PRISM: Complete Online Decentralized Multi-Agent Pathfinding with Rapid Information Sharing using Motion Constraints

TL;DR

PRISM tackles multi-task multi-agent pathfinding under constrained communication by integrating info packets carrying motion constraints into a Modified Conflict-Based Search framework. It proves completeness and deadlock avoidance and demonstrates scalability and robustness across diverse environments, including low-connectivity networks and narrow passages. Empirical results show PRISM outperforming centralized CBS in many settings and offering robust coordination where TPTS struggles, while closely matching solution quality in favorable conditions. The approach provides a practical, adaptable foundation for large-scale decentralized multi-agent planning in real-world scenarios such as search-and-rescue and logistics.

Abstract

We introduce PRISM (Pathfinding with Rapid Information Sharing using Motion Constraints), a decentralized algorithm designed to address the multi-task multi-agent pathfinding (MT-MAPF) problem. PRISM enables large teams of agents to concurrently plan safe and efficient paths for multiple tasks while avoiding collisions. It employs a rapid communication strategy that uses information packets to exchange motion constraint information, enhancing cooperative pathfinding and situational awareness, even in scenarios without direct communication. We prove that PRISM resolves and avoids all deadlock scenarios when possible, a critical challenge in decentralized pathfinding. Empirically, we evaluate PRISM across five environments and 25 random scenarios, benchmarking it against the centralized Conflict-Based Search (CBS) and the decentralized Token Passing with Task Swaps (TPTS) algorithms. PRISM demonstrates scalability and solution quality, supporting 3.4 times more agents than CBS and handling up to 2.5 times more tasks in narrow passage environments than TPTS. Additionally, PRISM matches CBS in solution quality while achieving faster computation times, even under low-connectivity conditions. Its decentralized design reduces the computational burden on individual agents, making it scalable for large environments. These results confirm PRISM's robustness, scalability, and effectiveness in complex and dynamic pathfinding scenarios.

Figures (8)

• Figure 1: This flowchart illustrates the three phases of PRISM and provides a high level outline. Included are line numbers that correspond to specific steps in Algorithm \ref{['alg:prism']}.
• Figure 2: Example of a system consisting of two local networks using bounded info packets. Greyed-out agents indicate resting agents. Bounded info packets are discarded once their flush time expires, the holder re-establishes communication with the packet’s origin agent, or the holder completes its task.
• Figure 3: Example of a system with three agents using infinite info packets. Greyed-out agents ($R_2$ and $R_3$) indicate those that have reached their resting positions, while the red square marks the goal position of the red agent ($R_1$). This perspective is from agent $R_1$; at timesteps 0-2, $R_1$ is unaware of $R_3$'s existence (indicated by $R_3$ being whited out). Upon receiving an infinite info packet, $R_1$ remembers $R_2$'s presence even upon leaving the network.
• Figure 4: Example of a conflict tree from a Modified CBS call involving three agents. $R_1^v$ is an info packet representing agent $R_1$, acting as a virtual agent within the conflict tree, while $R_2$ and $R_3$ are network agents. Conflicts between $R_1^v$ and network agents result in the creation of a single child node, preserving the static path of the virtual agent. In contrast, conflicts between network agents follow standard CBS behavior, generating two child nodes. Conflicts are highlighted in red and the solution node is highlighted in green.
• Figure 5: This figure shows the environments and examples of (a) 10% proximity and (b) LoS communication protocols. Note that for LoS, it is assumed agents can see all other agents within a 4 diameter grid cell proximity of itself to avoid immediate collisions. Agent positions are shown in red and the range of communication is shown in green.

Theorems & Definitions (16)

• Definition 1
• Definition 2
• Definition 3
• Definition 4
• Definition 5
• Definition 6
• Lemma 1
• proof
• Lemma 2
• proof