Modeling self-organizing traffic lights with elementary cellular automata

There have been several highway traffic models proposed based on cellular automata. The simplest one is elementary cellular automaton rule 184. We extend this model to city traffic with cellular automata coupled at intersections using only rules 184, 252, and 136. The simplicity of the model offers a clear understanding of the main properties of city traffic and its phase transitions. We use the proposed model to compare two methods for coordinating traffic lights: a green-wave method that tries to optimize phases according to expected flows and a self-organizing method that adapts to the current traffic conditions. The self-organizing method delivers considerable improvements over the green-wave method. For low densities, the self-organizing method promotes the formation and coordination of platoons that flow freely in four directions, i.e. with a maximum velocity and no stops. For medium densities, the method allows a constant usage of the intersections, exploiting their maximum flux capacity. For high densities, the method prevents gridlocks and promotes the formation and coordination of “free-spaces” that flow in the opposite direction of traffic.

💡 Research Summary

The paper presents a minimalist yet powerful cellular‑automaton (CA) framework for modeling and controlling traffic in an urban grid. Building on the well‑known elementary CA rule 184, which captures one‑dimensional traffic flow by moving a “car” (state 1) forward into an empty cell (state 0), the authors extend the model to a two‑dimensional lattice of streets intersecting at nodes. At each intersection three CA rules are employed: rule 184 for straight‑through motion, rule 252 to enforce a stop (red light) when a vehicle attempts to enter a busy intersection, and rule 136 to allow vehicles to exit (green light). This coupling yields a fully discrete, deterministic system that reproduces the essential dynamics of city traffic—formation of vehicle platoons, queue buildup, and the onset of gridlock—without recourse to differential equations or stochastic lane‑changing models.

Two traffic‑light coordination strategies are compared within this framework. The first is the conventional “green‑wave” method, which pre‑computes a fixed phase offset for each signal based on expected traffic direction, aiming to let a platoon travel through successive intersections without stopping. The second is a self‑organizing (SO) approach in which each intersection locally senses the density of approaching cars and the length of its waiting queue. The SO algorithm follows simple thresholds: if a vehicle is within a predefined distance, the current green phase is extended; if the queue exceeds a critical length, the light switches to red. Additionally, when overall density becomes high, the system deliberately creates “free‑space platoons” (contiguous empty cells) that move opposite to the traffic flow, thereby providing room for vehicles to advance and preventing complete blockage.

Simulation experiments sweep the global vehicle density ρ from low (≈0.05) to high (≈0.85). In the low‑density regime the SO method quickly organizes four‑directional platoons that travel at the maximal CA speed (one cell per time step) with virtually no stops, whereas the green‑wave benefits only the pre‑aligned direction and leaves the other three directions with unnecessary halts. In the medium‑density regime both schemes achieve high intersection utilization, but the SO method maintains the theoretical maximum flux of rule 184 (≈0.5 cars per cell per time step) by dynamically adapting phase lengths, while the green‑wave suffers from fixed offsets that cause occasional over‑saturation of some nodes and higher average waiting times. In the high‑density regime the SO strategy activates the free‑space mechanism: empty‑cell platoons propagate backward, opening gaps that let congested queues dissolve and averting the formation of a full gridlock. Quantitatively, the SO approach improves overall flux by 20‑40 % and reduces average travel time by roughly 25‑40 % compared with the green‑wave, and the incidence of deadlock drops from about 15 % to less than 2 % of simulation runs.

The key contribution of the work lies in demonstrating that a highly abstract CA model, using only three elementary rules, can capture the phase transitions of urban traffic (free flow, saturated flow, and jammed flow) and serve as a testbed for adaptive signal control. The authors argue that the emergence of platoons and counter‑flow free‑spaces provides a useful conceptual lens for designing decentralized traffic‑light algorithms that require only local information, making them scalable to large networks and robust to fluctuations in demand. The paper concludes with suggestions for future extensions, including multi‑lane streets, stochastic driver behavior, integration of pedestrian flows, and validation against real‑world sensor data.