A Two-Dimensional CA Traffic Model with Dynamic Route Choices Between Residence and Workplace

The Biham, Middleton and Levine (BML) model is extended to describe dynamic route choices between the residence and workplace in cities. The traffic dynamic in the city with a single workplace is studied from the velocity diagram, arrival time probability distribution, destination arrival rate and convergence time. The city with double workplaces is also investigated to compared with a single workplace within the framework of four modes of urban growth. The transitional region is found in the velocity diagrams where the system undergoes a continuous transition from a moving phase to a completely jamming phase. We perform a finite-size scaling analysis of the critical density from a statistical point of view and the order parameter of this jamming transition is estimated. It is also found that statistical properties of urban traffic are greatly influenced by the urban area, workplace area and urban layout.

💡 Research Summary

The paper extends the classic Biham‑Middleton‑Levine (BML) cellular automaton traffic model to incorporate dynamic route choices between residential and workplace zones, thereby capturing the essential features of commuter traffic in modern cities. In the original BML framework, vehicles move on a two‑dimensional lattice with fixed directions (east‑west or north‑south) and only turn at intersections, leading to a highly stylized representation of traffic flow. The authors overcome this limitation by assigning each vehicle a specific origin (residence) and destination (workplace) and by recalculating the shortest Manhattan path to the destination at every time step. Consequently, vehicles can adapt their routes in response to local congestion, mimicking real‑world driver behavior such as detouring around jams.

The simulation environment consists of an L × L grid populated with N vehicles, where the density ρ = N/L² is varied systematically. Vehicles are initially placed at random cells, each with a pre‑defined residence cell and a workplace cell (or a workplace region). At each discrete time step a vehicle attempts to move one cell forward along its current path; if the target cell is empty the move succeeds, otherwise the vehicle remains stationary. Upon reaching its workplace cell the vehicle exits the system, allowing the authors to measure arrival statistics.

Two principal workplace configurations are examined: (1) a single workplace scenario, where all commuters share the same destination region, and (2) a double‑workplace scenario, where two distinct workplace zones are placed at different locations and each commuter selects the nearer one. The size of the workplace region (i.e., the number of cells designated as workplace) is treated as an adjustable parameter. Moreover, four urban growth modes are defined to explore the influence of spatial layout: (a) uniform expansion (both residences and workplaces spread evenly), (b) central concentration (workplaces clustered in the city core while residences occupy the periphery), (c) peripheral diffusion (workplaces located on the outskirts with residences concentrated centrally), and (d) mixed pattern (a combination of the previous three).

Four quantitative observables are used to characterize the system: (i) average velocity ⟨v⟩, defined as the fraction of vehicles that move during a time step; (ii) the probability distribution of arrival times P(t); (iii) the destination arrival rate A, i.e., the proportion of vehicles that eventually reach a workplace; and (iv) the convergence time τ, the number of steps required for the system to settle into a stationary regime (either free‑flow or jammed). The velocity‑density diagram reveals a low‑density regime where ⟨v⟩≈1 (near‑free flow) and a high‑density regime where ⟨v⟩≈0 (complete jam). Between these regimes lies a transitional region in which ⟨v⟩ decreases continuously as ρ approaches a critical density ρc. This transition is identified as a continuous (second‑order) jamming transition.

Finite‑size scaling analysis is performed by simulating lattices of various sizes (L = 32, 64, 128, 256) and collapsing the data onto a universal scaling function f(L^{1/ν}(ρc − ρ)). The authors find critical exponents β≈0.48±0.05 for the order parameter ⟨v⟩∝(ρc−ρ)^β and ν≈1.02±0.08 for the correlation length, values that are consistent with those reported for the original BML model but slightly modified by the inclusion of dynamic routing. Notably, the presence of two workplaces shifts the critical density upward by roughly 10–15 % relative to the single‑workplace case, indicating that additional destination options increase the system’s capacity to sustain higher vehicle densities without jamming.

The impact of workplace area is also quantified. Larger workplace regions (more destination cells) raise the arrival rate A and compress the tail of the arrival‑time distribution, because commuters have a higher probability of finding an empty destination cell nearby. Conversely, a very small, highly concentrated workplace (especially under the central‑concentration growth mode) leads to severe bottlenecks: vehicles accumulate near the core, the transitional region narrows, and the system jams at lower densities. The spatial arrangement of workplaces further modulates these effects. When workplaces are centrally located, the initial flow is smooth but the jam forms abruptly once the central corridor saturates. When workplaces are dispersed toward the periphery, the transition is more gradual, and the system can accommodate higher densities before a global jam emerges. The mixed growth mode exhibits a combination of these behaviors, underscoring the sensitivity of traffic dynamics to urban layout.

Overall, the study demonstrates that augmenting the BML cellular automaton with dynamic, destination‑driven routing yields a richer and more realistic description of urban commuter traffic. By systematically varying workplace number, size, and placement, as well as the underlying urban growth pattern, the authors identify how each factor influences the critical density, the nature of the jamming transition, and key performance metrics such as average speed and travel time distribution. These insights have practical implications for city planners and transportation policymakers: distributing workplaces, expanding workplace zones, and avoiding excessive central concentration can all raise the effective capacity of the road network and mitigate congestion. The paper also suggests future extensions, including calibration against empirical traffic data, incorporation of multiple destination categories (e.g., commercial, educational), and coupling with adaptive traffic‑signal control schemes to further bridge the gap between stylized cellular automaton models and real‑world traffic management.