AI and COLREGs in Autonomous Surface Vessels

Autonomous Surface Vessels (ASVs) rely on advanced artificial intelligence to navigate the world's oceans safely and legally. This article explores how these uncrewed ships utilize sensor fusion, machine learning, and automated decision-making algorithms to adhere to the International Regulations for Preventing Collisions at Sea (COLREGs), while also addressing the primary challenges of translating human-centric maritime laws into precise computer code.

Understanding COLREGs in the Autonomous Era

The International Maritime Organization (IMO) established COLREGs as the "rules of the road" for the sea. These regulations dictate how vessels must maneuver when encountering other ships to prevent collisions, defining specific scenarios such as head-on, crossing, and overtaking situations. Because COLREGs were written for human mariners, they include subjective terms like "good seamanship" and "ample time." For an ASV to operate legally and safely alongside crewed vessels, its onboard artificial intelligence must interpret these human-centric rules and execute predictable, compliant maneuvers.

Perception and Sensor Fusion

Before an AI system can make a navigation decision, it must understand its environment. ASVs achieve situational awareness through a process called sensor fusion, which combines data from multiple hardware sources:

  • Radar and LiDAR: Measure the distance, speed, and trajectory of surrounding obstacles.
  • Optical and Infrared Cameras: Powered by computer vision, these identify the type of vessel (e.g., sailboat, fishing boat, cargo ship) and read navigation lights at night.
  • Automatic Identification System (AIS): Receives broadcast data from other ships regarding their identity, position, course, and speed.

Machine learning models process this combined data in real-time to classify targets. Identifying the type of vessel is critical because COLREGs establish a hierarchy of right-of-way. For example, a power-driven ASV must generally give way to a sailing vessel or a vessel engaged in fishing.

Translating Rules into Algorithms

Once the AI perceives the environment, it must determine if a collision risk exists. ASVs calculate the Closest Point of Approach (CPA) and Time to Closest Point of Approach (TCPA) for every detected obstacle. If a risk is identified, the AI classifies the encounter into one of three primary COLREGs scenarios:

  1. Overtaking (Rule 13): The ASV is approaching another vessel from behind and must keep out of the way.
  2. Head-on (Rule 14): Two power-driven vessels are meeting on reciprocal courses, requiring both to alter their course to starboard (right).
  3. Crossing (Rule 15): Two vessels are crossing paths. The vessel that has the other on its starboard side is the "give-way" vessel and must alter its course, while the other is the "stand-on" vessel and must maintain its course and speed.

Developers use rule-based expert systems and fuzzy logic to translate these regulations into mathematical constraints. Fuzzy logic is particularly useful as it allows the AI to handle the grey areas of maritime navigation, converting subjective concepts like "safe distance" into actionable variables based on vessel speed, weather, and sea state.

Decision-Making and Path Planning

With the scenario classified, the AI’s path-planning module calculates a new route. ASVs typically use a combination of deep reinforcement learning and optimization algorithms to generate a safe trajectory.

During training, reinforcement learning models are run through millions of simulated maritime encounters. The AI is rewarded for reaching its destination efficiently and heavily penalized for violating COLREGs or coming too close to other vessels.

Furthermore, Rule 8 of COLREGs states that any action taken to avoid a collision must be "large enough to be readily apparent to another vessel observing visually or by radar." Therefore, the AI is programmed to avoid making a series of small, incremental course adjustments. Instead, it executes distinct, sharp changes in course or speed so that human crews on nearby ships can clearly see the ASV's intentions.

Overcoming the Challenges of Ambiguity

The greatest hurdle for AI in maritime navigation is the unpredictability of human behavior and the ambiguity of the rules themselves. Rule 17, for instance, requires a "stand-on" vessel to maintain its course, yet demands that it take evasive action if it becomes apparent that the "give-way" vessel is not taking appropriate action.

Teaching an AI the exact moment to break one rule to satisfy another requires advanced predictive modeling. Modern ASVs use behavioral prediction algorithms to monitor the actions of human-crewed ships. If a human-operated vessel fails to yield as required by law, the ASV’s collision avoidance system overrides standard protocol and enters an emergency evasion mode, prioritizing physical safety over strict adherence to right-of-way rules.