Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HyperTree Planning: Enhancing LLM Reasoning via Hierarchical Thinking

Runquan Gui, Zhihai Wang, Jie Wang, Chi Ma, Huiling Zhen, Mingxuan Yuan, Jianye Hao, Defu Lian, Enhong Chen, Feng Wu

TL;DR

HyperTree Planning (HTP) introduces hypertree-structured planning outlines to enable hierarchical thinking in LLMs for complex planning tasks. It combines a top-down hypertree construction with self-guided planning and plan generation to address long reasoning chains and diverse constraints. Across TravelPlanner, PlanBench, and Natural Plan benchmarks, HTP delivers state-of-the-art accuracy and notable efficiency gains compared with strong baselines, including multi-path reasoning and agent-based approaches, particularly when paired with Gemini-1.5-Pro or GPT-4o backbones. The work demonstrates the practicality and impact of structured, hierarchical planning for autonomous agents and long-horizon problem solving.

Abstract

Recent advancements have significantly enhanced the performance of large language models (LLMs) in tackling complex reasoning tasks, achieving notable success in domains like mathematical and logical reasoning. However, these methods encounter challenges with complex planning tasks, primarily due to extended reasoning steps, diverse constraints, and the challenge of handling multiple distinct sub-tasks. To address these challenges, we propose HyperTree Planning (HTP), a novel reasoning paradigm that constructs hypertree-structured planning outlines for effective planning. The hypertree structure enables LLMs to engage in hierarchical thinking by flexibly employing the divide-and-conquer strategy, effectively breaking down intricate reasoning steps, accommodating diverse constraints, and managing multiple distinct sub-tasks in a well-organized manner. We further introduce an autonomous planning framework that completes the planning process by iteratively refining and expanding the hypertree-structured planning outlines. Experiments demonstrate the effectiveness of HTP, achieving state-of-the-art accuracy on the TravelPlanner benchmark with Gemini-1.5-Pro, resulting in a 3.6 times performance improvement over o1-preview.

HyperTree Planning: Enhancing LLM Reasoning via Hierarchical Thinking

TL;DR

HyperTree Planning (HTP) introduces hypertree-structured planning outlines to enable hierarchical thinking in LLMs for complex planning tasks. It combines a top-down hypertree construction with self-guided planning and plan generation to address long reasoning chains and diverse constraints. Across TravelPlanner, PlanBench, and Natural Plan benchmarks, HTP delivers state-of-the-art accuracy and notable efficiency gains compared with strong baselines, including multi-path reasoning and agent-based approaches, particularly when paired with Gemini-1.5-Pro or GPT-4o backbones. The work demonstrates the practicality and impact of structured, hierarchical planning for autonomous agents and long-horizon problem solving.

Abstract

Recent advancements have significantly enhanced the performance of large language models (LLMs) in tackling complex reasoning tasks, achieving notable success in domains like mathematical and logical reasoning. However, these methods encounter challenges with complex planning tasks, primarily due to extended reasoning steps, diverse constraints, and the challenge of handling multiple distinct sub-tasks. To address these challenges, we propose HyperTree Planning (HTP), a novel reasoning paradigm that constructs hypertree-structured planning outlines for effective planning. The hypertree structure enables LLMs to engage in hierarchical thinking by flexibly employing the divide-and-conquer strategy, effectively breaking down intricate reasoning steps, accommodating diverse constraints, and managing multiple distinct sub-tasks in a well-organized manner. We further introduce an autonomous planning framework that completes the planning process by iteratively refining and expanding the hypertree-structured planning outlines. Experiments demonstrate the effectiveness of HTP, achieving state-of-the-art accuracy on the TravelPlanner benchmark with Gemini-1.5-Pro, resulting in a 3.6 times performance improvement over o1-preview.
Paper Structure (36 sections, 2 equations, 5 figures, 7 tables, 1 algorithm)

This paper contains 36 sections, 2 equations, 5 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Reasoning Paradigms
  4. Planning Agents
  5. Preliminary
  6. Problem Statement
  7. HyperTree Definition
  8. Methodology
  9. From Tree Planning to HyperTree Planning
  10. Top-down Hypertree Construction Algorithm
  11. Self-guided Planning and Plan Generation
  12. Experiments
  13. Setups
  14. Main Results
  15. Comparison with powerful SOTA LLMs
  16. ...and 21 more sections

Figures (5)

  • Figure 1: An example of hierarchical thinking, demonstrating its ability to decompose complex reasoning chains into manageable components, effectively handle diverse constraints, and systematically manage multiple distinct sub-tasks.
  • Figure 2: An overview of HyperTree Planning (HTP). Compared to previous tree planning methods such as ToT yao2024tree and RAP hao2023reasoning, HTP introduces structural innovations that enable each edge to connect multiple child nodes, making it suitable for a divide-and-conquer strategy.
  • Figure 3: Flowchart of HTP, which consists of three parts: (1) HyperTree Construction; (2) Self-Guided Planning; (3) Plan Generation.
  • Figure 4: Success rates on the TravelPlanner benchmark categorized by problem instance difficulty and trip durations. Circles indicate instances with a success rate of 0 for clearer identification.
  • Figure 5: Success rates on the Blocksworld and Trip Planning benchmarks, categorized by different pruning strategies.