Read (x, y, z, yaw_deg) for an embodied agent from THOR metadata.

Read (x, y, z, yaw_deg) for an embodied agent from THOR metadata.

Apply environment-specific defaults after YAML merge.

golden_actions ， max_steps = 10 + 2 * n n actions Done ， steps task_config

/tmp/.X11-unix/X* X display `xdpyinfo` ； xdpyinfo socket None

Convert action parameters to a VirtualHome script string. Returns: (script_line, error_message). On success, error_message is Non

spatial-planning/envs/ai2thor/wrapper.py Returns:

Build success condition predicate from YAML configured success_condition Returns: Success condition predicate function

Approximate active first-person camera world position from char pose.

Retry PutBack/PutIn only if strict preconditions still hold.

Check if the task success condition is met using the metadata dict. Mirrors AI2ThorEnvWrapper._check_done logic; works on the AI2THOR-compati

Convert position payload to [x, y, z] list.

Collect (interactable_types, all_scene_types) from a graph snapshot.

Return a step reward (small bonus/penalty); task completion reward is added by caller.

（ ） spatial-planning/envs/ai2thor/wrapper.py

Calculate reward (generalized version, supports different tasks) **Uses reward parameters from YAML config**: - step_success_bonus: S

spatial-planning/envs/ai2thor/wrapper.py

Calculate reward based on metadata and action result

task_config success_conditions env success_predicate wrapper reset/step ， target_object_types target_description

EnvAction AI2-THOR spatial-planning/envs/ai2thor/wrapper.py

Convert EnvAction to AI2-THOR action sequence **Dual-mode support (optimized)**: 1. Priority mode: If action.thor_actions is non-empt

move/turn/interact spatial-planning/envs/ai2thor/wrapper.py

Traditional abstract action conversion logic (backward compatible) Uses "prioritize target object" strategy: 1. If target object type

（ ） spatial-planning/envs/ai2thor/wrapper.py For PickupObject: SEMANTIC_OBJECT_MAPPING

Convert interaction action For PickupObject: Supports semantic searching based on SEMANTIC_OBJECT_MAPPING. When model outputs PickupO

（ ） spatial-planning/envs/ai2thor/wrapper.py Args: action_name: magnitude: （

Convert navigation action Supported navigation actions: - Translation: MoveAhead, MoveBack, MoveLeft, MoveRight (use magnitude or def

，

Deep copy dictionary

Deep copy dictionary

Deep copy dictionary

Derive success/failure/null task_result for logs.

spatial-planning/envs/ai2thor/wrapper.py

Determine interaction action based on object properties and task semantics Args: obj: Object metadata Returns:

[ ] 3D ( )

[ ] 3D ，

[ ]

[ ]

[ ] ( + )

( ， )

[ ] ，

[ ] ( )

Agent UI

Avoid hard crashes on non-UTF8 terminals (e.g. Windows cmd GBK).

Register/update dedicated first-person cameras and return their IDs. Returns interact-compatible camera IDs: the last N indices from

pip virtualhome sys.path（ __init__.py ）

Evaluate a single condition Args: condition: Single condition configuration env: Environment instance met

Evaluate a single condition Args: condition: Single condition configuration env: Environment instance met

Evaluate a single condition Args: condition: Single condition configuration env: Environment instance met

Evaluate a single condition Args: condition: Single condition configuration env: Environment instance met

graph think_node fail_reason， （ ACTION） : Step 5 failed after 3 retries: Missing <ACTION> tag or empty ACTION conte

Find an object bbox containing the character body/camera probe.

Find evaluation result json for a specific environment.

spatial-planning/envs/ai2thor/wrapper.py

Automatically find suitable interaction target object Priority: 1. Visible target type objects (target_object_types) 2. Visib

Find task.json file for a given task name Searches in tasks/{task_name}/task.json Supports task ID format conversion (with/wi

Find task.json file for a given task name Searches in tasks/{task_name}/task.json Supports task ID format conversion (with/wi

Find task.json from standard dirs; support procthor000 / procthor_000. Search order: 1. <project_root>/dual_agent/task_mutil_procth

Find task.json file for a given task name Searches in tasks/{task_name}/task.json Supports task ID format conversion (with/wi

spatial-planning/envs/ai2thor/wrapper.py

Format action for display Args: action: Environment action Returns: Formatted action string

Format a compact human-readable action string for episode logs.

，

spatial-planning/envs/ai2thor/wrapper.py

Strict first-person text: only provides action feedback and task hints

spatial-planning/envs/ai2thor/wrapper.py

Original omniscient view text state

entry_pos: （ S， T） exit_stair_pos: （ None）

spatial-planning/envs/ai2thor/wrapper.py

Generate status summary of task-related objects Provides key information about task goals to help Agent understand current task progress

Read small/medium/large movement granularity from action metadata.

，

Return visible object ids from Unity get_visible_objects(camera_id).

Return a previously bound instance id for the object type, if still valid.

Pygame

Returns: Dict[str, bool]: ， ： - "real_time": - "supports_actio

（ reset） ： agent0 agent1 ``reset_*.png`` ``agent1/`` ``agent2/`` ， agent0 observation

Map seg_inst RGB colors to graph object ids.

Backward-compatible alias for surface distance.