Towards Zero-Knowledge Task Planning via a Language-based Approach

TL;DR

This paper introduces a novel zero-knowledge task planning framework using large language models to decompose instructions, generate behavior trees, and refine plans dynamically, demonstrating effectiveness in simulated environments.

Contribution

It formalizes the ZKTP problem and presents a pioneering LLM-based approach for task decomposition, plan generation, and on-the-fly refinement without task-specific knowledge.

Findings

Improved task performance in AI2-THOR simulator.

Effective decomposition of natural language instructions.

Dynamic plan refinement enhances robustness.

Abstract

In this work, we introduce and formalize the Zero-Knowledge Task Planning (ZKTP) problem, i.e., formulating a sequence of actions to achieve some goal without task-specific knowledge. Additionally, we present a first investigation and approach for ZKTP that leverages a large language model (LLM) to decompose natural language instructions into subtasks and generate behavior trees (BTs) for execution. If errors arise during task execution, the approach also uses an LLM to adjust the BTs on-the-fly in a refinement loop. Experimental validation in the AI2-THOR simulator demonstrate our approach's effectiveness in improving overall task performance compared to alternative approaches that leverage task-specific knowledge. Our work demonstrates the potential of LLMs to effectively address several aspects of the ZKTP problem, providing a robust framework for automated behavior generation with no task-specific setup.