Scalable Multi-Agent Coordination via Iterative Refinement
A novel framework for simultaneous target assignment and pathfinding (TAPF) separates target allocation from pathfinding through iterative refinement. In contrast to earlier methods that depend on Conflict-Based Search (CBS), which closely integrates these functions and restricts scalability, the new strategy utilizes rapid suboptimal MAPF solvers such as LaCAM. Within a specified time limit, it continuously resolves MAPF for the existing assignments, pinpoints bottleneck agents through feedback, and adjusts the assignments accordingly. Empirical findings indicate that this feedback-oriented cycle allows for greater scalability compared to CBS-based solvers while preserving solution quality.
Key facts
- The framework addresses the concurrent target assignment and pathfinding (TAPF) problem.
- Prior TAPF work exclusively used Conflict-Based Search (CBS), which is compute-intensive.
- The proposed approach decouples target assignment from pathfinding.
- It builds on modern, fast, suboptimal MAPF solvers such as LaCAM.
- Within a time budget, it repeatedly solves MAPF for the current target assignment.
- Bottleneck agents are identified via MAPF feedback.
- The assignment is refined based on feedback.
- Empirical results show the framework scales beyond CBS-based solvers with decent solution quality.
Entities
—