Multi-functional and Practical Adaptive Collision Avoidance Decision-making System for Autonomous Ships

In the current landscape, autonomous navigation systems face challenges due to incomplete functionality and low integration. They also lack a comprehensive, real-time, and accurate navigation risk assessment, while operating independently from one another. To address these issues, a multi-functional and practical adaptive collision avoidance decision-making system for autonomous ships is developed in this paper. The system primarily relies on the electronic navigational chart (ENC) to effectively display the ship’s navigation situation. It establishes a collision avoidance decision-making model centered around the own ship, offering collision risk analysis, avoidance methods, and optimal timing to ensure safe navigation. The system efficiently integrates and processes multimodal maritime data from various devices, utilizing ontology-based approaches for comprehensive navigation situation understanding, which integration provides invaluable support for intelligent decision-making processes. With a user-friendly interface, excellent portability, and cross-platform interoperability, this system has undergone collaborative efforts and joint debugging with China Shipbuilding Navigation Technology Co., Ltd. during the project’s first and second phases. Notably, the system has been successfully implemented on the “High-performance Integrated Bridge System” platform of China Shipbuilding Navigation, showcasing its potential for advancement from principle prototype development to actual equipment application. By offering accurate collision avoidance decision support, this system significantly contributes to enhancing the safety of ship navigation.


Introduction
With the continuous development of global trade, maritime transportation plays a vital role in international trade.Ships, as one of the primary modes of cargo transportation, have seen a constant increase in scale and quantity, leading to congested maritime traffic.However, maritime congestion brings challenges to ship navigation, and navigation safety issues have become increasingly prominent [1,2].Ship collisions and grounding accidents pose severe threats to human lives and property, making it urgent to seek a solution that can autonomously avoid collisions.Autonomous ships, as an innovative technology and a future development trend, have gathered significant attention.The autonomous collision avoidance navigation system is regarded as the "brain" behind ensuring the navigational safety of autonomous ships.It integrates various technologies, including navigation situation awareness, navigation behavior decision-making, motion control, intelligent supervision, and comprehensive testing [3,4].
In recent years, the fraunhofer center for maritime logistics and services CML (Fraunhofer CML) in Germany made significant advancements in the field of unmanned cargo ships through their research in the maritime unmanned navigation through intelligence in network (MUNIN) project, conducted between 2012 and 2015 [5].They developed and validated solutions for autonomous ships bridges, autonomous engine rooms, onshore operation centers, and communication architectures.Additionally, the Norwegian University of Science and Technology initiated the autonomous marine operations and systems (AMOS) project in 2013, with plans to develop safer, smarter, and more environmentally friendly ships and offshore intelligent platforms by the end of 2023 [6].Collaborating with Kongsberg, the SINTEF Ocean Laboratory involves developing the first electric propulsion unmanned container ship, the Yara Birkeland, which successfully completed autonomous navigation tests from the Herøya port to the Brevik port in 2018 [7].In addition, the center for Intelligent Maritime Vehicle of Dalian maritime university of China has developed the "zhihai-1" platform.It is equipped with attitude sensors, GPS receivers, microprocessors, and other electronic devices.It also utilizes wireless communication components to directly communicate with external mobile terminals [8].
With the advancement of technology, autonomous navigation systems are gradually maturing.However, current solutions in the market still face several challenges.Most autonomous navigation systems have high development and testing costs, limited functionality that fails to meet user requirements, and low integration [9,10,11].They lack a comprehensive, real-time, and accurate assessment of navigation risks, only calculating collision risk when a potential danger is visually detected [12,13,14].Moreover, there is a lack of open communication and interoperability among shipborne navigation instruments such as radar, AIS, ECDIS, and other onboard devices like autopilots [15,16].These systems operate independently, hindering information exchange [17].Therefore, at this stage, it is necessary to further expand the functionality and performance of autonomous navigation systems.This includes providing more features to meet diverse user needs.Additionally, there is a need to enhance the integration between different instruments and systems, enabling information transmission and interoperability, which will allow the system to comprehensively and real-time evaluate navigation risks and make corresponding decisions.Furthermore, enhancing the system's warning capability and intelligent collision avoidance decision-making ability is crucial, which would enable timely recognition of potential hazards and appropriate response measures [18].By addressing these challenges, autonomous navigation systems will be better equipped to handle complex navigational environments, improving the safety and efficiency of ships [19].
Therefore, to address the aforementioned issues, we have developed a multi-functional and practical intelligent collision avoidance decision-making system for autonomous ships.It primarily displays the ship's navigation situation against the backdrop of ENC.The system's main function is to establish a collision avoidance decision-making model centered around the own ship, taking into account the navigation mission requirements and encounter situations.The system integrates and processes multimodal maritime data from various devices, enabling comprehensive visualization.It utilizes ontology-based approaches to segment scenarios and achieve a holistic understanding of the navigation situation, laying the foundation for intelligent decision-making algorithms.By establishing the intelligent collision avoidance de cision-making algorithm based on operator's perspective (BOP) [20], the system enables intelligent collision avoidance decision, which provides collision risk analysis, avoidance methods, timing and magnitude of maneuvering, and the timing for resuming the original heading, ensuring the safe navigation of the ship.Additionally, it combines collision risk warning and trajectory estimation capabilities to dynamically plan the ship's route and generate real-time visualizations of algorithm results.The system features a user-friendly interface, strong portability, and cross-platform interoperability.Currently, the system has undergone collaboration and joint debugging with China Shipbuilding Navigation Technology Co., Ltd. in the first and second phases of the project, whic has been applied to the "High-performance Integrated Bridge System" platform of China Shipbuilding Navigation.Furthermore, the system has been recognized for its broad prospects, from principle prototype development to actual equipment application.The contributions of this paper are as follows.(1) The system's comprehensive integration of multimodal maritime data from various onboard devices addresses the challenge of low integration found in many existing systems.(2) The utilization of ontology-based approaches for scenario segmentation represents a novel aspect of our system.By categorizing navigation situations into entities and attributes, our system gains a holistic understanding of the maritime environment, setting the foundation for more intelligent decision-making algorithms.(3) The dynamic route planning and visualization capabilities of our system offer real-time adaptability to changing navigational conditions.(4) The user-friendly interface, strong portability, and cross-platform interoperability of our system contribute to its practicality and potential for wider adoption.

