one is chosen at random for each scene. Rooms are always rectangular
or square. The X and Z must each be within [2, 100] and the Y must be
within [2, 10]. The room's bounds will be [-X/2, X/2] and [-Z/2, Z/2].
Default: random X from 5 to 25, random Y from 3 to 8, random Z from 5 to 25

Simple Example:

room_dimensions: null

Advanced Example:

room_dimensions: x: 10 y: - 3 - 4 - 5 - 6 z: min: 5 max: 10

#### room_shape

(string): Shape of the room to restrict the randomzed room dimensions if
`room_dimensions` weren't configured. Options: `rectangle`, `square`.
Default: Use `room_dimensions`

Simple Example:

room_shape: null

Advanced Example:

room_shape: square

#### shortcut_agent_with_target

(bool, or [AgentTargetConfig](#AgentTargetConfig) dict, or list of bools
and/or AgentTargetConfig dicts): Each scene will have an agent carrying a
target object; you will need to use the InteractWithAgent action on the
agent to request it to produce the target object, so you can then use the
PickupObject action to pick up the target object.

Will use a target already configured via the `goal` option, or, if no
target object is configured, will automatically set a retrieval goal,
generate a soccer ball, and assign it as the retrieval target object.

If `true`, will randomize all of the agent configuration options.

Default: false

Simple Example:

shortcut_agent_with_target: False

Advanced Example:

shortcut_agent_with_target: agent: position: x: min: 1 max: 3 y: 0 z: [2, 3] bounds: - x: 0 z: 2 - x: 2 z: 0 - x: 0 z: -2 - x: -2 z: 0

#### shortcut_bisecting_platform

(bool or [BisectingPlatformConfig](#BisectingPlatformConfig)):
Creates a platform bisecting the room.  If True, the default behavior will
be that the performer starts on one end with a blocking wall in front of
them such that the performer is forced to make a choice on which side they
want to drop off and they cannot get back to the other side. This overrides
the `performer_start_position` and `performer_start_rotation`, if
configured. Note that the blocking wall can be disabled if needed.
Default: False

Simple Example:

shortcut_bisecting_platform: False

Advanced Example:

shortcut_bisecting_platform: has_blocking_wall: False

#### shortcut_double_door_choice

(bool or [DoubleDoorConfig](#DoubleDoorConfig)):
Creates a wall with 2 doors, a platform at the back wall and lava down
the length of the room. (a.k.a. door occluder, or "door-cluder" or
"doorcluder"). The performer starts on the platform on one end such that
the performer is forced to make a choice on which door they want to open.
Default: False

Simple Example:

shortcut_double_door_choice: True

Advanced Example:

shortcut_double_door_choice: add_freeze: False start_drop_step: min: 2 max: 5 occluder_wall_position_z: None occluder_distance_from_performer: 6.5 add_platform: True platform_width: 1 platform_height: 1.5 platform_length: 1 add_lava: True

#### shortcut_imitation_task

(bool or [ImitationTaskConfig](#ImitationTaskConfig)):
Creates a room with an imitation task.
The performer watches an agent open containers in a specific order. Then
the performer is kidnaped. After the kidnaping either the performer
is not moved and the containers are not moved, the containers are moved
and possibly rotated by 90 degrees but the performer does not move, or
the performer is moved but the containers stay the same. In all
cases the agent is moved away from the containers but still nearby to
be seen after the kidnap.
The room is always rectangular.
The short dimension is either 8, 9, 10 and the long is 16, 18, 20.
The room height is always 3.

Simple Example:

shortcut_imitation_task: False

Advanced Example:

shortcut_imitation_task: trigger_order: [left_right, middle_left, right_middle] containers_on_right_side: True kidnap_options: containers_rotate

#### shortcut_lava_room

(bool): Creates a room with lava on either side of the performer. The
performer will start in the center (non-lava) part. Default: False

Simple Example:

shortcut_lava_room: False

Advanced Example:

shortcut_lava_room: True

#### shortcut_lava_target_tool

(bool or [LavaTargetToolConfig](#LavaTargetToolConfig)):
Creates a room with a goal object on an island surrounded by lava.
There will also be a block tool to facilitate acquiring the goal object.
One dimension of the room must be 13 or greater. The other dimension must
be 7 or greater. Rectangular block tools are used on default, and for
these, the max width across for front + island_size + rear is 9. The min
width across for front + island_size + rear is 5. Lava can be asymmetric
but the same restrictions of width min: 5 and max: 9 apply for left +
island_size + right as well. By default, the target is a soccer ball
with scale between 1 and 3.

