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

2025-10-19

IEEE/RJS International Conference on Intelligent RObots and Systems

AI Summary

This study addresses the challenge of automatically generating action sequences to achieve goals without any task-specific prior knowledge. To this end, the authors propose a zero-shot task planning framework grounded in large language models (LLMs), which parses natural language instructions into subtasks and constructs behavior trees for execution. In the event of execution failure, the framework leverages the LLM to iteratively refine the behavior tree online, establishing a closed-loop optimization process. This work formally defines the zero-knowledge task planning problem for the first time and demonstrates significant performance gains over baseline methods that rely on task-specific knowledge, as evaluated in the AI2-THOR simulation environment. The results substantiate the framework’s robustness and generalization capabilities across diverse tasks.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Planning, Routing, and Scheduling: Planning with Language Models
• Multiagent Systems: Multiagent Planning

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

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.

Problem

• Zero-Knowledge Task Planning
• Task Planning
• Large Language Models
• Behavior Trees
• Natural Language Instructions

Innovation

• Zero-Knowledge Task Planning
• Large Language Models
• Behavior Trees
• Task Decomposition
• Online Refinement