SaGC dataset for CLARA paper

CLARA: Classifying and Disambiguating User Commands for Reliable Interactive Robotic Agents

Project & Arxiv

The main code can be found on Code

Dataset components

Dataset

The dataset is in data/agument.json The dataset is consisted of

{
    "INDEX_NUMBER":
        {
            "SCENE":{
                '''scene information'''
                floorplan list: []
                object lists: []
                people list: []
            }
            "GOAL": Language Instruction
            "LABEL": Uncertainty Label
            "TASK": Robot Type
        }
}

Scene

Description of the environment

• floorplan: list of area categories in the environment
• objects: list of objects seen in the environment
• people: list of peoeple in the scene. Each person is decripted as the color of the cloth wearing

Goal

Language instruction (command) given to the robot

Robot type

• cooking robot
• cleaning robot
• massage robot

Uncertainty label

• 0: clear
• 1: ambiguious
• 2: infeasible
• 3: ignore

Examples

for example,

"2": {
        "scene": {
            "floorplan": [
                "kitchen",
                "living room",
                "bedroom"
            ],
            "objects": [
                "water",
                "bacon",
                "bread",
                "pan",
                "coffee",
                "table",
                "orange juice",
                "sasuage",
                "banana",
                "apple"
            ],
            "people": [
                "person wearing blue shirt",
                "person wearing white shirt",
                "person wearing red shirt"
            ]
        },
        "goal": "Cook and serve bacon and toast on a plate.",
        "label": 0,
        "task": "cooking"
    }

Dataset Generation

1. Set your key in key/[your key file.txt]

2. Craft examples. The samples used in the dataset is in data/sample.json file

3. Run agument.ipynb file