Ship Interactive Collision Avoidance Decision-Making Under Manned And Unmanned Mixed Navigation Conditions

Ship collision avoidance decision-making is a critical research topic in the field of shipping.With the rapid development of artificial intelligence technology,autonomous shipping technology research and development powerhouses,such as Norway,the UK,the US,Japan,and China,have conducted real-ship tests in specific encounter scenarios in actual sea areas.For a long time in the future,the sea will be in a state of mixed navigation between Maritime Autonomous Surface Ship with different degrees of autonomy and traditional manned ships.Due to the lack of real-time interaction capabilities with surrounding manned ships and the uncertainty of human operators’ steering behavior,the safe navigation between unmanned and manned ships is more challenging,making the research of safety interaction and collision avoidance between unmanned vessels and manned vessels under mixed navigation conditions of significant importance.This thesis focuses on the research of ship interaction and collision avoidance under the conditions of mixed navigation between manned and unmanned ships,aiming to improve the humanization,interactivity,and intelligence of ship collision avoidance decision-making.The main research content of this thesis is as follows:(1)A mixed navigation scene comprehension and quantification method is proposed to address the differences in collision avoidance behavior between manned and unmanned ship during the collision avoidance process.Firstly,the mixed navigation scenario is defined,and the collision avoidance mechanism and safety challenges for vessels under mixed navigation between manned and unmanned ships are analyzed.Considering the differences in risk and benefit perception among manned ship operators in the same encounter situation,vessel operators are classified into three sailing styles: aggressive,normal,and conservative.Furthermore,according to the "International Regulations for Avoiding Collisions at Sea" and steering practices,the collision avoidance process of ships under mixed navigation is quantified,including the classification of ship encounter situations,calculation of vessel motion parameters,and determination of ship collision risk.This provides theoretical support for the subsequent construction of collision avoidance decision-making model.(2)A ship-to-ship interaction collision avoidance behavior model based on game theory has been developed specifically for mixed navigation scenarios.Game theory is applied to solve the problem of unmanned ships interacting with manned ships and autonomously avoiding collisions.The ship’s course change is considered as the game strategy,and ship operator demands are modeled,and finely tuned to integrate psychologically expected safety benefits,social benefits that conform to collision avoidance rules,and economic benefits that consider ship energy consumption.On the basis of meeting the collision avoidance requirements,the difference in sailing style is characterized by introducing weight coefficients to quantify the perception of collision avoidance strategy benefits during the collision avoidance process.Finally,the sequential action order of ship collision avoidance games is designed to solve the Nash equilibrium of the stage game.(3)A decision-making method for ship interaction collision avoidance based on Nash-Q learning is proposed to address the characteristic of continuous interaction during the multi-ship collision avoidance process in mixed navigation scenarios.The ship collision avoidance problem under the condition of manned and unmanned mixed navigation is regarded as a multi-agent and multi-state Markov decision process.The multi-agent reinforcement learning method constructs the expression paradigm of the Markov game ship interactive collision avoidance process under the condition of mixed manned and unmanned mixed navigation scene and then solves it based on the Nash-Q learning algorithm,considering each state as a stage game,and the ship can interact with other ships in each state to obtain the optimal balance of each ship in the current stage.The collision avoidance action is to solve the optimal action sequence of each ship in the whole collision avoidance process.The simulation results show that the method proposed in this thesis can realize the safe encounter of multiple ships under the condition of manned and unmanned mixed navigation,and each ship can adjust its own collision avoidance behavior strategy according to the collision avoidance actions taken by other ships,similar to human-like interacts with other ships and independently performs collision avoidance operations to resolve collision risks.