Informed Sampling-based Collision and Grounding Avoidance for Autonomous Marine Crafts

Maritime technology has undergone significant developments in the past decades, but the frequency of collision and grounding incidents has not decreased. A majority of incidents are still attributed to human error [1]. This thesis focusses on algorithms for
decision support and autonomous navigation that allow humans or an autonomous system to navigate safely. A central element towards this is an effective collision and grounding avoidance system. Any vessel, autonomous or not, is required to adhere to the rules of safe navigation, the IMO COLREGs. The COLREGs describe the required navigational behaviours and describe the obligations between vessels when the risk of collision is imminent. The collision avoidance system must therefore comply with the COLREGs, to ensure safe navigation. Restrictions posed by the surrounding environment must also be accounted for to protect against grounding. The objective of this thesis is to research and develop novel collision and grounding avoidance algorithms, for use during fully or partial autonomous operation. To achieve these objectives, the thesis first investigates the COLREGs and general practises for safe navigation. Techniques within sampling-based motion planning (SBMP) are then used to develop a collision and grounding avoidance framework capable of considering the highfidelity nature of the Electronic Navigational Charts (ENCs) and the complex nature of the COLREGs. Sampling-based motion planning is an established paradigm for solving challenging planning problems and is selected as the main approach to investigate the collision and grounding avoidance problem. The thesis employs standard navigation methods to assess the risk of collision and to investigate which COLREGs are applicable in a given situation. Custom ship domains based on Lamé curves are proposed to bias the path planner toward finding route deviations in compliance with COLREGs rules 8 & 13-17. Conventional ideas within sampling-based motion planning are challenged to be able to extend the method from path planning to collision avoidance. When collision avoidance is performed, it is desired to leverage the underlying nominal path or route. This is done to ensure that the objectives encoded by the nominal route are included within the computed optimal deviation. As a result, novel cost functions were required to calculate paths with minimum route deviation, which were subsequently developed and presented, leveraging ideas from existing track-control problems. Furthermore, a data-driven objective function, based on historical navigation information, is investigated to include the notion of “good seamanship” when computing route deviations, so that they mimic the behaviour of human navigators. Methods are investigated to increase the performance and convergence properties of sampling-based motion planning algorithms and are achieved by introducing novel sampling strategies. An informed sampling strategy is presented that accelerates the convergence towards solutions with minimum path deviation. In addition, a data-driven sampling strategy is proposed, which takes advantage of past experiences of others. This is shown to be effective in rapidly finding solutions within the vicinity of prior data. The informed and data-driven sampling strategies are demonstrated for cases related to marine crafts as well as general sampling-based motion planning problems. The architecture and functional descriptions for an autonomy stack are presented
in detail, with a specific emphasis on the role of the collision and grounding avoidance module to obtain partial or full autonomy. The Short Horizon Planner (SHP) is introduced as the module responsible for computing COLREGs compliant and safe route deviations.
In general, the proposed methods are demonstrated in the context of both the merchant fleet and for an autonomous harbour bus, the Greenhopper, primarily through high-fidelity simulations and hardware-in-the-loop testing using the proposed autonomy stack.
The research findings have been disseminated through publication or submission to international journals and also presented at international conferences. These scientific articles are part of the thesis.