WP 1 Innovation leads
Autonomous docking of marine surface vessels
Researcher/ Supervisor: Simon Lexau/ Anastasios Lekkas
Problem: One of the key challenges for autonomous operations is achieving safe and efficient autonomous docking in complex port environments.
Solution: New advanced planning algorithms for maneuvering ships through challenging environments and sophisticated low-level control systems that ensure the vessel follows the planned trajectory safely and with optimal energy efficiency.
Purpose & Value: Enable vessels to navigate safely through complex harbor settings amid wind, currents, and dynamic obstacles, while reducing energy consumption and minimizing wear on the vessel's thrusters.
Supervisory control of autonomous vessels through transparency and explainability
Researcher/ Supervisor: Joel Jose/ Erlend M. Coates
Problem: A key challenge in the development and deployment of autonomous ships is establishing trust with human users and society due to lack of transparency in autonomous decisions.
Solution: New interfaces for existing collision avoidance algorithms for to facilitate exchange of knowledge and interaction with human. Reliable AIbased collision avoidance systems that are interpretable to a human navigator and demonstrates innovative characteristics of providing explanations of its own actions in real-time.
Purpose & Value: Transparent collision and grounding avoidance systems that can provide clear explanations to validate their actions. This will increase confidence and foster trust with human users through effective, efficient and satisfactory interactions, hence enable adoption of the technology.
Combining Extended Object Tracking and Localization to Enhance Situational Awareness
Researcher/Supervisor: Martin Baerveldt/ Edmund Brekke
Problem: Autonomous vessels need reliable situational awareness from sensor data, but separating object tracking from localization and mapping causes poor data association and uncertainty handling. This fragmentation reduces safety and reliability in dense, constrained waterways, increasing collision risk and limiting safe autonomous deployment.
Solution: An integrated situational awareness system fuses all sensors to jointly perform tracking, mapping, and localization with explicit uncertainty handling. Using extended object tracking and robust data association, it accurately models vessels and static features while leveraging multiple sensors to improve reliability in complex waterways.
Purpose & Value: Partners developing autonomous vessel systems can use the solution as a framework to enhance sensor fusion, perception, and situational awareness capabilities. As such, it can provide new technical solutions that enable autonomous vessels to operate safely in more complex environments.
Radar based localization and mapping
Researcher/ Superviser: Henrik Flemmen/ Edmund Brekke
Problem: Global Navigation Satellite Systems (GNSS) is the primary technology for positioning for both manned and unmanned ships. Still, it might fail in rare cases, and GNSS redundancy by alternative solutions is needed.
Solution: A fallback solution for situations where the ship is in vicinity of land is developed based on radar information.
Purpose & Value: The solution may represent a potential automatic fallback solution for position location under loss of GNSS provided position data.
Caspar – Multi sensor detection
Researcher/Supervisor: Emil Martens/Annette Stahl
Problem: Robotic tasks like planning, estimation, and calibration are often nonlinear optimization problems. Unconstrained optimization is typically preferred in real-time robotics because it is simpler and more efficient, especially when problem structure allows constraints to be handled implicitly
Solution: The Caspar software library provides a high-performance GPU-based nonlinear solver for robotics by generating optimized CUDA kernels from symbolic Python expressions. Built on SymForce, it lets users define residuals symbolically and achieves 5–20× speedups over leading bundle-adjustment solvers.
Purpose & Value: Making large optimization problems run 5-20 times faster than SOTA and up to 1000 times faster than established solutions like GTSAM, opens up a word of new possibilities in real time robotics by enabling real-time optimization of a new class of problems.
Safe-Space Estimation for Autonomous Surface Vessels
Researcher/ Supervisor: Johannes Skarø/ Edmund Brekke
Problem: While existing situational awareness systems estimate vessel states and motion, they do not provide a direct representation of which regions of the environment will remain safe for navigation. Current approaches insufficiently account for vessel interactions, spatial extent, uncertainty in detections, and situations where AIS data are unavailable.
Solution: A situational-awareness solution that probabilistically predicts current and future safe navigable space for autonomous vessels, unifying dynamic traffic, static obstacles, and constraints. It identifies collision-free regions over time, enabling consistent risk assessment and maneuver planning across perception, prediction, and planning systems.
