Researchers propose AdaPlan-H, a self-adaptive hierarchical planning mechanism for LLM agents that adjusts planning granularity based on task complexity.

AdaPlan-H initiates with a coarse-grained macro plan and progressively refines it based on task complexity, generating self-adaptive hierarchical plans tailored to varying difficulty levels of different tasks.

The method can be optimized by imitation learning and capability enhancement, mimicking human planning strategies inspired by the principle of progressive refinement in cognitive science.

Experimental results demonstrate the method significantly improves task execution success rates while mitigating overplanning at the planning level, providing a solution for multi-step complex decision-making tasks.

Code and data will be made publicly available; the submission was posted on 25 Apr 2026.