Exploring Metamodels and LLM Agents for Planning Tasks in Modular Material Handling Systems
Year: 2026
Type: conference-paper
Abstract: LLMs are increasingly being used as decision making agents, but applying them to intralogistics remains challenging because warehouse operations require reliable state awareness and feasibility checking. This paper studies the integration of a ReAct agent into a modular material handling (MMH) system, where the agent performs high-level task planning from natural language requests and executes plans through tool calls operating over a meta-model abstraction of the system. To improve the reliability of tool selection, we model the MMH system using a knowledge graph that represents components, relations, states, and operational constraints of the system. The knowledge graph is integrated into the agent workflow to validate actions before tool execution and to update the system state afterward. The approach is evaluated first on the real MMH platform and then in simulation with and without knowledge graph guidance. The results show improved decision consistency, higher success on complex multi step tasks, and fewer unnecessary tool calls and delays. These findings suggest that structured knowledge can improve the reliability of LLM based control in cyber-physical intralogistics systems. Additional details on the implementation and experimental setup are available in the project repository at https://github.com/MakaremHind/warehouse-agent-with-knowledge-graph.
Authors: Hind Makarem, Lars Ohnemus
Language: English
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7222473566
