Reproduce “Multi-Goal Multi-Agent Pickup and Delivery” 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
Qinghong Xu, Jiaoyang Li, Sven Koenig, Hang Ma
Xu et al. introduce LNS-PBS and LNS-wPBS, two decoupled algorithms designed for efficient coordination of multiple agents in multi-goal pickup and delivery tasks. LNS-PBS offers completeness guarantees for well-formed problem instances, while LNS-wPBS achieves superior scalability, effectively handling thousands of agents and tasks with reduced service times and stable runtimes compared to existing methods.
Key insights
- Combining sophisticated task sequence assignment with sequence-of-goal pathfinding significantly improves overall system efficiency in multi-agent pickup and delivery.
- A practical trade-off exists between theoretical completeness and real-world scalability, where windowed planning (LNS-wPBS) can achieve remarkable scalability for thousands of agents and tasks.
- Leveraging even partial knowledge of future tasks in a semi-online setting consistently reduces average service times, demonstrating the value of look-ahead planning in dynamic environments.
- The integrated Large Neighborhood Search effectively minimizes service times for complex Multi-Goal MAPD instances through its iterative refinement process for task assignments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.