Intelligent Collision Avoidance Decision-Making System Architecture
The intelligent collision avoidance decision-making system is designed to display the navigational situation based on ENC.Its primary function is to form a collision avoidance decision-making model centered around the own ship, considering the requirements of the navigation task and the encountered situations.It provides analysis of collision risk, avoidance methods, timing and magnitude of maneuvering, and the timing for resuming the original heading, ensuring the safe navigation of the own ship.The system comprehensively processes the information from onboard sensors, including own ship's data, target ship's information, and other obstacles' data, and utilizes collision avoidance decision algorithms to calculate real-time control commands.The system presents the operations through various means, such as text, instruments, graphical, and sound.The intelligent collision avoidance decision-making system includes three subsystem: Collision Avoidance Decision-Making Demonstration Subsystem (SW1), Collision Avoidance Decision-Making Subsystem (SW2), and Comprehensive Ship Information Display Subsystem (SW3).
Among these, SW1 is a demonstration subsystem that achieves collision avoidance effects based on ENC.It first builds a navigation situation awareness model based on ontology.Subsequently, it calculates ship speed, heading, Time to Closest Point of Approach (TCPA), Distance at Closest Point of Approach (DCPA), encounter situations, and collision risk parameters for the own ship and different target ships using geometric methods.The system then utilizes the BOP algorithm to construct a collision avoidance decision-making model and derive decision results.SW2 primarily implements functional modules for encounter siuation scenarios' configurations, aids to navigation (AtoN) query, collision risk warning, and display mode switching.Moreover, SW2 provides a trajectory estimation function, which can estimate collision avoidance trajectories for the next 3-15 minutes.Based on socket communication services, it establishes information exchange among the three subsystems and employs big data technology to visualize DCPA, TCPA, collision risk, and changes in distance between ships.SW3 achieves real-time display of comprehensive ship information through modular web design using JavaScript technology.It mainly utilizes various means such as instruments, text, graphics, and sound for presentation.
The connection among the three subsystems is established through the same local area network, and the logical architecture of the system is shown in Fig. 1.The overall logical architecture of the system is divided into four layers: the data layer, service layer, application layer, and support layer.The data layer is primarily responsible for storing and managing chart data, navigation situation data, algorithm parameters, and service data.The service layer  The interaction among the three subsystems is depicted in Fig. 2. SW1 receives data related to the ship's navigation situation data and initializes encounter scenarios settings before transmitting them to SW2.If SW1 detects a collision risk, it applies geometric algorithms to compute the risk parameters and utilizes the BOP algorithm to determine collision avoidance decisions.The decision-making process is visually displayed, and the risk parameters are communicated to SW2 for visualizing DCPA, TCPA, collision risk, and distance.If SW2 needs to perform trajectory estimation, it sends the configured parameters to SW1.Lastly, SW3 comprehensively displays ship information based on the data parameters received from SW1 and SW2.

