Behavior-Based COLREGs Collision Avoidance

• Collision Regulations (COLREGs) Behavior-Based Approach is a method that models human maritime rules through explicit behavior-based protocols for real-time navigation.
• It employs rule-based arbitration, cost function shaping in optimization, and hybrid planning techniques to enforce legal maritime maneuvers.
• The approach integrates discrete rule logic with sensor-based state estimation, ensuring safe, efficient, and socially compliant navigation in mixed environments.

A behavior-based approach to COLREGs (International Regulations for Preventing Collisions at Sea) compliance relies on explicit modeling, classification, and execution of collision-avoidance maneuvers that mimic human seafaring rules in autonomous surface vehicles (ASVs) and other maritime robotics. This paradigm encompasses algorithmic frameworks that represent COLREGs as a hierarchy of behavioral protocols, typically realized through rule-based arbitration, cost function shaping in optimization routines, local replanning mechanisms, or direct embedding in control primitives. The approach differs from pure kinematic or physical modeling by incorporating discrete rule logic—such as give-way, stand-on, overtaking, or head-on situations—into the real-time navigation pipeline, ensuring not just geometric collision avoidance but explicit rule compliance for multi-agent and mixed human-autonomous environments.

1. Core Principles and Motivations

Behavior-based COLREGs strategies embed legal and social conventions of maritime navigation into the core architecture of collision avoidance algorithms. Fundamental principles include:

• Explicit mapping of encounter types (overtaking, head-on, crossing, stand-on/give-way) to formal rule representations (e.g., COLREGs Rules 13–17).
• Hierarchical or distributed arbitration, where ship-to-ship geometry determines which behavioral protocol is activated.
• Local, interpretable maneuver generation aligned with human-expectant actions, yielding social compliance and predictability.
• Modular composition, enabling integration with lower-level trajectory generation, optimization, or safety filters.

This approach is motivated by the recognition that geometric collision avoidance is insufficient in dense, human-interactive waterways, and that scalable, interpretable, and verifiable algorithms must directly reflect seafaring conventions (Thakar et al., 2023, Vries et al., 2022, Dubey et al., 2021, Lee et al., 27 Apr 2025, Paine et al., 16 Jan 2026).

2. Algorithmic Realizations: Architectures and Data Structures

Two-Phase Conflict-Resolution (COMCORE)

The COLREGs-Compliant Conflict-Resolving (COMCORE) strategy exemplifies a layered behavior-based approach (Thakar et al., 2023). It features:

• Phase 1: Independent shortest-path plans per agent using A* search on a discrete workspace grid.
• Phase 2: Identification and classification of conflicts with behavioral replanning:
  • Conflicts are characterized as head-on or crossing by analyzing joint paths on a 3×3 grid.
  • Labeling as give-way or stand-on is based on relative vessel positions per COLREGs Rule 15.
  • Local waypoint insertions realize protocol-compliant maneuvers (e.g., anti-clockwise detours for both head-on vessels, asymmetric detours for give-way vessels in crossing).

A solution tree maintains candidate joint plans, updating paths locally and validating for new conflicts in a recursive loop.

Optimization-based and Continuous Behavior Encoding

Alternative strategies encode COLREGs compliance as shaped cost functions within model predictive control (MPC) frameworks (Vries et al., 2022), where Gaussian-shaped “behavioral costs” penalize violation of rule-specific zones, e.g., port or starboard sectors, priority vessels, or overtaking corridors. In this paradigm:

• The relative bearing determines which regulation cost activates (head-on, crossing, or overtaking).
• The continuous cost landscape is parameterized so that rule-favored behaviors become cost-minimizing, ensuring social navigation even in dense or canal environments (Vries et al., 2022).

Hybrid and Sampling-based Planning

The hybrid VORRT-COLREGs algorithm combines rapidly-exploring random trees (RRT) with Velocity Obstacles (VO) (Dubey et al., 2021):

• RRT proposes candidate waypoints, while VO-based filtering prunes any extensions that violate collision or rule-compliance constraints (e.g., minimum Closest Point of Approach (CPA), required starboard detours).
• Each extension is forward-simulated and checked for admissibility under rule-specific angular bounds or motion constraints expressed by the COLREGs.

Control Barrier Function Approaches

Recent advances formulate behavior-based COLREGs compliance as switching sets of control barrier functions (CBFs) mapped to encounter types (Lee et al., 27 Apr 2025, Paine et al., 16 Jan 2026):

• For each dynamic scenario (e.g., head-on, crossing from starboard), a CBF encodes the correct avoidance direction (left/right turn), integrated as a linear constraint into a quadratic program enforcing collision-free, rule-compliant local control.
• In distributed multi-ASV settings, behavioral layers (e.g., MOOS-IvP colregs_bhv) synthesize nominal commands, which are then filtered by CBF constraints to guarantee safety even when other vessels are non-cooperative (Paine et al., 16 Jan 2026).

3. Formal Rule Encoding, Encounter Classification, and State Machines

Behavior-based methods require precise mathematical and computational representations of COLREGs situations. Typical components are:

• Encounter Type Classification: Using relative bearing, heading, and CPA/TCPA thresholds to partition space into head-on, overtaking, crossing, stand-on, give-way, and emergency/ambiguous modes (Thakar et al., 2023, Eriksen et al., 2019, Eriksen et al., 2019).
• Behavior Arbitration Automata: Band-structure or finite-state machines governing transition among behavior modules (e.g., stand-on, give-way, evasive action), with explicit handling of rule inversion in restricted contexts such as Rule 9 (narrow channels) (Hansen et al., 2022). The arbitration resolves conflicts between multiple simultaneous rules and handles changing situational geometry.
• Rule-to-Behavior Libraries: For each rule, a corresponding maneuver primitive is specified—e.g., starboard turns for head-on, port detours for overtaking, unaltered course for stand-on (Thakar et al., 2023, Dubey et al., 2021). Some methods enforce compliance by soft constraints through cost shaping, others as hard constraints via trajectory library selection or QP safety filters.

