Agent-based Simulation of Human Movement Shaped by the Underlying Street Structure

Relying on random and purposive moving agents, we simulated human movement in large street networks. We found that aggregate flow, assigned to individual streets, is mainly shaped by the underlying street structure, and that human moving behavior (either random or purposive) has little effect on the aggregate flow. This finding implies that given a street network, the movement patterns generated by purposive walkers (mostly human beings) and by random walkers are the same. Based on the simulation and correlation analysis, we further found that the closeness centrality is not a good indicator for human movement, in contrast to a long standing view held by space syntax researchers. Instead we suggest that Google’s PageRank, and its modified version - weighted PageRank, betweenness and degree centralities are all better indicators for predicting aggregate flow.

💡 Research Summary

The paper presents an agent‑based simulation study that investigates how the underlying street network structure shapes aggregate human movement. Two types of agents are defined: (1) random walkers that move without a destination, selecting the next street purely at random, and (2) purposive walkers that are assigned a specific destination and travel along shortest‑path or minimum‑cost routes. Both agent groups are deployed on large, real‑world street networks comprising thousands of streets and intersections. Over tens of thousands of simulation runs, the authors record the cumulative number of visits to each street – the “aggregate flow.”

The first major finding is that the distribution of aggregate flow across streets is governed primarily by topological properties of the network. Streets that occupy central positions, have many connections, or lie on many alternative routes consistently attract higher flow, regardless of the agents’ decision rules. The second finding is that the flow patterns generated by random walkers and purposive walkers are statistically indistinguishable. In other words, even when agents behave like real pedestrians with purposeful destinations, the macroscopic flow pattern is still dictated by the street layout; individual behavioral nuances have only a marginal effect on the overall distribution.

A third, more methodological contribution concerns the evaluation of centrality measures commonly used in space‑syntax research. The authors test several metrics – closeness, degree, betweenness, PageRank, and a weighted version of PageRank – against the simulated flow using correlation analysis. Contrary to the long‑standing belief that closeness centrality (a measure of average topological distance to all other nodes) is a strong predictor of pedestrian movement, the results show a weak correlation. By contrast, PageRank and weighted PageRank, which incorporate both the number of incoming links and the importance of those links, exhibit the highest explanatory power. Betweenness centrality (capturing the extent to which a street lies on shortest paths between other streets) and simple degree centrality also outperform closeness.

These findings have practical implications for urban planning and transportation engineering. First, when forecasting pedestrian volumes or designing interventions to redistribute foot traffic, analysts should prioritize PageRank‑based indicators rather than rely on closeness centrality. PageRank’s ability to capture the recursive influence of highly connected streets makes it a more realistic proxy for the “attractiveness” of a street in a networked environment. Second, the similarity between random‑walker and purposive‑walker flow suggests that relatively simple stochastic models can be employed to approximate aggregate movement patterns, reducing computational complexity while still delivering reliable macro‑level insights.

The paper concludes by outlining future research avenues: (i) validating the results across cities with different morphological characteristics, (ii) incorporating temporal dynamics such as peak vs. off‑peak periods and event‑driven surges, and (iii) extending the framework to multimodal networks that integrate pedestrians, cyclists, vehicles, and public transit. By addressing these extensions, the authors aim to refine the hypothesis that street structure dominates movement and to translate the insights into actionable tools for city managers and designers.