H-AIM: Orchestrating LLMs, PDDL, and Behavior Trees for Hierarchical Multi-Robot Planning

TL;DR

H-AIM addresses long-horizon planning for heterogeneous robot teams by translating high-level instructions into executable parallel behavior trees through a cascaded pipeline. It combines an LLM-driven PDDL File Generator, a Hybrid Planner that merges semantic reasoning with classical planning, and a Behavior Tree Compiler that yields fault-tolerant execution, all coordinated via a shared blackboard. On the MACE-THOR benchmark, H-AIM substantially improves task success rate and goal recall over the strongest baseline, demonstrating robust, dynamic multi-robot coordination in household scenarios ($SR$ rising from $12\%$ to $55\%$ and $GCR$ from $32\%$ to $72\%$). This work enables scalable, end-to-end autonomous planning-execution loops for diverse robot teams and lays groundwork for real-world deployment with further extensions to perception and re-planning under partial observability.

Abstract

In embodied artificial intelligence, enabling heterogeneous robot teams to execute long-horizon tasks from high-level instructions remains a critical challenge. While large language models (LLMs) show promise in instruction parsing and preliminary planning, they exhibit limitations in long-term reasoning and dynamic multi-robot coordination. We propose Hierarchical Autonomous Intelligent Multi-Robot Planning(H-AIM), a novel embodied multi-robot task planning framework that addresses these issues through a three-stage cascaded architecture: 1) It leverages an LLM to parse instructions and generate Planning Domain Definition Language (PDDL) problem descriptions, thereby transforming commands into formal planning problems; 2) It combines the semantic reasoning of LLMs with the search capabilities of a classical planner to produce optimized action sequences; 3) It compiles the resulting plan into behavior trees for reactive control. The framework supports dynamically sized heterogeneous robot teams via a shared blackboard mechanism for communication and state synchronization. To validate our approach, we introduce the MACE-THOR benchmark dataset, comprising 42 complex tasks across 8 distinct household layouts. Experimental results demonstrate that H-AIM achieves a remarkable performance improvement, elevating the task success rate from 12% to 55% and boosting the goal condition recall from 32% to 72% against the strongest baseline, LaMMA-P.

Figures (6)

• Figure 1: Overview of H-AIM. The diagram illustrates the core composition of our approach: the technical foundation (LLM, PDDL, BT), the executing heterogeneous robot team, and representative application scenarios.
• Figure 2: H-AIM architecture. The framework orchestrates three LLM-driven modules (PFG, HP, BTC) to convert language instructions into executable plans.
• Figure 3: PFG. This module transforms natural language instructions into structured PDDL problem files by parsing the input, decomposing the task, allocating subtasks, and formalizing planning elements.
• Figure 4: HP. This module orchestrates classical and LLM-driven planning stages to generate optimized, robust action sequences.
• Figure 5: Multi-robot Parallel Behavior Tree. The top-level Parallel node coordinates individual robot subtrees, with a shared blackboard enabling communication and state synchronization.