With the increasing interest in the maritime industry in unmanned systems and maritime robotics, attention is shifting towards extremely safe automated collision avoidance systems (ACAS). An ACAS combines the sensors for situational awareness with specific algorithms that prevents a vessel from hitting other objects, usually other nearby vessels. The current state of technology is such that this is one of the big challenges for unmanned systems, as most ACAS are not sufficiently reliable to replace a human helmsman, especially in busy shipping lanes. In this paper we present an approach to ACAS which uses swarming algorithms to prevent vessels of colliding. Swarming algorithms have the characteristic that in itself simple – and therefore robust- rules for the individual participants in the swarm, can result in complex and intelligent behavior of the swarm. The hypothesis that was tested is that an individual ACAS will perform relatively poor in comparison to a solution where all the vessels within a certain range are made subject to a swarming strategy. In the simulations, this hypothesis was tested, not only for collision avoidance, but also with efficient routing and minimal mutual hindrance. The solution we propose is based on a layered approach, where steering actions are abstracted to a number of evasive strategies, on which machine learning can be applied. These strategies correspond to the rules that define the entire swarm. As a special measure of robustness, the swarm should be able to cater for nodes that do not conform to these rules, for instance floating objects or vessels that are not equipped with ACAS.
A typical characteristic of a swarm algorithms is that all the participants in the swarm are equal. As vessels can be mutually different, we propose an additional layer of complexity where the models that the vessels make of the other participants are based on theories from evolutionary and social theories. The solution has been developed for the ship simulator of RH Marine on RDM Campus, and could be the onset for automated support for the helmsman, but with the eventual goal to develop ACAS for unmanned vessels. We have run extensive simulations with vessels in a shipping lane, where individual ACAS strategies were compared with a swarming solution.