For hooked and isosceles tools, different min/max rules apply. See
LavaTargetToolConfig for details.

The tool is a pushable/pullable tool object with a length equal
to or greater than the span from the front of the lava over the island to
the back of the lava. It will have a width of either 0.5, 0.75, 1.0.

Default: False

Simple Example:

shortcut_lava_target_tool: False

Advanced Example:

shortcut_lava_target_tool: island_size: min: 1 max: 3 front_lava_width: 2 rear_lava_width: [2, 3] left_lava_width: 2 right_lava_width: min: 2 max: 4 guide_rails: [True, False]

#### shortcut_seeing_leads_to_knowing

(bool): Shortcut for the seeing leads to knowing task. The performer
starts with four open containers in view. An agent will then walk into
view. Placers will descend into each container, with only one having
the soccer ball. After the placers are finished moving, the agent
will attempt to go to the container that has the target object. The
performer will then have to pick whether the scene is plausible or
implausible based on the agent's behavior (note that for training,
all generated scenes are plausible).

Default: False

Simple Example:

shortcut_seeing_leads_to_knowing: False

Advanced Example:

shortcut_seeing_leads_to_knowing: target_behind_agent: True

#### shortcut_start_on_platform

(bool): Ensures that the performer will start on top of a platform. In
order for this to work, `structural_platforms` needs to be specified
using the label `start_structure`. This ignores any configured
`performer_start_position` setting. Default: False

Simple Example:

shortcut_start_on_platform: False

Advanced Example:

shortcut_start_on_platform: True

#### shortcut_tool_choice

(bool, ToolChoiceConfig): Creates a room with a bisecting platform, with
two mirrored lava islands containing targets and a different tool option
(one valid tool that can be used to obtain the target, and one incorrect
option). If True, the default behavior will be that the performer starts
on one end, having to eventually pick a side, and they cannot get back
to the other side. This overrides the `performer_start_position` and
`performer_start_rotation`, if configured. Note that this shortcut
requires a minimum x-axis room size of 20.

Default: False

Simple Example:

shortcut_tool_choice: False

Advanced Example:

shortcut_tool_choice: improbable_choice: 'no_tool'

#### shortcut_triple_door_choice

(bool or [TripleDoorConfig](#TripleDoorConfig)):
Creates a platform bisecting the room with a wall with 3 doors
(a.k.a. door occluder, or "door-cluder" or "doorcluder").
The performer starts on one end such that the performer is forced to make
a choice on which door they want to open.  By default, the doors will be
restricted so only one can be opened.  Will increase the Y room dimension
to 5 if it is lower than 5.  Default: False

Simple Example:

shortcut_triple_door_choice: False

Advanced Example:

shortcut_triple_door_choice: start_drop_step: min: 2 max: 5 add_lips: True add_freeze: [True, False] restrict_open_doors: True

#### side_wall_opposite_colors

(bool, or list of bools): Makes three of the room's walls the same color,
and the other wall (either the left or the right) an opposite color.
Overrides all the `wall_*_material` config options. Default: False

Simple Example:

side_wall_opposite_colors: False

Advanced Example:

side_wall_opposite_colors: True

#### sidesteps

(list of [SidestepsConfig](#SidestepsConfig) dicts): When a sequence of
sidesteps will occur, the label of the object the performer will sidestep
around, and the degrees around the object the performer will sidestep.
Degrees must be in increments of
90 or -90: [90, 180, 270, 360, -90, -180, -270, -360].
If a consecutive list all labels MUST be identical.
The performer must be 3 distance away from the object's center
for sidesteps to work.
*Important Note*: this property will OVERRIDE 'performer_look_at'
and `performer_start_rotation` forcing the performer to look at
this object on scene start.

Simple Example:

sidesteps: null

Advanced Example:

sidesteps:

begin: 50
object_label: turntable
degrees: 90

• begin: 1 object_label: turntable degrees: [90, 180, 270, 360, -90, -180, -270, -360]

#### specific_agents

([AgentConfig](#AgentConfig) dict, or list of
AgentConfig dicts): One or more specific agents (with one
or more possible variations) that will be added to each scene.

