Technical Description
This invention introduces an autonomous marine navigation system that smoothly integrates path following and dynamic obstacle avoidance using an Epsilon Greedy-based Twin Delayed Deep Deterministic Policy Gradient (TD3) reinforcement learning framework. By pairing a 3-Degrees-of-Freedom ship dynamics model with non-linear multi-objective rewards, the system calculates continuous, jitter-free rudder adjustments while strictly complying with international maritime collision regulations (COLREGs).
Problems Addressed
- Absence of Continuous Learning
- Rigidly Programmed Failure Modes
- Regulatory Compliance Transgressions
- Brittle Handling of Complex Scenarios
- Lack of Comprehensive, Integrated Architecture
- Function Approximation Errors & Jerky Steering
Tech Features
- Epsilon-Greedy TD3 Core Architecture
- Multi-Objective Interactive Non-Linear Reward Product
- Safe Pass Index Boundary Regulation & Enforcement
- Hybrid Recurrent Reinforcement Learning Network
- Urgency-Triggered Proximity Module Switching
Target Audience
- Oceanographic Research Groups
- Environmental Monitoring Agencies
- Maritime Defense & Security Fleets
- Commercial Logistics Shipping Lines
Tech ID: P38-2096 TRL 2 Patent Status: Granted Available For Exclusive and Non-exclusive License
