Autonomous Software Agents Project

This project was developed for the Autonomous Software Agents course at the University of Trento during the 2024/2025 academic year.

by
Fabio Missagia
Alessandro Sartore

The goal is to implement an autonomous agent using the Belief-Desire-Intention (BDI) model, which allows the agent to perceive the environment, form goals, plan actions, and continuously adapt to dynamic conditions.

The agents operate in a simulated game environment where they must pick up and deliver parcels efficiently, as shown in the examples below.

In addition, the project consists of two parts:

• Single Agent: A standalone agent that independently explores the environment, updates its beliefs, and makes decisions to fulfill its goals.

  

• Multi-Agent: A system of two cooperating agents that share information, coordinate strategies, and divide tasks to complete deliveries more effectively.

  

Performance is evaluated across multiple game levels. Results and insights are included in the report under the Evaluation and Results section.

The environment is based on the Deliveroo.js simulation framework.

🗂️ Project Structure

Here's a description of the project structure.

.
├── README.md
├── config.js
├── main.js
├── package.json
├── package-lock.json
├── yalc.lock
├── update.sh
├── docs/
│   ├── report.pdf             # Formal report
│   └── presentation.pdf       # Project presentation slides
└── src/
    ├── agent.js               # Agent initialization and lifecycle
    ├── beliefs/
    │   └── beliefs.js         # Stores the agent's perceived entities
    ├── desires/
    │   └── desires.js         # Goal representation
    ├── plans/
    │   ├── PDDLplan.js        # PDDL planner integration
    │   ├── delivero-domain.pddl # PDDL domain definition
    │   └── pathfinder.js      # Pathfinding logic
    ├── coordination/          # Multi-agent coordination
    │   ├── area_manager.js         # Manages area division between agents
    │   ├── communication_handler.js # Handshake initiator for agent communication
    │   └── handover_coordinator.js  # Coordinates roles in corridor-like maps
    ├── strategies/            # Agent decision-making strategies
    │   ├── delivery.js             # Best strategy for parcel delivery
    │   ├── exploration_manager.js  # Best strategy for map exploration
    │   ├── parcel_selector.js      # Best strategy for selecting parcels
    │   └── planner.js              # Core of the agent logic
    └── utils/
        ├── message.js         # Defines the message structure for communication
        └── utils.js           # Contains constants, thresholds, and helpers

🚀 Installation & Usage

To try the autonomous agents, follow these steps:

1. Clone the repository:

   git clone https://github.com/Sartor02/Autonomous_Software_Agents.git

2. Navigate to the project directory:

   cd Autonomous_Software_Agents

3. Install dependencies:

   npm install

4. Start the project:

   npm start

You will be prompted to choose whether to launch a single agent or a multi-agent system.

Additionally, you will be asked to provide agent tokens: make sure to paste the token corresponding to the agent(s) you created on the selected platform!

📚 Additional Information

If you want more information about the technical aspects, such as the communication mechanism, the exploration algorithms, or how the agent selects the best parcel, take a look to the presentation slides! For a formal write-up, see the final report instead.