Autonomous Collision Avoidance at Sea: A Survey

Abstract

In this survey, results from an investigation on collision avoidance and path planning methods developed in recent research are provided. In particular, existing methods based on Artificial Intelligence, data-driven methods based on Machine Learning, and other Data Science approaches are investigated to provide a comprehensive overview of maritime collision avoidance techniques applicable to Maritime Autonomous Surface Ships. Relevant aspects of those methods and approaches are summarized and put into suitable perspectives. As autonomous systems are expected to operate alongside or in place of conventionally manned vessels, they must comply with the COLREGs for robust decision-support/-making. Thus, the survey specifically covers how COLREGs are addressed by the investigated methods and approaches. A conclusion regarding their utilization in industrial implementations is drawn.

On the implementation of a rule-based system to perform assessment of COLREGs onboard maritime autonomous surface ships

“…Implementing the rules formulated in COLREGs can be achieved by several approaches. [7] gives an overview about a variety of actual approaches including which parts of COLREGs are covered, even though not detailing the criteria for the traffic analyses in detail. Most approaches are concentrating solely on rule 13 to 15 during the traffic assessment, which is often based on the Closest Point of Approach and derived metrics.…”

COLREGs-based Situation Awareness for Marine Vessels - a Discrete Event Systems Approach

“…where f β is the bearing density function. From ( 21) and (15) it is possible to calculate the probability that vessel j needs to give way, i.e. as the sum of probabilities of all the elements of G ⊂ S 2 G = {(HO, HO), (HO, OT ), (HO, P S), (SB, HO), (SB, SB), (SB, OT ), (SB, OT ), (OT, OT ), (P S, OT ), (P S, P S)}, (27) where G contains all the elements of S 2 where the second element q in the tuple (p, q) is q = 1.…”

“…The use of COLREGs was researched extensively in the field of Collision Avoidance (COLAV), where [15] employed several methods. Further, Reinforcement Learning (RL) algorithms were investigated to learn a COLREGs-compliant control policy for COLAV, combining simulations of specific scenarios [16], random ones [17], and randomly generated from historical AIS data [18].…”

“…Similarly, the output of U 2 , U 9 are determined by the violation of TCPA threshold values (t aware , t act ). Lastly, the output of U 4 depends on the relative bearing mapping given in (15).…”

Collision avoidance for autonomous ship using deep reinforcement learning and prior-knowledge-based approximate representation

“…To solve these problems, many scholars have proposed soft computing methods, such as genetic algorithm (Tsou et al, 2010), velocity Obstacle (Wang et al, 2020), fuzzy logic (Fiskin et al, 2021), geometric calculation (Ding et al, 2021), and model predictive control (Yuan and Gao, 2022). However, these soft computing methods have exposed their limitations in the MASS collision avoidance applications, among which is the difficulty of tackling new collision avoidance risks due to lack of scene adaptability after a MASS attempt to avoids multiple ships successively (Burmeister and Constapel, 2021).…”