Flow of autonomous traffic on a single multi-lane street

We investigate the behaviour of an original traffic model. The model considers a single multi-lane street, populated by autonomous vehicles directed from either end to the other. Lanes have no intrinsic directionality, and the vehicles are inserted at random at either end and any lane. Collision avoidance is fully automatic and, to enhance the transport capacity of the street, vehicles form_trains_ in which they may travel at high speed quite close to the vehicle in front. We report on the transit times for vehicles under a wide variety of conditions: vehicle insertion probability & imbalance and their maximum speed distribution. We also outline an interesting feature of the model, that the complex interactions of many vehicles are considerably more powerful than a simple “keep left” directive which each vehicle should obey.

💡 Research Summary

The paper presents a novel traffic simulation model that explores the dynamics of autonomous vehicles on a single multi‑lane street where lanes have no predefined direction. Vehicles are inserted randomly at either end of the street and may occupy any lane. Each vehicle is assigned an individual maximum speed drawn from a prescribed distribution, and a global insertion probability controls how frequently new vehicles appear. The core of the model is a fully automatic collision‑avoidance system that monitors the distance and relative speed to the vehicle ahead. When a following vehicle comes within a safety threshold, it synchronises its speed with the leader and joins a “train” – a tightly spaced convoy that travels at the leader’s speed. This mechanism allows high‑speed travel while keeping inter‑vehicle gaps minimal, thereby increasing the effective throughput of the street.

The authors conduct a systematic series of experiments varying three key parameters: (1) the insertion probability (from 0.01 to 0.20), (2) the imbalance factor that skews vehicle insertion toward one end of the street (0 % to 50 %), and (3) the statistical distribution of maximum speeds (e.g., normal N(30 km/h, 5 km/h) versus uniform U(20 km/h, 40 km/h)). For each configuration they record the transit time of every vehicle, compute average travel times, delay distributions, and overall throughput.

Results show that at low insertion rates the average transit time grows roughly linearly with traffic density, but beyond a critical density a sharp increase in train formation occurs. Trains act as moving buffers: they absorb fluctuations in arrival rates and smooth out speed heterogeneity, which prevents the system from entering a severe congestion regime. When the speed distribution is wide, fast vehicles are frequently forced to join slower convoys, which paradoxically reduces overall variance in travel times and improves average performance. Conversely, a narrow speed distribution yields fewer trains and a more pronounced sensitivity to insertion probability.

A particularly striking finding concerns the “keep‑left” rule, a simple lane‑selection heuristic that would force every vehicle to stay in the leftmost lane whenever possible. Simulations that imposed this rule produced higher average transit times and more frequent deadlocks than the baseline model, which allows vehicles to change lanes freely as part of the train‑formation process. The authors argue that the emergent, self‑organising behaviour of many interacting autonomous agents is far more effective at optimising lane utilisation than any static directive.

In the discussion, the authors relate these observations to concepts from complex‑systems theory, emphasizing that local interaction rules (collision avoidance and train joining) give rise to global optimisation without central coordination. They suggest that future urban traffic infrastructure could benefit from abandoning fixed lane directions in favour of dynamic, vehicle‑driven lane allocation, especially when autonomous fleets dominate. Potential extensions include coupling the model with multi‑intersection networks, integrating traffic‑signal control, and validating the approach with real‑world sensor data.

Overall, the study demonstrates that autonomous vehicle convoys on a direction‑agnostic multi‑lane street can dramatically increase transport capacity, and that the collective dynamics of many such vehicles outperform simplistic lane‑keeping policies. The insights provide a compelling argument for re‑thinking street design and control algorithms in the era of fully autonomous mobility.