SmartCity Multi-Agent Simulation

This project explores multi-agent systems for smart city traffic management.
Using JADE for agent communication and JENA for ontology reasoning, we built a simulation of vehicles, traffic lights, and cloud-level coordination to mimic an edge-cloud-user architecture.

Overview

The case study models a grid of city blocks with intersections managed by traffic light agents.
Key participants include:

• Vehicle Agents (user level): Move through streets, avoid collisions, request traffic lights to turn green.
• Traffic Light Agents (edge level): Control intersections, negotiate green times with other lights.
• Cloud Agent (cloud level): Mediates and coordinates between traffic lights using negotiation protocols.
• Environment Agent: Maintains shared world state using a blackboard model for real-time updates.

All knowledge is encoded in an ontology, created in Protégé, and queried via SPARQL to initialize the environment.

Architecture

Figure – Demo of the SmartCity multi-agent traffic simulation.

Communication & Coordination

• Blackboard model: Shared environment state (streets, vehicles, lights).
• Message Passing: Direct FIPA-ACL message exchange between agents.
• Negotiation (Contract-Net protocol): Traffic lights negotiate switching times.
• Consensus: Cloud agent enforces synchronized decisions across multiple lights.
• Pareto Optimality Heuristic: Balances waiting vs. flowing vehicles when deciding switch times.

Implementation Details

• Framework: JADE (Java Agent DEvelopment Framework)
• Ontology: Built in Protégé, loaded with Apache JENA
• Reasoning: SPARQL queries to extract semantic knowledge (streets, directions, agent placement)
• Concurrency: Agents run with reactive TickerBehaviour and negotiation behaviours (AchieveREInitiator, Contract-Net)

Example Use Case

• Streets are one-way and single-lane.
• At intersections, vehicles approaching a red light request a green phase.
• Traffic light agents negotiate via the cloud to coordinate switching.
• Vehicles stop if the distance to the next car is below a threshold, simulating safety measures.
• Yellow-light states prevent collisions during phase transitions.

Running the Simulation

From the repo folder:

cd SmartCity
make
make gui
make entorno

Or launch directly via the JADE GUI. The environment agent will initialize all others automatically.

Results & Extensions

• The system successfully models distributed traffic light negotiation, improving flow fairness.

• Introduced advanced concepts like Pareto optimality and consensus protocols for realistic coordination.

• Possible extensions:

  • Larger city maps
  • Integration with real IoT data streams
  • Alternative coordination strategies (e.g., reinforcement learning)

Tech Stack

• JADE – Multi-agent system platform
• JENA – Ontology loading & SPARQL querying
• Protégé – Ontology design
• Java – Core implementation

📂 Full source code: GitHub Repository