Simple Example:

specific_agents: null

Advanced Example:

specific_agents: num: 15 type: [agent_male_02, agent_female_02] agent_settings: chest: 2 eyes: 1 position: x: [1, 0, -1, 0.5, -0.5] y: 0 z: [1, 0, -1] rotation_y: [0, 10, 350] actions: - step_begin: [1, 2] step_end: 7 is_loop_animation: False id: ['TPM_clap', 'TPM_cry'] - step_begin: [13, 14] step_end: 17 is_loop_animation: True id: ['TPM_clap', 'TPM_cry'] movement: animation: TPF_walk step_begin: [2, 4] bounds: - x: 2 z: 0 - x: 0 z: 2 - x: -2 z: 0 - x: 0 z: -2 num_points: 5 repeat: True

#### specific_interactable_objects

([InteractableObjectConfig](#InteractableObjectConfig) dict, or list of
InteractableObjectConfig dicts): One or more specific objects (with one
or more possible variations) that will be added to each scene.

Simple Example:

specific_interactable_objects: null

Advanced Example:

specific_interactable_objects:

num: 1
material: ["WOOD_MATERIALS", "PLASTIC_MATERIALS"]
scale: 1.2
shape: "table_2"
position:
  x: 3
  z: 3

• num: min: 10 max: 30 material:
  • "WOOD_MATERIALS"
  • "PLASTIC_MATERIALS"
  • "AI2-THOR/Materials/Metals/Brass 1" scale: min: 0.8 max: 1.4 shape:
  • "car_1"
  • "racecar_red"
  • "crayon_blue" position: x: min: -4 max: 1 z: min: -4 max: 1

#### structural_droppers

([StructuralDropperConfig](#StructuralDropperConfig) dict, or list of
[StructuralDropperConfig](#StructuralDropperConfig) dicts): Template(s)
containing properties needed to create a droppers.  Default: None

Simple Example:

structural_droppers: num: 0

Advanced Example:

structural_droppers: num: min: 0 max: 3 position_x: min: -1 max: 1 position_z: - 1 - 2 - 5 drop_step: 2 projectile_material: 'AI2-THOR/Materials/Metals/Brass 1' projectile_shape: soccer_ball projectile_scale: 0.85

#### structural_l_occluders

([StructuralLOccluderConfig](#StructuralLOccluderConfig) dict, or list of
[StructuralLOccluderConfig](#StructuralLOccluderConfig) dicts): Template(s)
containing properties needed to create an L-shaped occluder.  Default: None

Simple Example

structural_l_occluders: num: 0

Advanced Example:

structural_l_occluders: num: 2 position: x: 1 y: 0 z: 2 rotation_y: 30 material: 'AI2-THOR/Materials/Metals/Brass 1' scale_front_x: 0.3 scale_front_z: [0.4, 0.5, 0.6] scale_side_x: min: 0.1 max: 2.1 scale_side_z: 0.6 scale_y: 0.7

#### structural_moving_occluders

([StructuralMovingOccluderConfig](#StructuralMovingOccluderConfig) dict,
or list of
[StructuralMovingOccluderConfig](#StructuralMovingOccluderConfig) dicts):
Template(s) containing properties needed to create a structural moving
occluders.  Default: None

Simple Example:

structural_moving_occluders: num: 0

Advanced Example:

structural_moving_occluders: num: min: 0 max: 3 wall_material: 'AI2-THOR/Materials/Metals/Brass 1' pole_material: 'AI2-THOR/Materials/Metals/Brass 1' position_x: - 1 - 2 - 1.5 position_z: min: -3 max: 2.5 origin: top occluder_height: 0.9 occluder_width: 1.1 occluder_thickness: 0.07 repeat_movement: true repeat_interval: 5 rotation_y: 90

#### structural_notched_occluders

([StructuralNotchedOccluderConfig](#StructuralNotchedOccluderConfig), or
list of [StructuralNotchedOccluderConfig](#StructuralNotchedOccluderConfig)
dict) --
Occluders made of 3 cube. Formed with a notch in the middle to leave space
for a platform to pass under.