4. Integration with Optimization, Machine Learning, and Probabilistic Inference

Behavior-based COLREGs compliance can be embedded in several algorithmic substrates:

Model Predictive Control (MPC)

• Penalty Shaping: Rule-specific penalty regions (often elliptical or sector-based around moving obstacles) are constructed to reflect correct passing maneuvers (Vries et al., 2022, Eriksen et al., 2019, Eriksen et al., 2019).
• Constraint Enforcement: Discrete COLREGs state variables can be held fixed or switched as discrete modes in multiphase or receding-horizon optimizations (Bergman et al., 2020, Eriksen et al., 2019). Transitions are governed by encounter classifiers and finite-state diagrams.

Learning-Based and Probabilistic Approaches

• Deep Reinforcement Learning (DRL): COLREGs are encoded in reward shaping (e.g., negative penalty for entering wrong sectors or failing to maneuver early) or as hard constraints via action masking with formal temporal logic (Larsen et al., 2021, Krasowski et al., 2024, Meyer et al., 2020).
• Dynamic Bayesian Networks (DBN): Probabilistic models infer the intention of other vessels, conditioning trajectory predictions and maneuver compatibility on inferred rule-compliance and route/grounding hazards (Mahipala et al., 1 Apr 2025).
• LSTM-based Intention Models: Learned classifiers predict whether obstacles will pass port or starboard, allowing the ASV to plan active information-seeking and rule-consistent maneuvers under high uncertainty (Jeong et al., 2024).
• LLMs: LLMs (e.g., GPT-4) interpret input state (risk indices, relative heading, CPA/TCPA) and generate interpretable, text-based rule-compliant maneuver commands, integrated into lower-level trajectory and control modules (Agyei et al., 2024).

5. Representative Workflows and Pseudocode Patterns

Behavior-based COLREGs compliance algorithms typically follow the workflow:

1. State Estimation: Fusion of sensor data to build an up-to-date state representation (ownship, obstacles, static map) (Meyers et al., 2022).
2. Encounter Classification: For each obstacle, compute geometric/motion features, classify the encounter, and label the rule in effect.
3. Behavior Arbitration: Evaluate predicates for rule activation and select the applicable behavior module (e.g., keep course, alter course starboard, evasive maneuver) based on a priority scheme.
4. Maneuver or Trajectory Generation: Use rule-labeled collision avoidance/replanning to produce waypoints, control signals, or QP optimization with behavior-specific constraints.
5. Execution and Validation: Apply the chosen control action, validate for new conflicts, and repeat.

This workflow can be realized in a combination of event-driven (discrete) automata, continuous control updates, or sampling-based planning cycles (Thakar et al., 2023, Dubey et al., 2021, Meyers et al., 2022).

6. Performance, Validation, and Limitations

Empirical validation across a range of environments and agent counts demonstrates that behavior-based COLREGs compliance strategies:

• Achieve zero collision incidents and high command compliance rates in structured and unstructured simulations, and in real-world multi-vessel and adversarial scenarios (Thakar et al., 2023, Paine et al., 16 Jan 2026, Dubey et al., 2021).
• Scale efficiently to up to 10–15 vessels in general grid- or sampling-based planners, with sub-second planning times and negligible increase in cost or detour length relative to unregulated path planners (Thakar et al., 2023, Dubey et al., 2021, Lee et al., 27 Apr 2025).
• Preserve socially compliant trajectories and “human-like” decisiveness; however, limitations include the need for discrete scenarios (ignoring physical dynamics in some implementations), limited coverage of some COLREGs rules (e.g., overtaking in COMCORE), and the requirement of accurate state information and synchronized timing (Thakar et al., 2023, Lee et al., 27 Apr 2025, Paine et al., 16 Jan 2026).
• Extensions in recent work include intent-aware behavior under uncertainty (Jeong et al., 2024), integrated probabilistic inference and learning (Mahipala et al., 1 Apr 2025), and explainable decision-making via LLMs (Agyei et al., 2024).

7. Prospects and Open Research Directions

Emergent themes in state-of-the-art behavior-based COLREGs compliance research include:

• Extension to more complex rule sets (overtaking, restricted visibility), multi-ship roundabouts, and high-density traffic (Thakar et al., 2023, Dubey et al., 2021).
• Seamless integration with learning-based methods for intent inference, uncertainty quantification, and data-driven prediction of non-cooperative agents (Jeong et al., 2024, Mahipala et al., 1 Apr 2025).
• Real-time formal guarantees of safety, correctness, and completeness via hybrid CBF and temporal logic (Krasowski et al., 2024, Lee et al., 27 Apr 2025, Paine et al., 16 Jan 2026).
• Robustness in the presence of imperfect sensing, reactive human operators, and heterogeneous platform dynamics (Paine et al., 16 Jan 2026).
• Enhanced explainability and transparency using LLM-driven or interpretable symbolic encodings (Agyei et al., 2024).
• Generalization to restricted domains (narrow channels, grounding hazards), dynamic route planning, and context-dependent rule inversion (Hansen et al., 2022, Mahipala et al., 1 Apr 2025).

The behavior-based approach thus provides a modular, interpretable, and extensible recipe for embedding navigational rules in ASV systems, directly supporting scalability to realistic, congested, and socially critical marine domains.