Collision Avoidance Decision-Making Demonstration Subsystem (SW1)
SW1 is developed using the standardized S-57 ENC, enabling the visualization of ship collision avoidance processes on the ENC.SW1 offers users an intuitive and user-friendly interface facilitated by graphical, dialog boxes, menus, and toolbars.The display dialog box located on the right side of SW1 provides ship details, including longitude, latitude, heading, and speed, as well as target ship information such as MMSI, ship name, ship type, longitude, latitude, heading, speed, relative bearing, relative distance, DCPA, TCPA, collision risk, collision risk warning, and more.The visual interface of SW1 is presented in Fig. 4. The computational principles of SW1, as illustrated in Fig. 5, begin by employing the "ontology" theory to categorize the ship's navigation situation into entities and attributes.This ontology-based approach enables the system to represent the maritime data in a standardized and semantically meaningful manner.The ontology provides a structured framework for organizing and classifying encounter scenarios based on the International Regulations for Preventing Collisions at Sea (COLREGS), thereby laying the foundation for collision avoidance decision-making.Furthermore, SW1 employs geometric algorithms to real-time calculate the speed, heading, TCPA, DCPA, encounter situations, and collision risk between the own ship and surrounding target ships.By analyzing the relative geometric relationship between the own ship and target ship, SW1 uses the BOP algorithm to minimize the total path for ship collision avoidance as the objective function, while considering ship maneuverability and rules as constraints, which quantifies the avoidance turning interval and solves the optimal collision avoidance strategy.

Collision Avoidance Decision-Making Subsystem (SW2)
The visual interface of SW2, as depicted in Fig. 6, provides a user-friendly platform for a range of functionalities, fostering efficient interaction with the system.Among its key features are encounter scenario settings, trajectory estimation settings, AtoN query, visualizations, and display mode switching.In Fig. 6, the primary focus lies in configuring the IP addresses between the three interconnected systems.This crucial step ensures seamless communication and data exchange between SW1, SW2, and other relevant components.Additionally, users can set various system parameters within SW2, such as defining the safe radius for ship navigation, simulation speed, the number of dynamic and static obstacle targets, simulation time, and overall path configuration.Furthermore, users have the flexibility to select and implement specific algorithms like B-splines for path planning.Another significant aspect of SW2 is the ability to set data storage names and pathways.This feature allows users to effectively manage and store the processed maritime data, enabling future analysis and reference.Moreover, the display mode switching feature of SW2 enhances the visual experience for users, offering three distinct modes: daytime, twilight, and nighttime.These modes cater to different lighting conditions, ensuring optimal visibility and usability throughout various operational scenarios.

Comprehensive Ship Information Display Subsystem (SW3)
The visual interface is shown in Fig. 8. SW3 is developed using JavaScript technology and adopts a modular design approach, categorizing and displaying relevant information.The main interface of SW3 is comprised of 10 display areas: software version information display area, routerelated information display area, own ship's navigational situation information display area, target ship's navigational situation information display area, predicted changes in own ship's heading and speed area, external environmental information display area (wind direction, wind speed, current direction, current speed, etc.), instrument panel area, collision risk information display area, collision avoidance decision information display area, the historical and predicted trajectories display area for own ship and target ships, and multi-ship collision risk warning area.

Figure 8. The visual interface
The operational principles of SW3 are shown in Fig. 9.In the instrument panel area, SW3 presents information such as heading, speed, wind speed, wind direction, current speed, and current direction.Additionally, it displays text information including time, position, heading, speed, chart name, scale, bearing and distance to the next turning point, route information, as well as details about target ships such as name, type, heading, speed, ship speed ratio, position, DCPA, TCPA, and collision risk.

