GraSIF Repository

This folder contains three graph datasets saved as pickle files, used for the evaluation of methods. Each dataset is a list of dictionaries containing the task name, initial state, and goal state represented as NetworkX graphs. Additionally, each dictionary includes specific information relevant to the dataset.

Data	Number of Tasks	Mean nodes	Actions
SayPlan Office	25	202.6	2.1
Behaviour-1K	186	12.1	4.9
VirtualHome	347	195.7	1.6

Table 1: Dataset comparison. Actions represent the mean number of nodes changed between the initial and goal graph.

To load a dataset, use the following code snippet:

import pickle

with open('./datasets/<name>.pkl', 'rb') as file:
    tasks = pickle.load(file)

SayPlan Office

The SayPlan Office dataset represents graphs and tasks defined in SayPlan. Each task consists of a dictionary with the following structure:

• name: The name of the task.
• human: Human-readable task description (same as name for SayPlan).
• detailed: Detailed task description (same as name for SayPlan).
• init: Initial state as a NetworkX graph.
• goal: Goal state as a NetworkX graph.
• actions: A list of ground-truth actions to complete the task.

Behaviour-1K

The Behaviour-1K dataset represents tasks defined in Behaviour1K. For each task defined in BDDL, a subgraph was constructed to represent the environment. Using this subgraph, the goal graph was created, and human-readable as well as detailed task descriptions were added.

The dataset contains a total of 186 tasks, each represented by a dictionary with the following structure:

• name: The name of the task from Behaviour1K.
• human: Human-readable task description.
• detailed: Detailed task description.
• init: Initial state as a NetworkX graph.
• goal: Goal state as a NetworkX graph.

VirtualHome RobotHow

The VirtualHome dataset represents tasks from the RobotHow dataset. For each task, the VirtualHome graph was reconstructed into a structure compatible with our methods. This was achieved using the graph_parser.py script available in the utils folder of the repository.

Additionally, an ids dictionary maps nodes from the initial NetworkX graph to VirtualHome IDs. For example, the node ('fridge', 1) in the initial graph corresponds to the fridge node with ID 67 in the VirtualHome backend graph. This mapping is useful when using the dataset with the VirtualHome simulator.

Each task is represented by a dictionary with the following structure:

• name: The name of the task from RobotHow.
• human: Human-readable task description. (Same as name for RobotHow)
• detailed: Detailed task description.
• init: Initial state as a NetworkX graph.
• goal: Goal state as a NetworkX graph.
• ids: Mapping of nodes to VirtualHome IDs.