Maritime Autonomous Surface Ships (MASS)

• Maritime Autonomous Surface Ships (MASS) are self-navigating vessels that combine onboard autonomy with remote control options for versatile maritime operations.
• They integrate modular sensor arrays, robust communication links, and advanced control algorithms to navigate safely while complying with COLREGs.
• Field validations and redundant systems, such as dual controllers and power architectures, ensure reliable performance in diverse marine environments.

A Maritime Autonomous Surface Ship (MASS) is defined as a surface vessel capable of operating independent of direct human control, encompassing all levels of autonomy from remotely controlled to fully autonomous operation. MASS systems integrate onboard autonomy, distributed sensing, advanced guidance, navigation, and control (GNC) stacks, high-throughput perception, and robust communication architectures to support long-duration missions across diverse and often challenging marine environments.

1. Fundamental Architecture and Operating Modes

MASS platforms are characterized by highly modular architectures that support multiple operating modes: manual onboard control (factory joystick), manual remote control (dedicated RF systems), autonomous waypoint navigation (with path planning computed either onboard or via a Ground Control Station), and autonomous velocity control using control algorithms such as PID, adaptive, or model-based controllers (Moulton et al., 2018). This multi-modal operation is implemented via dual-path control integration: a primary manual system (often retained for compliance and fail-safe operation) is augmented with an autonomous computing stack (e.g., Pixhawk PX4 micro-controller interfaced to the actuation chain via PWM) and middleware such as ROS for mission-level commands, sensor data fusion, and system health monitoring.

The ability to switch seamlessly between manual and autonomous control is critical in both routine and emergency scenarios. Safety circuits, such as remote engine shutoff implemented via relays triggered by autonomous controllers, ensure fault tolerance. Power system architecture typically involves a dual-battery series configuration, delivering, for example, a 24V supply for heavy-duty sensor suites and step-down regulated rails for sensitive embedded electronics (Moulton et al., 2018).

2. Sensor Integration and Situational Awareness

Robust situational awareness is realized through fusion of heterogeneous sensor modalities: above-water sensors (multimodal cameras, LIDAR, anemometers, radar, GNSS), and underwater sensors (bathymetric transducers, current meters, sidescan sonar, depth sensors). Communication links include long-range radio modems, WiFi for ad hoc local networks, and specialized remote control links. Sensor placement and mounting—especially for underwater elements—are optimized to avoid adverse hydrodynamic effects like cavitation, with modular mounting platforms enabling reconfiguration for mission-specific payloads.

A recurring challenge is the suppression of electromagnetic interference (EMI) from high-voltage marine powertrains, mitigated by careful wiring, grounding, and antenna placement (e.g., use of elevated masts for improved radio line-of-sight) (Moulton et al., 2018). Data from distributed sensor arrays are centrally processed using middleware (e.g., MAVROS on ROS) to support navigation, collision avoidance, and mission execution.

3. Autonomous Navigation and COLREGs Compliance

Achieving robust autonomy demands compliance with the International Regulations for Preventing Collisions at Sea (COLREGs). This involves a multilayered perception-to-action pipeline:

• Sensor data (radar, INS, Electronic Navigational Chart overlays) are fused to build an occupancy grid, separating static obstacles (e.g., land) from dynamic ones (other vessels).
• Dynamic obstacle trajectories are tracked via multiple hypothesis tracking using Kalman filters with constant velocity models, where state prediction is given by:

$$x' = F x + \nu, \qquad F = \begin{bmatrix} 1 & 0 & \Delta t & 0 \ 0 & 1 & 0 & \Delta t \ 0 & 0 & 1 & 0 \ 0 & 0 & 0 & 1 \ \end{bmatrix}$$

• Heading differences are used to classify COLREGs encounter situations (head-on, overtaking, crossing), and virtual forbidden polygons are constructed to represent the regulatory “keep-out” zones in the occupancy grid (Meyers et al., 2022).

For local path planning, reactive algorithms such as Visibility Graph Inspired Path Planning (VGIPP) operate directly on the updated binary grid, generating collision-free waypoints while respecting COLREGs-induced constraints. This approach provides sub-second replanning capability on standard CPUs and has shown operational success (e.g., 3.4 nmi missions with real-time regulatory compliance in constrained channels).

4. Learning-Based Perception, Control, and Multi-Modal Scene Understanding

Deep learning (DL) is integrated at multiple levels in MASS. For perception, deep convolutional networks are trained on datasets such as MassMIND—a long-wave infrared (LWIR) segmented dataset capturing coastal and deep-sea domains, annotated with categories required for navigation: sky, water, obstacles, living obstacles, bridges, and self-image (Nirgudkar et al., 2022). Models such as DeepLabv3, UNet, and PSPNet are evaluated via IoU, F1, and recall metrics, with DeepLabv3 excelling in navigationally critical class segmentation.

DL methods are applied to:

• Sensor fusion (multi-modal, cross-domain),
• State estimation (e.g., via domain-adapted CNN architectures capable of out-of-distribution detection),
• Guidance (learning optimal global trajectories from large-scale AIS data, reinforcement learning for local, COLREGs-aware planning),
• Control (adaptive neural controllers, hybrid DL-model–based MPCs), offering model-free compensation for vessel-environment uncertainties (Qiao et al., 2022).

Reward functions for DRL in guidance and control must encode both safety and rule compliance, and their optimal design remains a core research problem. Hybrid control architectures use neural estimators for uncertainties embedded within provably stable MPC or PID loops.

5. Mission-Scale System Integration, Redundancy, and Field Validation

Practical deployment of MASS hinges on rigorous field validation and robust redundancy:

• All mission-critical systems (controllers, sensors, power) are equipped with hardware redundancy, including dual controllers with hierarchical failover (primary/backup) and independent power/communication rails (Moulton et al., 2018, Arnaoot, 2 Apr 2025).
• Standardized wiring, labeling, documentation, and fleet-level quality assurance protocols are enforced to avoid hard-to-diagnose errors during fleet deployment.
• Pre-deployment checklists and routine maintenance regimes for system readiness (including health of engine, batteries, all communication links, and sensor arrays) are established following lessons from early field trials (Moulton et al., 2018).
• Realized test platforms, such as the AFRL Jetyak, are performance-validated for long-duration operation (up to 18 hours at low speed), with successful deployment in adverse environments.
• Adaptive control schemes incorporating learned disturbances (e.g., via Gaussian Processes) are highlighted as essential directions for future, high-performance MASS platforms.

6. Trade-Offs, Limitations, and Path Forward

MASS development encounters technical trade-offs:

• Modularity and sensor flexibility must be balanced against integration complexity and EMI risks, particularly for high-power vessels.
• Communication architecture must ensure low-latency overrides, redundancy, and safety circuity, but longer range introduces susceptibility to spectrum congestion and interference.
• Sensor placement, particularly for underwater payloads, requires careful mechanical and hydrodynamic analysis to achieve valid data under varying vessel dynamics.
• Calibration and redundancy (e.g., dual compasses for INS) are vital for reliable state estimation, especially in geomagnetically disturbed environments or GNSS-limited settings.
• Building a fleet amplifies the need for standardized integration procedures and wire color schemes to prevent idiosyncratic failures.

Future directions include the incorporation of advanced learning-based adaptive controllers, expanded multi-modal situational awareness, and more comprehensive field studies, as well as the codification of deployment best-practices for certification and regulatory acceptance. The advancement of holistic, fail-safe, and scalable autonomous surface ship architectures forms the foundation for the transition from experimental ASVs to fully realized MASS platforms (Moulton et al., 2018).