Simple Example:

structural_notched_occluders: num: 1

Advanced Example

structural_notched_occluders: num: 1 labels: [test_label] material: AI2-THOR/Materials/Metals/BrushedAluminum_Blue height: 2 position_z: 2 down_step: 6 up_step: 61 notch_width: 2 notch_height: 2

#### structural_occluding_walls

([StructuralOccludingWallConfig](#StructuralOccludingWallConfig) dict,
or list of
[StructuralOccludingWallConfig](#StructuralOccludingWallConfig) dicts):
Template(s) containing properties needed to create a occluding walls.
Requires a goal and target.  Default: None

Simple Example:

structural_occluding_walls: num: 0

Advanced Example:

structural_occluding_walls:

• num: min: 0 max: 3 material: 'AI2-THOR/Materials/Walls/DrywallGreen' keyword_location: keyword: between relative_object_label: target
• num: 2 material: 'WALL_MATERIALS' position: x: [1, -5] y: 2 z: 1.5 rotation_y: min: 30 max: 60 scale: x: 3 y: 4

#### structural_platforms

([StructuralPlatformConfig](#StructuralPlatformConfig) dict, or list of
[StructuralPlatformConfig](#StructuralPlatformConfig) dicts): Template(s)
containing properties needed to create a platform.  Default: None

Simple Example:

structural_platforms: num: 0

Advanced Example:

structural_platforms: num: [1, 2, 4] position: x: 1 y: 0 z: 2 rotation_y: 30 material: 'PLASTIC_MATERIALS' lips: front: True back: True left: False right: False scale: x: 1.1 y: [0.5, 1] z: min: 0.3 max: 1.3 auto_adjust_platforms: True long_with_two_ramps: False adjacent_to_wall: ['front_left_corner, 'right']

#### structural_ramps

([StructuralRampConfig](#StructuralRampConfig) dict, or list of
[StructuralRampConfig](#StructuralRampConfig) dicts): Template(s)
containing properties needed to create a ramp.  Default: None

Simple Example:

structural_ramps: num: 0

Advanced Example:

structural_ramps: num: min: 0 max: 3 position: x: 1 y: 0 z: 2 rotation_y: 30 material: 'AI2-THOR/Materials/Metals/Brass 1' angle: 30 width: 0.4 length: 0.5

#### structural_throwers

([StructuralThrowerConfig](#StructuralThrowerConfig) dict, or list of
[StructuralThrowerConfig](#StructuralThrowerConfig) dicts): Template(s)
containing properties needed to create a droppers.  Default: None

Simple Example:

structural_throwers: num: 0

Advanced Example:

structural_throwers: num: min: 0 max: 3 wall: [front, left] position_wall: min: -1 max: 3 height: - 1 - 2 - 1.5 rotation_y: 0 rotation_z: min: 3 max: 7 throw_force: [600, 1000, 1200] throw_step: 2 projectile_material: 'AI2-THOR/Materials/Metals/Brass 1' projectile_shape: soccer_ball projectile_scale: 0.85

#### structural_tube_occluders

([StructuralTubeOccluderConfig](#StructuralTubeOccluderConfig), or list
of [StructuralTubeOccluderConfig](#StructuralTubeOccluderConfig) dict) --
One or more templates containing properties needed to generate tube
occluders. Default: None


Simple Example:

structural_tube_occluders: - num: 0

Advanced Example:

structural_tube_occluders: - num: min: 1 max: 3 - num: 1 position: x: [1, 2] y: 0 z: min: -3 max: 3 radius: 4 material: WOOD_MATERIALS down_step: 21 up_step: 81

#### structural_turntables

([StructuralTurntableConfig](#StructuralTurntableConfig), or list
of [StructuralTurntableConfig](#StructuralTurntableConfig) dict) --
Groups of structural turntable configurations and how many should be
generated from the given options.
Default: 0


Simple Example:

structural_turntables: - num: 0

Advanced Example:

structural_turntables: - num: min: 1 max: 3 - num: 1 position: x: [1,2] y: 0 z: min: -3 max: 3 turntable_height: 0.5 turntable_radius: 1 turntable_movement: step_begin: 1 step_end: 10 rotation_y: 20 material: PLASTIC_MATERIALS - num: [1, 3] position: x: [4, 5] y: 0 z: min: -5 max: -4 turntable_height: min: .5 max: 2 turntable_radius: [.75, 1.25] turntable_movement: step_begin: 1 step_end: 20 rotation_y: 10 material: AI2-THOR/Materials/Metals/BrushedAluminum_Blue

#### structural_walls

([StructuralWallConfig](#StructuralWallConfig) dict, or list of
[StructuralWallConfig](#StructuralWallConfig) dicts): Template(s)
containing properties needed to create an interior wall.  Default: None

