Skip to content

README.md

Latest commit

 

History

History
86 lines (71 loc) · 2.87 KB

README.md

File metadata and controls

86 lines (71 loc) · 2.87 KB

Dynamic Retrocausal Simulator

(TemporalOS)

Overview

This repository implements a retrocausality simulation exploring how predictive modeling and rule-based behavior affect multi-agent interactions in a constrained grid environment.

The project simulates agents navigating a grid, using a Temporal Convolutional Network (TCN) to predict and avoid collisions. The goal is to demonstrate how retrocausal reasoning (predicting future states to influence current behavior) can lead to emergent coordination without explicit communication.

Project Structure

Dynamic-Retrocausal-Simulator/
├── config.py                 # Configuration settings
├── requirements.txt          # Python dependencies
├── archived_data/            # Old data
├── main.py                   # Main script for complete pipeline
├── src/                      # Source code
│   ├── agents.py            # Agent behavior and rules
│   ├── model.py             # Simulation model and TCN architecture
│   ├── tcn.py               # TCN training code
│   ├── data_gen.py          # Data generation utilities
│   ├── evaluate.py          # Model evaluation and analysis
│   └── visualize.py         # Visualization and animation
├── data/                    # Data files
├── models/                  # Trained model files
├── results/                 # Results and outputs
└── notebooks/               # Jupyter notebooks for exploration
    └── Explore_and_Debug.ipynb

Quick Start

Prerequisites

  • Python 3.8+
  • pip

Installation

pip install -r requirements.txt

Run Complete Pipeline

python main.py

This will:

  1. Generate training data
  2. Train the TCN model
  3. Evaluate performance
  4. Create visualization

Individual Steps

  • Generate data: python -m src.data_gen
  • Train model: python -m src.tcn
  • Evaluate: python -m src.evaluate
  • Visualize: python -m src.visualize

Methodology

Simulation Environment

  • Configurable grid size (default: 5×5)
  • Variable number of agents (default: 3)
  • Built using Mesa framework

Data Preparation

  • Agents collect position and relative position data over multiple timesteps
  • Sequences of 5 timesteps used for prediction
  • Features: agent position (2) + relative positions of other agents (10)

Model

  • Temporal Convolutional Network (TCN)
  • Predicts collision probability based on recent history
  • Used for retrocausal decision making

Rules

  • With Rules: Agents use TCN predictions to avoid collisions
  • Without Rules: Random movement with basic collision avoidance

Results

Check the results/ directory for:

  • Model evaluation metrics
  • Confusion matrices
  • Feature importance analysis
  • Simulation animations (GIF)

Development

Use the Jupyter notebook in notebooks/ for interactive exploration and debugging.