Desire Paths — see where people naturally walk

Desire paths (also called desire lines, social trails, or goat tracks) are informal routes created by people repeatedly choosing a path that better matches convenience, comfort, or directness than planned infrastructure. They emerge when official paths force detours, cross green spaces, or don't align with natural movement patterns.

Desire paths form through a positive feedback loop: the first few pedestrians take a shortcut, wearing down vegetation or soil. This worn surface becomes easier to walk on, encouraging more people to follow. Over time, repeated use solidifies the informal route into a visible path. Factors like terrain slope, existing infrastructure quality, and destination proximity influence where desire paths emerge.

Agent-based modeling (ABM) simulates the actions and decisions of individual agents — in this case, virtual pedestrians — to understand how collective behavior emerges. Each agent follows simple rules: choosing paths based on terrain resistance, route efficiency, and accumulated wear from other agents. The simulation produces emergent desire path networks that mirror real-world pedestrian routing behavior.

Urban planners can use desire path simulations to predict where pedestrians will create new routes before they exist. This helps in designing park layouts that align with natural foot traffic, identifying walkable versus impassable zones around facilities, and deciding where to build official paths to prevent environmental degradation from informal trails. A dense desire-line network signals high pedestrian permeability; gaps show where connections are severed.

H3 is a hexagonal hierarchical spatial indexing system developed by Uber. It divides the globe into hexagons at multiple resolution levels. In this simulator, H3 resolution 15 provides approximately 0.88-meter spacing, creating a fine-grained grid for precise pedestrian flow simulation over real-world terrain data.

Affordance theory, coined by psychologist James Gibson, refers to the action possibilities that an environment offers an organism. In pedestrian simulation, paved surfaces have high affordance (agents prefer them naturally), while grass and vegetation start lower but accumulate wear over time. The model tracks how traversed cells become more attractive — creating emergent desire paths through positive feedback.

Yes. The simulator exports flow networks as GeoJSON files containing hex cells with desireScore values. These can be imported directly into QGIS, ArcGIS, or other GIS tools for further spatial analysis, cartographic rendering, or integration with urban planning workflows.

This simulator extends the work of Bossowski et al. (2025): "Predicting Desire Paths: Agent-Based Simulation for Neighbourhood Route Planning," published in Computer Environment and Urban Systems (CEUS). It builds on Ma et al.'s agent-based model, adding weighted preferences for route length versus affordance trade-offs, and an obstacle-aware Dijkstra gradient pathfinding system.

Adjustable in the Walking behaviour tab: Comfort preference biases walkers toward worn-in ground; Shortcut preference trades a direct line for an easier surface; Look-ahead and Viewing arc set how far and wide a walker scans; Crowd size sets how many virtual walkers set out from each place; Grid resolution sets the H3 cell detail; Spontaneity adds chance to route choice; Paths wear in lets used ground grow more inviting.

At H3 resolution 15 a single cell is only about 1 meter across — thinner than a real footway and narrow enough that adjacent path cells can become disconnected. To keep linear features (footways, paths, tracks) connected and realistically wide, they are rasterized as a corridor widened by a landcover-aware Gaussian blur (radius 1, σ=1.0) around each cell along the line. The blur accumulates weight from every corridor source within radius, producing a smooth falloff that yields a roughly 3-meter-wide path — equivalent to a single resolution-14 cell — and improves path connectivity. The BFS does not propagate through cells with friction ≥ dense vegetation (4.0), so paths respect landcover boundaries and do not force walkers through bush or keep-off terrain. The widening radius is derived from a target path width and the resolution's cell spacing, so it stays approximately constant across resolutions: radius 1 at res 15, and 0 at coarser resolutions where one cell is already wide enough.