Simple Example:

structural_walls: num: 0

Advanced Example:

structural_walls: num: min: 1 max: 3 position: x: 1 y: 0 z: 2 rotation_y: 30 material: 'PLASTIC_MATERIALS' width: .5 height: .3

#### swivels

(list of [StepBeginEnd](#StepBeginEnd) dicts): When a swivel
action should occur.  A swivel forces the performer agent to only
`"['LookDown', 'LookUp', 'RotateLeft', 'RotateRight']"` for a
range of steps.  User should try to avoid swivel (and freeze) overlaps
but if using ranges and choices, the ILE will retry on overlaps.
This field must be blank or an empty array if `passive_scene` is `true`.

Simple Example:

swivels: null

Advanced Example:

swivels:

begin: 7
end: 9

• begin: [16, 18, 20] end: min: 26 max: 30

#### teleports

(list of [TeleportConfig](#TeleportConfig) dicts): When a
kidnap/teleport will occur and where the player agent should be teleported.
This field must contain either both position fields or the `rotation_y`
field or an exception will be thrown.  This field must be blank or an empty
array if `passive_scene` is `true`.

Simple Example:

teleports: null

Advanced Example:

teleports:

step: 5
position_x: 3
position_z: 6
rotation_y: 45

• step: [34, 36] position_x: min: -3 max: 3 position_z: [3, 5, 7]
• step: min: 41 max: 48 rotation_y: [30, 120, 270]

#### tools

([ToolConfig](#ToolConfig), or list of [ToolConfig](#ToolConfig) dict) --
Groups of large block tool configurations and how many should be generated
from the given options.
Default: 0


Simple Example:

tools: - num: 0

Advanced Example:

tools: - num: min: 1 max: 3 - num: 1 shape: tool_rect_1_00_x_9_00 position: x: [1,2] y: 0 z: min: -3 max: 3 - num: [1, 3] shape: - tool_rect_0_50_x_4_00 - tool_rect_0_75_x_4_00 - tool_rect_1_00_x_4_00 position: x: [4, 5] y: 0 z: min: -5 max: -4

#### trapezoidal_room

(bool, or list of bools): Makes the room trapezoidal, so the left and right
walls will be angled inward. Currently only supported for room_dimensions
of X=12 and Z=16. Default: False

Simple Example:

trapezoidal_room: False

Advanced Example:

trapezoidal_room: True

#### turntables_with_agent_and_non_agent

([TurntablesAgentNonAgentConfig](#TurntablesAgentNonAgentConfig) dict):
Useful for the Spatial Reference task. Rotates a turntable underneath
a non-agent object so it faces another object in the scene.

Simple Example:

turntables_with_agent_and_non_agent: null

Advanced Example:

turntables_with_agent_and_non_agent: agent_label: my_agent non_agent_label: my_rotating_object turntable_labels: [my_turntable_1, my_turntable_2] direction_labels: [my_static_object_1, my_static_object_2]

#### wall_back_material

(string, or list of strings): The material for the back wall, or list of
materials, from which one is chosen for each scene. Default: random

Simple Example:

wall_back_material: null

Advanced Example:

wall_back_material: "Custom/Materials/GreyDrywallMCS"

#### wall_front_material

(string, or list of strings): The material for the front wall, or list of
materials, from which one is chosen for each scene. Default: random

Simple Example:

wall_front_material: null

Advanced Example:

wall_front_material: "Custom/Materials/GreyDrywallMCS"

#### wall_left_material

(string, or list of strings): The material for the left wall, or list of
materials, from which one is chosen for each scene. Default: random

Simple Example:

wall_left_material: null

Advanced Example:

wall_left_material: "Custom/Materials/GreyDrywallMCS"

#### wall_material

(string, or list of strings): The material for all room walls, or list of
materials, from which one is chosen for each scene. Is overridden by the
individual `wall_*_material` config options. Default: ROOM_WALL_MATERIALS

Simple Example:

wall_material: null

Advanced Example:

wall_material: "Custom/Materials/GreyDrywallMCS"

#### wall_right_material

(string, or list of strings): The material for the right wall, or list of
materials, from which one is chosen for each scene. Default: random

Simple Example:

wall_right_material: null

Advanced Example:

wall_right_material: "Custom/Materials/GreyDrywallMCS"