Streaming Multi-agent Pathfinding

TL;DR

The paper addresses multi-agent pathfinding in assembly-line contexts where agents appear periodically and share the same action sequence, potentially yielding unbounded planning horizons. It formalizes Streaming MAPF (S-MAPF) with cycle time $c$ and introduces ASCBS, a two-level algorithm that enforces cyclic vertex/edge constraints to guarantee collision-free, optimal, and complete solutions. Key contributions include the S-MAPF formalization, the cyclic constraint mechanisms, exploration of disjoint splitting within ASCBS, and extensive experiments showing ASCBS outperforms traditional MAPF solvers under long working hours. The work also outlines extensions to Stay-in-Environment scenarios and Non-Uniform Cycle Time, broadening applicability to real-world assembly lines and human-robot collaboration, while maintaining computational efficiency.

Abstract

The task of the multi-agent pathfinding (MAPF) problem is to navigate a team of agents from their start point to the goal points. However, this setup is unsuitable in the assembly line scenario, which is periodic with a long working hour. To address this issue, the study formalizes the streaming MAPF (S-MAPF) problem, which assumes that the agents in the same agent stream have a periodic start time and share the same action sequence. The proposed solution, Agent Stream Conflict-Based Search (ASCBS), is designed to tackle this problem by incorporating a cyclic vertex/edge constraint to handle conflicts. Additionally, this work explores the potential usage of the disjoint splitting strategy within ASCBS. Experimental results indicate that ASCBS surpasses traditional MAPF solvers in terms of runtime for scenarios with prolonged working hours.

Figures (6)

• Figure 1: A snapshot of the S-MAPF problem where the cycle time is $2$. The snapshot is taken at a time step of $2 \times k$ where $k$ is a sufficiently large integer. White cells are feasible, while grey cells are infeasible. The dashed square marks the start point of the agent stream, and the dashed circle indicates the goal point. The solid circle represents an agent in the stream, with the number denoting the stream ID. Arrows depict the cells that the path crosses. The initial start times for agent streams 0, 1, and 2 are 0, 0, and 1, respectively, with action sequences 'RRRRRUUWUUU', 'LLLLLL', and 'DDDD'. 'U' stands for up, 'D' for down, 'L' for left, 'R' for right, and 'W' for wait.
• Figure 2: Examples of the conflicts where the cycle time is 2. The time step for the snapshot is $2 \times k$, where $k$ is a sufficiently large integer. The initial start times of agent streams $0$ and $1$ are $0$ and $1$. The conflict and the action sequences are shown at the top of the figure.
• Figure 3: Success rate and average running time for different implementations of ASCBS on the instances with cycle time $3$.
• Figure 4: Average running time of ASCBS and CBS, the number of feasible instances by both two algorithms, and average relative error of ASCBS compared to CBS on the instance with $10$ agent streams and cycle time $3$.
• Figure 5: Success rate and average running time for different implementations of ASCBS on the instances with $7$ agent stream.

Theorems & Definitions (14)

• Theorem 1
• Theorem 2
• Lemma 1
• proof
• Theorem 1
• proof
• Lemma 2
• proof
• Lemma 3
• proof