Vialidad-IO

Adaptive traffic-light control for a signalised intersection, modelled with Constraint Satisfaction (CSP) and Multi-Agent Simulation (Mesa).

The CSP layer enumerates every safe phase configuration (no two crossing arms green at once); the multi-agent layer runs vehicles through the intersection and uses those configurations for queue-adaptive phase switching.

---

Results

Valid CSP configurations	CSP vs. fixed-phase comparison

Simulation metrics (CSP control)

Combined animation

---

Project structure

Vialidad-IO/
├── notebooks/
│   ├── CSP_Traffic.ipynb       # CSP solver — enumerates valid phases
│   ├── modelL.ipynb            # Mesa multi-agent simulation
│   └── CSP_MultiAgent.ipynb    # Integrated model (CSP + agents)
├── models/
│   ├── main.tex                # Formal mathematical model (LaTeX)
│   └── main.pdf                # Compiled PDF
├── reports/
│   ├── csp_configurations*.png
│   ├── csp_multiagent_metrics.png
│   ├── csp_vs_fixed_comparison.png
│   └── figures/
│       └── csp_multiagent_simulation.gif
├── data/
├── docs/
└── requirements.txt

---

Architecture

CSP layer (CSP_Traffic.ipynb)

• Variables: W, E, NW, NE — binary state (0 = Red, 1 = Green) for each arm.
• Constraints: crossing arms cannot both be green; at least one arm must be green (liveness).
• Solver: hand-written recursive backtracking (backtrack() + is_valid()).
• Output: 6 valid phase configurations as list[frozenset[Arm]].

The formal definition is in models/main.tex / models/main.pdf.

Multi-agent layer (modelL.ipynb)

Built on Mesa 3.5. Three classes:

Class	Role
Arm (Enum)	Maps arm names to intersection geometry
TrafficLight	One per arm; holds state = "green" | "red"
Vehicle	Follows a 5-waypoint route; stops at the stop line when red
Cruce (Model)	Spawns vehicles on a schedule; runs adaptive phase control each step

Intersection center at (10, 10), plaza circle radius R = 2.2. Phase control switches to the CSP-valid phase with the longest queue after min_green steps, or forces a switch after max_green steps.

Integration (CSP_MultiAgent.ipynb)

Cruce accepts csp_phases: list[frozenset[Arm]] — the direct output of the CSP solver — as its set of allowable phases. Passing FIXED_PHASES (2-phase avenue + diagonal) reproduces baseline behaviour for comparison.

---

Setup

Two Python environments are used:

Environment	Purpose
.venv/ (Python 3.14)	CSP_Traffic.ipynb — matplotlib, numpy, pandas
analisis-vial/venv/	modelL.ipynb, CSP_MultiAgent.ipynb — requires Mesa 3.5

Install dependencies (analisis-vial venv)

pip install -r requirements.txt

Launch Jupyter

# Mesa notebooks
/Users/angelluna/Documents/STEM_PROJECTS/py/analisis-vial/venv/bin/jupyter notebook

# CSP-only notebook
source .venv/bin/activate && jupyter notebook

Compile the formal model

cd models && latexmk -pdf main.tex

---

Notebooks

Notebook	Description	Kernel
CSP_Traffic.ipynb	Enumerates and visualises all valid phase configurations	.venv
modelL.ipynb	Mesa simulation with fixed phases	analisis-vial
CSP_MultiAgent.ipynb	Full integrated model with CSP-driven adaptive control	analisis-vial

The FutureWarning about seed vs rng in Mesa output is expected and non-breaking.