Autonomous Vessel Navigation

A comparative study of classical and reinforcement learning approaches to autonomous maritime navigation.

Phase 1: Classical Navigation (COMPLETE & TESTED)

Implemented:

• A* pathfinding
• Grid world environment
• Static obstacles
• Path Optimization - String pulling (Line of Sight)
• Vessel physics (Kinematic & Nomoto models)
• Path following (Pure Pursuit & ILOS)
• Dynamic obstacles
• Collision detection (CPA/TCPA)
• Collision avoidance (COLREGs-inspired)

Phase 2: Reinforcement Learning (IN PROGRESS)

Coming next:

• RL environment wrapper
• DQN/PPO implementation
• Training infrastructure
• Performance comparison

Setup

# Install dependencies
uv sync

# Run demos
uv run python examples/navigation_with_avoidance.py

Project Structure

• src/environment/ - Grid world and simulation environment
• src/pathfinding/ - A* and other pathfinding algorithms
• src/vessel/ - Vessel physics models
• src/visualization/ - Plotting and animation tools