Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive-Horizon Conflict-Based Search for Closed-Loop Multi-Agent Path Finding

Jiarui Li, Federico Pecora, Runyu Zhang, Gioele Zardini

TL;DR

The paper tackles scalable closed-loop multi-agent path finding (MAPF) with formal guarantees. It introduces ACCBS, an adaptive-horizon solver built on a finite-horizon CBS that grows the planning horizon online while reusing a single constraint tree. The authors prove a cost-invariance property and establish completeness and asymptotic optimality, delivering an anytime planner that balances solution quality and computation. Empirical results on MAPF benchmarks show ACCBS achieving near-optimal performance within practical budgets and robustness to disturbances, outperforming reactive baselines while remaining scalable. This work bridges the gap between optimal but expensive planners and fast open-loop methods, enabling safe, robust deployment of large robot fleets in warehouses and logistics.

Abstract

MAPF is a core coordination problem for large robot fleets in automated warehouses and logistics. Existing approaches are typically either open-loop planners, which generate fixed trajectories and struggle to handle disturbances, or closed-loop heuristics without reliable performance guarantees, limiting their use in safety-critical deployments. This paper presents ACCBS, a closed-loop algorithm built on a finite-horizon variant of CBS with a horizon-changing mechanism inspired by iterative deepening in MPC. ACCBS dynamically adjusts the planning horizon based on the available computational budget, and reuses a single constraint tree to enable seamless transitions between horizons. As a result, it produces high-quality feasible solutions quickly while being asymptotically optimal as the budget increases, exhibiting anytime behavior. Extensive case studies demonstrate that ACCBS combines flexibility to disturbances with strong performance guarantees, effectively bridging the gap between theoretical optimality and practical robustness for large-scale robot deployment.

Adaptive-Horizon Conflict-Based Search for Closed-Loop Multi-Agent Path Finding

TL;DR

The paper tackles scalable closed-loop multi-agent path finding (MAPF) with formal guarantees. It introduces ACCBS, an adaptive-horizon solver built on a finite-horizon CBS that grows the planning horizon online while reusing a single constraint tree. The authors prove a cost-invariance property and establish completeness and asymptotic optimality, delivering an anytime planner that balances solution quality and computation. Empirical results on MAPF benchmarks show ACCBS achieving near-optimal performance within practical budgets and robustness to disturbances, outperforming reactive baselines while remaining scalable. This work bridges the gap between optimal but expensive planners and fast open-loop methods, enabling safe, robust deployment of large robot fleets in warehouses and logistics.

Abstract

MAPF is a core coordination problem for large robot fleets in automated warehouses and logistics. Existing approaches are typically either open-loop planners, which generate fixed trajectories and struggle to handle disturbances, or closed-loop heuristics without reliable performance guarantees, limiting their use in safety-critical deployments. This paper presents ACCBS, a closed-loop algorithm built on a finite-horizon variant of CBS with a horizon-changing mechanism inspired by iterative deepening in MPC. ACCBS dynamically adjusts the planning horizon based on the available computational budget, and reuses a single constraint tree to enable seamless transitions between horizons. As a result, it produces high-quality feasible solutions quickly while being asymptotically optimal as the budget increases, exhibiting anytime behavior. Extensive case studies demonstrate that ACCBS combines flexibility to disturbances with strong performance guarantees, effectively bridging the gap between theoretical optimality and practical robustness for large-scale robot deployment.
Paper Structure (10 sections, 3 theorems, 5 equations, 6 figures, 1 table, 2 algorithms)

This paper contains 10 sections, 3 theorems, 5 equations, 6 figures, 1 table, 2 algorithms.

Table of Contents

  1. Introduction
  2. Problem Definition and Terminology
  3. Finite-horizon CBS
  4. Horizon selection dilemma
  5. Active prefix and constraint-tree reuse
  6. Active prefix
  7. Details of algorithm
  8. Discussion and theoretical analysis
  9. Experiments
  10. Conclusion and Future Work

Key Result

Lemma III.1

Let $n$ be a node generated while the running horizon is $h$, and assume that for each agent $a_i$ the suffix $[v^{a_i}_h,\ldots,v^{a_i}_{H_{\max}}]$ follows a shortest path to $\rho_g^t(a_i)$. Then $J_h(n) = J_{h+1}(n).$ Consequently, $J_h(n) = J_{H_{\max}}(n)$ for all $h \in \{1,\dots,H_{\max}\}$.

Figures (6)

  • Figure 1: Constraint tree in . Colors indicate successive stages of the search. The same constraint tree is reused as the planning horizon grows, enabling seamless transitions between different horizon lengths (\ref{['prop:tree-reuse']}).
  • Figure 2: Example of active prefix. Only conflicts inside the active prefix (shaded region) are resolved; conflicts in the tail are ignored until the running horizon grows to cover them.
  • Figure 3: When the time budget increases, can produce better performance and approach the optimum, which is very expensive to compute.
  • Figure 4: can naturally deal with lifelong with changing goals. Similar to one-shot case, more time budget can lead to better throughput performance.
  • Figure 5: Tradeoff between planning time and performance. Performance gains from longer "thinking" time may not justify the cost, making budget selection non-trivial.
  • ...and 1 more figures

Theorems & Definitions (18)

  • Definition II.1: instance
  • Definition II.2: Trajectories and conflicts
  • Definition II.3: system, unified problem, and algorithms
  • Definition III.1: $H$-step trajectory
  • Definition III.2: Vertex and edge constraints
  • Definition III.3: Constraint satisfaction
  • Definition III.4: Cost function
  • Remark III.1: -style formulation of the cost
  • Definition III.5: Node in finite-horizon constraint tree
  • Definition III.6: Finite-horizon constraint tree
  • ...and 8 more