Lifelong Multi-Agent Path Finding in Large-Scale Warehouses (2005.07371v2)

Abstract: Multi-Agent Path Finding (MAPF) is the problem of moving a team of agents to their goal locations without collisions. In this paper, we study the lifelong variant of MAPF, where agents are constantly engaged with new goal locations, such as in large-scale automated warehouses. We propose a new framework Rolling-Horizon Collision Resolution (RHCR) for solving lifelong MAPF by decomposing the problem into a sequence of Windowed MAPF instances, where a Windowed MAPF solver resolves collisions among the paths of the agents only within a bounded time horizon and ignores collisions beyond it. RHCR is particularly well suited to generating pliable plans that adapt to continually arriving new goal locations. We empirically evaluate RHCR with a variety of MAPF solvers and show that it can produce high-quality solutions for up to 1,000 agents (= 38.9\% of the empty cells on the map) for simulated warehouse instances, significantly outperforming existing work.

• The paper proposes the Rolling-Horizon Collision Resolution (RHCR) framework to decompose and solve lifelong MAPF efficiently.
• It demonstrates significant improvements in computational speed and throughput in simulated large-scale warehouses with up to 1,000 agents.
• Key innovations include dynamic goal allocation and windowed planning, which adapt to ever-changing tasks while reducing computational load.

Lifelong Multi-Agent Path Finding in Large-Scale Warehouses

The paper "Lifelong Multi-Agent Path Finding in Large-Scale Warehouses" by Jiaoyang Li et al. addresses a complex variant of the Multi-Agent Path Finding (MAPF) problem, where agents are continuously assigned new goal locations over time, as opposed to the standard one-shot pathfinding task. This lifelong MAPF is particularly relevant in dynamic environments such large-scale automated warehouses where mobile robots constantly receive new tasks.

Problem Context and Challenges

MAPF involves finding collision-free paths for multiple agents traversing a graph, aiming for minimal makespan or flowtime. This optimization problem is classified as NP-hard when aiming for optimal solutions. The lifelong variant adds complexity by integrating the continual reassignment of goals, necessitating paths that are adaptable and can be reorganized efficiently.

Current methods for addressing lifelong MAPF generally fall into three categories:

1. Solving it as a whole in an offline setting, which is limited by scalability constraints.
2. Decomposing the problem into sequences of MAPF instances by replanning for all agents at each timestep.
3. Decomposing it similarly but restricting recalculations only to agents receiving new goals.

These approaches either suffer from inefficiencies in computational demand or limitations in application scope due to environmental specificity.

Proposed Solution: Rolling-Horizon Collision Resolution (RHCR)

The paper introduces a new framework: Rolling-Horizon Collision Resolution (RHCR). This approach decomposes lifelong MAPF into a sequence of Windowed MAPF instances. A Windowed MAPF solver in RHCR focuses on resolving collisions within a bounded time horizon, while ignoring interactions beyond that scope, thus reducing unnecessary computational efforts.

Crucial components of RHCR include:

• Windowed Planning: Collisions are resolved within a windowed view of the problem, effectively decreasing planning horizons and runtime without significantly compromising throughput.
• Dynamic Goal Allocation: Incorporating dynamic goal allocation whereby agents receive new assignments before idling occurs.
• Flexible Integration: RHCR can be integrated with various MAPF solvers like CA*, PBS, ECBS, and CBS, adapting to both complete and bounded-suboptimal solutions.

Empirical Validation

The authors demonstrate RHCR's efficiency in simulated warehouse environments with up to 1,000 agents. Empirical evaluations were conducted using two types of warehouse maps: fulfiLLMent centers and sorting centers. The results indicate that RHCR not only provides solutions faster than existing approaches but also maintains high throughput, a critical criterion for warehouse efficiency.

A notable finding is that optimal horizon lengths do not necessarily correlate with extensive collision foresight. RHCR with reduced time horizons performs comparably to, or even better than, solutions considering the entire lifespan of the system in certain settings.

Future Directions

The paper suggests several avenues for future research:

• Development of adaptive strategies for horizon length adjustments based on environmental congestion or computational constraints.
• Exploring parallelization techniques for agent group pathfinding to further enhance scalability.
• Investigating incremental search techniques that leverage prior computations for future path recalculations.

Conclusion

By proposing RHCR, which adeptly balances between computational efficiency and adaptive planning requirements, the work provides a robust framework for lifelong MAPF problems in highly dynamic, large-scale environments. The implications are significant for real-world applications in warehousing and logistics, marking a valuable contribution to the paper of adaptive multi-agent systems.