LLM Robot Planning Dataset

This repo offers the dataset of the this paper and the possibility to create new datasets.

Dataset

The dataset can be found here

The Dataset Structure

The dataset is structured as follows:

1. Each sentence has an associated JSON file containing the ground truth movements.
2. The sentence.json file contains all sentences along with additional information about the sentences.

Create new Dataset

To create a new dataset several files are important:

1. objects.json: This JSON file describes the rules, for each rule the corresponding objects and the location.
2. gt_paraml.yml : This file contains all config parameters that are needed to create the dataset. Hint: The rule names in this file should be the same as in the objects.json file.

After the setup of these files run the script create_gt.py to start the dataset creation. Hint: In this file you can choose the directory name and the path.

python3 src/create_gt.py 

Read the Dataset

The script gt_reader.py is used to read the dataset. Adjust the dataset path in the script.

python3 src/gt_reader.py