WP 4 Innovation leads
Marine cybernetics and safety for autonomous ships
Researcher/ Supervisor: Emir Cem Gezer/ Roger Skjetne
Problem: Safety assessment of autonomous multi-mode operation of a vessel where switching between a defined set of nominal operating modes and relevant safety fallback modes is a challenge when developing vessel control systems.
Solution: A modular testbed environment that enables efficient and secure testing of ship autonomy that will reduce cost and risk connected with testing and enable integration between different modules of a system solution.
Purpose & Value: Reduce costs and safety concerns connected with safety assurance of control systems for autonomous vessels.
AI and data-driven safety management in operation
Researcher/ Supervisor: Jon Estil Krågebakk/ Ekaterina Kim
Problem: Accurate sea-state estimation is essential for autonomous vessels, but traditional single-point methods miss key wave phenomena. Advanced spatiotemporal wave-field models enabling full 3D surface, velocity, and pressure reconstruction are needed, yet such methods are largely absent in current marine systems.
Solution: A novel approach with use of a polarimetric stereo-camera rig with GNSS and IMU data, combined with a machine-learning model, to infer surface elevation and flow fields. Missing input and pressure data are generated using REEF3D-based simulations through the open source Ægir-Frames dataset pipeline.
Purpose & Value: Predicting vessel trajectory based on predicted environmental influence on planned trajectory.
COLREG evaluation system
Researcher/ Supervisor: Sreekant Sreedharan/ Børge Rokseth
Problem: Ensuring an autonomous vessel behaves safely regardless of operational pattern is a complex challenge. The vessel’s location i.e. high seas, territorial waters, internal waters, port, or inland waters determines applicable regulations. In the high seas, international laws apply, but closer to shore, territorial and domestic rules take precedence due to security, safety, and traffic concerns.
Solution: The new solution encodes maritime rules in the Legata DSL and evaluates autonomous navigation via large-scale simulations using runtime resolvers and rule automata. It outputs quantitative compliance scores and scales through containerized HPC simulation capsules supporting digital twins and sensor-based testing.
Purpose & Value: The platform delivers machine-readable regulations, scalable simulation, and quantitative compliance scoring, enabling autonomy developers, class/assurance bodies, regulators, and researchers to test behaviours at scale, generate repeatable evidence, audit rule interpretations, explore operational concepts, and build shared benchmarks for MASS evaluation.
Risk-aware AI agents for safety-critical operations
Researcher/ Supervisor: Paul Lee/ Ingrid B. Utne
Problem: AI’s growing role in autonomous, safety-critical systems demands risk-aware agents. This research examines what AI risk awareness means and how to design and control such agents for safe autonomous ship operations, supporting real-time assurance frameworks for industry partners.
Solution: Risk awareness gaps are addressed through a literature review synthesizing a safety-based definition for AI. A novel deep reinforcement learning (DRL) framework trains autonomous-ship agents to recognize collision risk, produce explainable risk estimates, and incorporate this awareness directly into decision-making.
Purpose & Value: A novel deep reinforcement learning framework that embeds measurable, explainable risk awareness into autonomous-ship decision-making. Relevant for implementation in AI agents providing real-time risk monitoring, control of risk aware agents and assistance in certification processes.
Online risk modeling of autonomous ships
Researcher/ Supervisor: Susanna D. Kristensen/ Ingrid B. Utne
Problem: Developing ships with higher levels of autonomy includes developing systems that can perform more tasks independently from human operators. This means that a ship operating in a higher level of autonomy must itself be capable of monitoring and assessing the risk related to the operation it is performing, to make decisions for safe and efficient operations.
Solution: New online risk models with focus on collision and grounding risk are developed. The solution improve autonomous surface vessel (ASV) safety by integrating online risk models into supervisory risk control. These models, focused on collision and grounding hazards, support planning and decision-making. Their effectiveness is demonstrated through simulations and real-world trials, and the impact of different risk metrics is analyzed. The approach involves a full pipeline: hazard identification, model development, integration, and evaluation.
Purpose & Value: The new online risk models, focusing on collision and grounding risks consider various accident scenarios and influencing factors, enabling risk-aware and proactive decision-making. Testing shows the models help ASVs make safe and efficient decisions, both in simulations and real trials. Additionally, the models improve decision transparency and can be adapted for other autonomous systems with similar safety and efficiency requirements.
Methods and tools for the mitigation of drift grounding risk
Researcher/ Supervisor: Spencer Dugan/ Ingrid B. Utne
Problem: Failure of the propulsion, electrical generation, or electrical distribution systems onboard ships can lead to loss of command. This may lead to serious consequences, such as grounding, if command is not restored, or appropriate mitigation actions do not occur. For uncrewed ships loss of primary functions is especially critical.
Solution:
- Empirical insights on losses of command. Quantitative estimates of frequency of various vessel types adjusted for activity.
- Improved identification of risk influencing factors that impact the likelihood of loss of command and maritime causalities.
- Addressing biases and limitations in maritime accident data. • Improved decision–making of autonomous navigation systems using risk information and machinery parameters.
Purpose & Value:
- Insights useful for the design and operational practices to mitigate risk of both conventional and autonomous ships.
- Improved quality of accident and incident data collection and navigational risk assessments.
- Insights useful for the development of guidelines and standards related to autonomous ships.
- Safer and more intelligent autonomous navigation systems.