metadata

license: apache-2.0
pretty_name: Embodied Spatial Reasoning
dataset_info:
  - config_name: 1.0.1
    features:
      - name: task_id
        dtype: string
      - name: query
        dtype: string
      - name: label
        dtype: bool
      - name: constraint_name
        dtype: string
      - name: image_url
        dtype: string
      - name: objects
        sequence: string
      - name: related_task_id
        dtype: string
      - name: scene_id
        dtype: string
    splits:
      - name: question_answering
        num_examples: 6000
    dataset_size: 6000
configs:
  - config_name: 1.0.1
    data_files:
      - split: question_answering
        path: transformed_all_v2.json
train-eval-index:
  - config: 1.0.1
    splits:
      eval_split: question_answering
    col_mapping:
      task_query: text
      task_label: label
language:
  - en
task_categories:
  - visual-question-answering
tags:
  - spatial reasoning
  - embodied AI
size_categories:
  - 1K<n<10K

Embodied Spatial Reasoning Tasks

Dataset Description

This dataset is part of the embodied-spatial-reasoning project, where the agent has to actively explore the environment to determine if certain spatial relationships hold true. The tasks involve spatial reasoning with various objects and scenes. Each task includes a query about the spatial relationships between objects within a scene, which the agent must verify through exploration.

Dataset Structure

The dataset is organized so that each entry represents a single task associated with a specific scene.

Features

task_id: (string) Unique identifier for each task.
query: (string) A query describing the spatial relationship to be verified (e.g., "The object sofa is to the left of wardrobe.").
label: (bool) Indicates whether the spatial relationship described in the query is correct (True) or not (False).
constraint_name: (string) Name of the spatial relationship (e.g., "left").
image_url: (string) URL of the image associated with the task.
objects: (list) List of objects involved in the scene.
related_task_id: (string) Identifier of a related task. Each positive task has a corresponding negative task.
scene_id: (string) Identifier for the scene where the task occurs.

Example Entry

{
    "query": "The object sofa is to the left of wardrobe.",
    "label": false,
    "constraint_name": "left",
    "image_url": "https://embodied-spatial-reasoning.storage.googleapis.com/images/left/False/718c876a-a8d1-4297-b552-24ce1f136ed4.png",
    "task_id": "718c876a-a8d1-4297-b552-24ce1f136ed4",
    "objects": [
        "wardrobe",
        "sofa"
    ],
    "related_task_id": "579c553b-21dd-4a44-bcef-1da9e5fada9f",
    "scene_id": "83ce1d71-506e-43aa-8fda-13911afa6e44"
}