Sustained Heterogeneity: an emergent collective mechanism in LLM-driven traffic
Multiagent Systems
Summary
The authors studied how large language models (LLMs) control multiple vehicles on a circular road and found that the cars show human-like stop-and-go traffic waves. They ruled out several usual causes and discovered a new phenomenon called Sustained Heterogeneity, where small differences in speed choices by the LLMs persist and spread through the traffic flow. This behavior depends on traffic density and the proportion of cars controlled by LLMs. Their analysis shows that while LLM agents think carefully about safety, unpredictable differences still happen, meaning traffic stability needs direct control measures. This work is the first to identify this collective effect and describe when it appears based on car density.
Large language modelsMulti-agent systemsTraffic flowStop-and-go wavesSustained HeterogeneityRing road experimentTraffic densityCollision avoidanceChain-of-thought reasoningPhase transition
Authors
Yujun Qi, Yangyang Guan
Abstract
Large language models (LLMs) are increasingly adopted as closed-loop controllers in physical multi-agent systems, yet their emergent collective dynamics remain incompletely characterised. We deploy 22 LLM agents as direct, real-time target-speed controllers (per 0.5 s cycle, with IDM as collision-avoidance clamp) on a 230 m ring road under the Sugiyama 2008 paradigm, reproducing human-like stop-and-go waves. Six matched controls spanning stochasticity (white noise, OU noise, temperature), population variance, and dynamical instability (delay, OV model) are systematically excluded. The surviving phenomenon, termed Sustained Heterogeneity (SH), is the persistent, approximately temperature-insensitive (approx. 8 percent across a 6x T sweep), per-cycle divergence in LLM-chosen target-speed adjustments, propagating through a three-stage cascade of drift, gap erosion, and nonlinear braking. Across four traffic densities, the critical LLM penetration fraction p_c decreases monotonically from no transition at density 43.5 veh/km to p_c approx 0.23 at density 95.7 veh/km, consistent with an initiation-threshold model governed by trigger distance, stochasticity, and fleet size. Chain-of-thought analysis of 39,600 decisions across three seeds shows agents engage in multi-factor safety reasoning, yet systematic divergence persists, implying stability must be enforced at the dynamics layer. This is the first study to identify a previously uncharacterised collective mechanism in LLM-controlled traffic and map a density-dependent phase boundary p_c(rho).