/DeepTraject

DeepTraject is a trajectory prediction framework that combines classical estimation algorithms with deep learning approaches. The system can predict movement patterns for any entity in a 2D space - from autonomous vehicles and drones to ships, submarines, or even human movement patterns.

Primary LanguageHTMLMIT LicenseMIT

DeepTraject: Multi-Model Trajectory Prediction Framework

Python 3.8+ PyTorch Code style: black License: MIT

🎯 Overview

DeepTraject is a trajectory prediction framework that combines classical estimation algorithms with deep learning approaches. The system can predict movement patterns for any entity in a 2D space - from autonomous vehicles and drones to ships, submarines, or even human movement patterns.

System Architecture

Key Features

  • 🤖 Hybrid Prediction Architecture

    • Kalman Filter for state estimation
    • LSTM with attention mechanism for pattern learning
    • Ensemble approach combining both methods

  • 🎮 Grid-Based Simulation Environment

    • Configurable grid size and resolution
    • Support for continuous movement within discrete grid
    • Boundary handling and collision avoidance

  • 📊 Multiple Trajectory Types

    • Circular patterns
    • Figure-eight patterns
    • Lane-change maneuvers
    • Random walks with smoothing
    • Custom trajectory support

  • 📈 Real-Time Analysis

    • Interactive dashboards
    • Live prediction visualization
    • Performance metrics tracking
    • Uncertainty estimation

💡 Use Cases

  • 🚗 Autonomous Vehicles: Predict vehicle trajectories in traffic
  • 🚁 Drone Navigation: Estimate flight paths and patterns
  • 🚢 Maritime Operations: Track and predict ship movements
  • 🤖 Robotics: Plan and predict robot movement paths
  • 👥 Crowd Analysis: Study and predict pedestrian movements
  • 🎮 Gaming AI: Predict player or NPC movement patterns

🛠️ Technical Architecture

Grid System

  • Configurable 2D grid environment
  • Continuous movement within discrete grid cells
  • Customizable grid resolution and size
  • Boundary handling and wraparound options

Prediction Models

  1. Kalman Filter

    • 4-state tracking (x, y, vx, vy)
    • Real-time state estimation
    • Uncertainty quantification

  2. LSTM Neural Network

    • Sequence-based prediction
    • Attention mechanism for pattern recognition
    • Adaptive learning rate

  3. Ensemble Predictor

    • Weighted combination of predictions
    • Dynamic weight adjustment
    • Model confidence scoring

📊 Performance Metrics

Trajectory Type Kalman Error (m) LSTM Error (m) Ensemble Error (m)
Circular 0.45 0.38 0.31
Figure-Eight 0.52 0.41 0.35
Lane-Change 0.38 0.44 0.33

🚀 Getting Started

Prerequisites

Python 3.8+
PyTorch 2.0+
NumPy
Matplotlib
Plotly

Installation

cd deeptraject/public
pip install -r requirements/requirements.txt

Quick Start

from deeptraject.simulation import SimulationRunner
from deeptraject.utils.types import TrajectoryType

# Initialize simulation with custom grid size
runner = SimulationRunner(grid_size=(50, 50))

# Run single trajectory prediction
trajectory_type = TrajectoryType.CIRCULAR
await runner.run_simulation(
    trajectory_type,
    steps=100,
    radius=10.0
)

# Run full benchmark suite
await runner.run_benchmarks()

📈 Visualization Examples

Real-Time Prediction

Prediction Animation

Performance Dashboard

Dashboard Example

📁 Project Structure

public/
├── src/              # Core source code
├── tests/            # Test suites
├── docs/             # Documentation
└── experiments/      # Experiment results

🔬 Technical Details

Grid System

  • Configurable dimensions (default: 50x50)
  • Continuous position tracking within grid
  • Customizable boundary conditions
  • Support for multiple entities

Prediction Pipeline

  1. State observation
  2. Multiple model predictions
  3. Ensemble combination
  4. Error calculation
  5. Model adjustment

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

👤 Author

Michael La Rosa