Purpose & Value: The result can be implemented by technology providers as part of perception and prediction stacks, enabling planners to directly query future safe navigable space rather than relying solely on object-level predictions. In addition, the solution can be applied in decision-support tools for manned vessels to improve navigational safety in complex traffic environments.
Mission Planning for MASS systems
Researcher/ Supervisor: Mikkel Bergstrand/ Anastasios Lekkas
Problem: Mission planning tools represents a void in the MASS autonomy stack; a high-level module which is able to inform the low-level system in how to proceed in solving tasks at a more abstract level, such as transferring passengers or cargo, and respond to contingency scenarios in a methodical way.
Solution (concept): Develop and evaluate real-time MASS mission planning by testing existing AI planners on milliAmpere, collaborating with partners on industry-relevant problems, and exploring new approaches including improved temporal planning, uncertainty modeling, reinforcement learning, and LLM-assisted generation of planning domains and problems
Purpose & Value: Integrating automated mission planning into MASS reduces manual workload, improves route efficiency, lowers energy use, and enhances remote operations. Increased autonomy enables fewer operators to manage more vessels, while automated re-planning during contingencies improves safety and reduces the risk of human error.
Mission Planning for Autonomous Surface Vessels
Researcher/ Supervisor: Miguel Hinostroza/ Edmund Brekke.
Problem: A central challenge in achieving Autonomous Surface Vessel (ASV) autonomy lies in the mission planning system, an architectural layer responsible for interpreting high-level commands, reasoning over spatiotemporal constraints, and generating executable tasks. Systematic validation that integrates both simulation-based studies and real-world experimental evaluation remains limited.
Solution: By combining temporal planning with standardized middleware and real-world testing, a proposed ROS-based mission planning framework addresses key limitations of existing approaches related to robustness, comparability, and practical deployability in maritime mission planning.
Purpose & Value: The developed mission planning system has been implemented onboard the milliAmpere 1 research ferry prototype, and provides a practical and flexible platform for researchers and partners in the SFI to develop, test, and validate autonomous navigation, control, and decision-making algorithms under real operational conditions.
Digital Twin for Situational Awareness and Optimal Control
Researcher/ Supervisor: Daniel Menges/ Adil Rasheed
Problem: Reliable environmental perception and awareness of a vessel's internal state, such as engine conditions, are critical for improving decision-making and preventing outages during operations. Digital twins may enable this but must combine accurate models, real-time data streams, and intelligent algorithms to deliver optimal, realworld decisions.
Solution: Adaptive tools for estimating environmental forces, proactive control methods for collision avoidance, and efficient algorithms for anomaly detection and big data processing. A focus is placed on creating digital twins for predictive monitoring and utilizing learning-driven frameworks for adaptive control and model identification.
Purpose & Value: Optimize decision-making and operational efficiency.
Simulation framework and software environment for evaluating automatic ship collision avoidance algorithms
Researcher/Supervisor: Trym Tengesdal/Tor Arne Johansen
Problem: systematic and extensive testing and validation of collision and grounding avoidance systems also in a simulation environment is missing. This is paramount for regulatory bodies and the public to gain trust in autonomous navigation systems.
Solution: A tailor-made framework for simulation and evaluation of maritime vessels with an automatic collision avoidance and anti-grounding (motion planning) system. The framework can be used for verification and assurance, and further enables automatic adjustment of collision avoidance systems online or offline
Purpose & Value: Framework for independent testing of new autonomous navigation algorithms. Standard and user-defined scenarios can be deployed for stress testing performance of a system, and hence contribute to validation of alternative solutions.
Sensor Rig for data acquisition in maritime environments
Researcher/Supervisor: Emil Martens/Anette Stahl.
Problem: High-quality datasets are critical for advancing ASV algorithms, but traditional data collection methods tend to be labor-intensive. Faster and better ways to acquire this data has been identified as a focus area.
Solution: A portable, standalone sensor rig was developed for easy acquisition of high-quality maritime data.
Purpose & Value: Planned to be used as a tool for data acquisition in further research in SFI Autoship.
