Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Bootstrapping Human-Like Planning via LLMs

Created by
  • Haebom

Author

David Porfirio, Vincent Hsiao, Morgan Fine-Morris, Leslie Smith, Laura M. Hiatt

Outline

This paper studies an accessible method for robot task assignment. We investigate how to combine two common end-user programming paradigms: drag-and-drop interfaces and natural language programming. Specifically, we construct a large-scale language model (LLM)-based pipeline that takes natural language as input and outputs human-like task sequences at a human-like granularity, and compare the generated task sequences to a dataset of manually assigned task sequences. Experimental results show that larger models perform better in generating human-like task sequences, but smaller models also achieve satisfactory performance.

Takeaways, Limitations

Takeaways:
Possibility of increasing efficiency in robot task assignment using natural language
Demonstrating the potential for effective combination of drag-and-drop interfaces and natural language programming through LLM-based pipelines
Provides criteria for selecting appropriate models by verifying the correlation between model size and performance.
Limitations:
The scope of the study may be limited to specific types of tasks and robots.
Lack of sufficient consideration of errors and limitations that may occur when applied to actual robot systems.
Lack of testing across diverse user groups limits generalizability.
👍