Large-Scale Task Planning

This project explores combining environmental knowledge with the commonsense knowledge encoded in large language models (LLMs) to reduce the search space in task planning problems. This approach is particularly useful in large-scale settings, as it enables the selection of a relevant set of objects over which the planning problem is addressed.

1. Relevant-Object Selection via LLMs, Graphs, and Object Hierarchies

In this project, we endow LLMs with a structured world representation to address the scalability challenges present in task planning problems. By reducing the problem size before planning begins, we focus on the key idea that, despite the presence of thousands of objects in complex environments, only a few are essential for completing specific tasks.

Paper (preprint):

@misc{perez2024leveraging,
  title={Leveraging LLMs, Graphs and Object Hierarchies for Task Planning in Large-Scale Environments},
  author={P{\'e}rez-Dattari, Rodrigo and Li, Zhaoting and Babu{\v{s}}ka, Robert and Kober, Jens and Della Santina, Cosimo},
  howpublished={arXiv preprint arXiv:2409.04775},
  year={2024}
}

      

Links:

Project members:
R. Pérez-Dattari, Z. Li, R. Babuska, J. Kober and C. Della Santina
ongoing