Reproduce “Coordinated Control of Multiple Construction Machines Using LLM-Generated Behavior Trees with Flag-Based Synchronization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Akinosuke Tsutsumi, Tomoya Itsuka, Yuichiro Kasahara, Tomoya Kouno, Kota Akinari, Genki Yamauchi, Daisuke Endo, Taro Abe and 3 more
Researchers from Kyushu University, Public Works Research Institute, and University of Tsukuba developed an LLM-driven workflow to automatically generate Behavior Trees for coordinated control of multiple construction machines. Their approach enabled safe and successful cooperative earthwork operations with real excavators and dump trucks, achieving high success rates in complex simulated scenarios.
Key insights
- Decomposing complex natural language instructions into an intermediate Action Sequence significantly aids LLMs in generating coherent, coordinated plans for heterogeneous multi-machine operations.
- Explicit flag-based synchronization through a Global Blackboard provides a robust and scalable method for managing inter-machine dependencies and shared operational states in dynamic, real-world construction environments.
- Advanced Large Language Models (e.g., GPT-5.2, Claude-Opus-4.6) are highly effective in generating correct and executable multi-machine Behavior Trees, but less capable models exhibit substantial performance degradation without human intervention.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.