Multi-Agent Coordination in Autonomous Vehicle Routing: A Simulation-Based Study of Communication, Memory, and Routing Loops

TL;DR

This paper identifies routing loops as a fundamental failure mode in memoryless decentralized rerouting for autonomous vehicle coordination and shows that naive information sharing can worsen performance. It introduces Object Memory Management (OMM), a lightweight protocol that maintains a distributed blacklist of blocked nodes to enable memory-aware routing and prevent repetitive detours. Through a 72-scenario experimental study on a graph-based AV routing simulator, OMM achieves substantial improvements, reducing travel and wait times by up to approximately 70-90% relative to memory-less baselines and maintaining stable recalculation costs. The findings underscore memory persistence as essential for robust multi-agent coordination in dynamic environments and have implications for decentralized routing across robotics, networks, and AI systems.

Abstract

Multi-agent coordination is critical for next-generation autonomous vehicle (AV) systems, yet naive implementations of communication-based rerouting can lead to catastrophic performance degradation. This study investigates a fundamental problem in decentralized multi-agent navigation: routing loops, where vehicles without persistent obstacle memory become trapped in cycles of inefficient path recalculation. Through systematic simulation experiments involving 72 unique configurations across varying vehicle densities (15, 35, 55 vehicles) and obstacle frequencies (6, 20 obstacles), we demonstrate that memory-less reactive rerouting increases average travel time by up to 682% compared to baseline conditions. To address this, we introduce Object Memory Management (OMM), a lightweight mechanism enabling agents to retain and share knowledge of previously encountered obstacles. OMM operates by maintaining a distributed blacklist of blocked nodes, which each agent consults during Dijkstra-based path recalculation, effectively preventing redundant routing attempts. Our results show that OMM-enabled coordination reduces average travel time by 75.7% and wait time by 88% compared to memory-less systems, while requiring only 1.67 route recalculations per vehicle versus 9.83 in memory-less scenarios. This work provides empirical evidence that persistent, shared memory is not merely beneficial but essential for robust multi-agent coordination in dynamic environments. The findings have implications beyond autonomous vehicles, informing the design of decentralized systems in robotics, network routing, and distributed AI. We provide a comprehensive experimental analysis, including detailed scenario breakdowns, scalability assessments, and visual documentation of the routing loop phenomenon, demonstrating OMM's critical role in preventing detrimental feedback cycles in cooperative multi-agent systems.

Figures (8)

• Figure 1: V2X communication architecture showing interactions among vehicles, infrastructure, pedestrians, and networks. Blue arrows represent short-range direct links (V2V, V2I, V2P), while orange arrows represent long-range network-assisted communications (V2N, I2N). Effective multi-agent coordination leverages this multi-modal communication infrastructure (adapted from yoshizawa2023survey).
• Figure 2: The graph-based road network used in all experiments, consisting of 86 nodes (intersections) and 161 directed edges (road segments). Blue nodes represent normal intersections, red nodes indicate obstacle locations in certain scenarios, and colored paths show example vehicle trajectories. This abstracted representation allows focused analysis of routing decisions without the confounding effects of detailed vehicle dynamics.
• Figure 3: Visual representation of the six experimental configurations showing the systematic progression from uncoordinated (Config 2) to fully coordinated (Config 6) multi-agent systems. Configuration 4 represents the critical failure case where reactive rerouting without persistent memory creates pathological routing loops, resulting in 542% performance degradation. Configuration 6 achieves optimal performance through the combination of communication, adaptive rerouting, and Object Memory Management.
• Figure 4: Sequential visualization of the routing loop phenomenon in Config 4 (Reroute without OMM). The sequence shows Vehicle 8 (orange) and Vehicle 6 (purple) on a graph where obstacles appear as red nodes. (Top-left) Initial state: both vehicles traveling toward their destinations. (Top-center) Vehicle 8 encounters obstacle at node 40, broadcasts information, and reroutes after 8s. (Top-right) Vehicle 8's new route encounters obstacle at node 37, and it reroutes again. (Bottom-right) Without memory of obstacle 40, the rerouting algorithm selects a path leading back toward it. (Bottom-center) Vehicle 8 returns to the vicinity of obstacle 40, "forgetting" it was blocked. (Bottom-left) The cycle repeats, with the vehicle trapped in an inefficient loop between the two obstacles. This visualization demonstrates why memory-less reactive rerouting degrades performance: locally optimal decisions at each step produce globally pathological behavior.
• Figure 5: Travel time distributions for 15 vehicles with 6 obstacles (left-to-right pattern). Note the extreme outliers in Config 4 (Reroute w/o OMM), where several vehicles exceed 100s due to routing loops. Config 6 (Reroute + OMM) shows tight distribution around 25-30s, indicating consistent performance.