Reproduce “HiMAP-Travel: Hierarchical Multi-Agent Planning for Long-Horizon Constrained Travel” 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
Viet Bui, Wenjun Li, Yong Liu
HiMAP-Travel presents a hierarchical multi-agent framework for long-horizon planning tasks with complex, coupled constraints. It mitigates 'Constraint Drift' by separating strategic resource allocation from tactical execution, proactively enforcing constraints via a synchronized global state, and enabling parallel processing. The framework achieved a 17.65 percentage point improvement in final pass rate on a travel planning benchmark and demonstrated a 2.63x speedup on 7-day trips compared to sequential baselines.
Key insights
- Decoupling strategic and tactical planning through a hierarchical multi-agent design, combined with context isolation for executors, effectively mitigates 'Constraint Drift' in long-horizon tasks.
- Proactive, deterministic constraint enforcement via a synchronized global state and atomic operations is more robust and efficient than post-hoc error correction.
- A cooperative bargaining protocol enables flexible re-planning and adaptation to infeasible sub-goals or progressively revealed constraints, enhancing overall system robustness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.