Args: callback: ， Action

Args: callback: ， FrameData

Args: callback: ， GameState

Args: screenshot: base64 prompt_text: （ eval ） step_number: （ 1 ）

base64 Args: image_input: PIL.Image Returns: base64 Raises: ValueError:

OpenAI ， API Args: response: OpenAI API api_type: API （"responses" "chat"）， None Returns:

Get list of all available task names Combines tasks from both task_presets and task.json files Returns:

Returns: Dict[str, Any]:

Args: **kwargs: Returns: bool:

graph.py parse_action_string() ： - MoveAhead,RotateLeft,MoveAhead - PickupObject(Apple),PutObjec

- OpenGL 2D

Return a graph node by id from the current snapshot.

Normalize an angle to [-pi, pi].

Save image frame (same naming as single-agent: ``{prefix}_{timestamp}.png``). When ``agent_count > 1``, saves under ``agent{k}/`` for the emb

scripts/carla/interact_carla.py teleport_forward ： hop（ 2m） ， ** **， “ 8m ”

Align multi-agent with single-agent: always use ``Event`` and ``event.frame``. ``controller.reset`` / ``step`` return ``Event`` when ``agent_

Create evaluator from configuration Supports both: 1. Single success_condition (legacy format) 2. Multiple success_conditions with succes

Evaluate task completion Args: env: Environment instance metadata: Environment metadata Returns:

JSON（ log.json， episode_*.json）

Find result JSON (log.json or episode_*.json) in task output dir. Returns: Path to JSON or None.

Find result JSON (log.json or episode_*.json) in task output dir. Returns: Path to JSON or None.

Find result JSON (log.json or episode_*.json) in task output dir. Returns: Path to JSON or None.

Render the current world state from a specific embodied agent's camera. Uses a no-op Pass so AI2-THOR returns that agent's frame and metadata

Deactivate current agent and hand off to the partner if possible.

init.json task.json init.json ， Args: task_file: ， tasks/procthor109/task.js

Mark current agent as failed and maybe update global failure hints.

task id， ai2thor_04000 / ai2thor04000

Parse a VirtualHome-style action string into an action dictionary. Supported formats: TurnLeft -> body turn 90° (defau

Print configuration information Args: config: ConfigLoader instance section: Section to print, None means print all

JSON

Set CARLA weather; override_name wins, else fallback_name.

ProcTHOR house （ ） ProcTHOR house rooms ， room floorPolygon ， ， Args: env

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

Check if task is complete (generalized version, based on task config) Determines task completion through target_object_types and success_pred

Args: metadata: AI2-THOR Returns: str:

Generate text state description (supports first_person / omniscient modes)

Return current character [x, y, z] from the cached graph snapshot.

Check if the object is currently held by the first character.

Extract world-space bounding-box center and size from a node.

Match interact mode by compensating VH walk-induced yaw drift.

Multi-agent: ``MultiAgentEvent`` per-agent ``Event``（ ``.frame``） spatial-planning/envs/ai2thor/wrapper.py — ``agent_count==1``

Prefer UTF-8 for terminal I/O without failing on redirected streams.

Completed：true= evaluate；false= （ ACTION）；null=

Decide CSV Completed status, considering both failure_type and fail_reason. Rules: - api_error/env_error/external_error → null (external issu

Decide CSV Completed status, considering both failure_type and fail_reason. Rules: - api_error/env_error/external_error → null (external issu

Determine failure reason; prefer failure_type from JSON. Returns one of: api_error, parse_error, env_error, action_error, model_error, exter

Determine failure reason; prefer failure_type from JSON. Supports both log.json (metadata.failure_type) and episode_*.json (root failure_typ

Determine failure reason; prefer failure_type from JSON. Returns one of: api_error, parse_error, env_error, action_error, model_error, exter