The collaborative efforts and joint debugs
During the initial phase of collaboration with China Shipbuilding Navigation Technology Co., Ltd., the primary objective was to implement the functionalities of the three systems, which encompassed the display of navigational situation information, recognition and querying of AtoN, computation of collision risk information, design of trajectory estimation.Furthermore, provisions were made for external interfaces in anticipation of the second phase of collaboration.The collaborative efforts and joint debugs is shown in Fig. 10.

Figure 10. The collaborative efforts and joint debugs
During the second phase, the primary focus was on targeted enhancements based on the achievements of the first phase project.The main objective was to achieve interconnectivity and supervision among multiple intelligent platforms.Ontology technology was applied to estimate the navigational situation.And the system was augmented with the BOP algorithm to facilitate secondary optimization of collision avoidance decision-making, thereby enhancing the decision outcomes.Additionally, the system was designed to meet the requirements of radar plotting standards in maritime practices, including a system response time of less than 6 minutes.The Intelligent Collision Avoidance Decision-Making System encompasses various parameters, as illustrated in Table 1.To verify the overall reliability and usability of the system, a series of simulation experiments was conducted using the ENC of Tianjin Port's navigation channel.The experiments covered classic maritime encounter scenarios, including head-on encounters, overtaking encounters, and crossing encounters.Different numbers of target ships were designed for each encounter scenario.This paper focuses on presenting a specific set of simulation experiments involving a crossing encounter scenario.The initial parameters for the own ship and target ships were configured by the SW2, as shown in Table 2.In the initial stage, the system maintained the settings for the own ship unchanged, including the positioning of the initial turning point.The parameters of the target ship were modified to create a crossing encounter situation between the two ships.Partial screenshots showcasing key moments during the collision avoidance process are presented in Fig. 11.
The ship's motion trajectories were analyzed for each time period during the crossing encounter.In Fig11(a), the own ship followed the planned route in the initial navigation phase.In Fig11(b), the own ship detected the approaching ship and determined, through calculations, that the encounter situation was a crossing encounter.The target ship was positioned ahead and to the right of the own ship, making the own ship the give-way ship with the responsibility to yield.In Fig11(c), the own ship had the option to turn left or right, but considering the reference radar turning diagram, a right turn was considered safer as it allowed the own ship to pass astern of the target ship.The trajectory analysis reveals that the own ship successfully crossed behind the target ship, maintaining a safe distance throughout.In Fig11(d), after yielding, the own ship proceeded towards the target turning point, ultimately completing the voyage task.
During the joint debugging process, the system was tested using simulated encounter scenarios on an electronic chart background to verify its performance.The results indicated that the system effectively handled approaching ships from various directions and addressed collision avoidance issues in different encounter scenarios.The ship's trajectories demonstrate adherence to the principles of "Early, Large, Wide, and Clear" during evasive actions, aligning with the requirements of COLREGS.This indicates a high level of intelligence in ship operations.The method proves to be effective in dealing with potential collision situations in complex maritime traffic environments, thereby achieving intelligent ship collision avoidance.

Conclusion
The developed system addresses existing challenges in current autonomous navigation systems by offering a multifunctional and practical solution.It effectively integrates and visualizes multimodal data, utilizes scenario segmentation, and applies ontological principles to comprehend navigation situations, thereby enabling intelligent decision-making.Through the combination of high-precision ENC and BOP algorithms, the system provides various functionalities, including intelligent display of collision avoidance decisions, collision risk warnings, and trajectory estimation.It enables real-time dynamic planning of navigation routes and offers visual representations of algorithmic outcomes.By providing comprehensive situational analysis and decision support for collision avoidance, the system ensures the safety of ship navigation.Moreover, its user-friendly interface, strong portability, and cross-platform interoperability contribute to its wide-ranging applicability.Collaborative efforts and joint  Future work will focus on further optimizing system performance and conducting additional validation and application in real-world navigation scenarios to achieve an even higher level of maritime safety and efficiency.

Acknowledgments
The work is supported by the Dalian Science and Technology Innovation Fund (2022JJ12GX015).

1 .Figure 1 .
Figure 1.The logical architecture of the system

Figure 3 .
Figure 3.The physical architecture

Figure 4 .
Figure 4.The visual interface of SW1

Figure 9 .
Figure 9.The operational principles of SW3

Figure 11 .
Figure 11.The the collision avoidance process

Table 1 .
The navigation situation parameters

Table 2 .
The initial parameters