Design of a guidance system for waterborne AGVs in accordance with navigation rules (COLREGS)

Design of a guidance

sys-tem for waterborne AGVs in

accordance with navigation

rules (COLREGS)

Ningjing Zhou

D

ESIGN OF A GUIDANCE SYSTEM FOR WATERBORNE

AGV

S IN

ACCORDANCE WITH NAVIGATION RULES

(COLREGS)

by

Ningjing Zhou

4253264

Master of Science

in Transport, Infrastructure and Logistics

Thesis committee: Prof. dr. ir. Gabriel Lodewijks Dr. R. R. Negenborn

Dr. ir. Winnie Daamen

Huarong Zheng (PhD candidate)

P

REFACE

This research is the final proof of competence for obtaining the Master of Science (MSc) degree in Civil Engi-neering, with a specialization in Transport, Infrastructure and Logistics (TIL) from University of Technology Delft, The Netherlands.

The purpose of this research is to build a dynamic guidance model of waterborne AGVs in the Port of Rotterdam area according to applicable navigation rules (primarily COLREGS), and compare the result of this model with the data of real vessel paths in practice. To achieve this general objective, the research has carefully reviewed the main navigation rules, determined the collision avoidance strategies, constructed a combined path planner and tested its effectiveness during a series of simulation experiments. The simulated results have been compared with AIS data of real marine vessels to examine the similarities and differences. The proposed model proves its advantages on ensuring the optimality of the final path, while there still re-mains much space for future improvements.

Here, I would like to take this opportunity to express my gratitude to Prof. dr. ir. Gabriel Lodewijks, Dr. R. R. Negenborn and Dr. ir. Winnie Daamen for their time, patience and guidance. I really benefit a lot from their advices to successfully proceed my thesis work. In addition, I would like to thank Huarong Zheng, my PhD supervisor, for her kind reminders on every detail of my work and her generous help when I was faced with bottlenecks. A final thanks is contributed to my dear family and my best friend Spoon who have given me strong spiritual support.

Ningjing Zhou Delft, December 2014

C

ONTENTS

List of Figures vii

List of Tables ix

1 Introduction 1

1.1 Research background. . . 1

1.2 Literature review . . . 2

1.2.1 Research on path-planning algorithms . . . 3

1.2.2 Research on navigation of marine vessels . . . 4

1.2.3 Summary of the literature review . . . 7

1.3 Research focus and contribution . . . 9

1.4 Research objective . . . 10

1.5 Research questions . . . 10

1.6 Research approaches . . . 11

1.6.1 Theoretical framework. . . 11

1.6.2 Methods and data . . . 12

1.7 Scope of the research . . . 13

1.8 Thesis outline. . . 13

2 Navigation rules and regulations 15 2.1 Applicable rules at the Port of Rotterdam. . . 15

2.1.1 International rules. . . 15

2.1.2 National rules . . . 15

2.1.3 Port rules . . . 16

2.2 COLREGS as the ’Rules of the Road’. . . 16

2.2.1 Action to avoid collision . . . 17

2.2.2 Rules in narrow channels . . . 18

2.2.3 Overtaking. . . 19

2.2.4 Head-on situation . . . 19

2.2.5 Crossing situation . . . 20

2.2.6 Action by give-way vessel . . . 20

2.2.7 Action by stand-on vessel . . . 21

2.3 The BPR as Inland Navigation Police Regulations. . . 21

2.4 Summary of navigation rules at the Port of Rotterdam . . . 22

3 Collision avoidance strategies 23 3.1 Ship characteristics. . . 23

3.2 Collision risk evaluation . . . 24

3.3 Rule selection. . . 25

3.4 Summary of collision avoidance strategies . . . 27

4 Path planning algorithm 31 4.1 Main idea of the path planning algorithm. . . 31

4.2 Global planning. . . 33

4.2.1 The A* algorithm. . . 33

4.2.2 Geometry used in the algorithm . . . 36

4.2.3 Adding turning constraints of a vessel . . . 37

4.2.4 Pseudo code of local A* . . . 39

4.2.5 Summary of global planning. . . 40

4.3 Local planning . . . 41

4.3.1 Considering safe passing distance. . . 41

4.3.2 Adding turning constraints. . . 42

4.3.3 Incorporating rule-based requirements . . . 42

4.3.4 Pseudo code of local A* . . . 42

4.3.5 Summary of local planning . . . 43

4.4 Summary. . . 43

5 Simulation results 45 5.1 Map initialization and assumptions. . . 45

5.2 Simulation with navigation rules . . . 45

5.2.1 Additional forbidden zone in the map setup. . . 46

5.2.2 Global planning . . . 46

5.2.3 Local planning. . . 49

5.3 Simulation without navigation rules . . . 59

5.3.1 Map setup . . . 60

5.3.2 Global planning . . . 60

5.3.3 Local planning. . . 61

5.4 Summary of the simulation results . . . 68

6 Case study: comparison with real data in a head-on situation 69 6.1 Selected area . . . 69

6.2 Selected vessels. . . 70

6.3 Map configuration . . . 71

6.3.1 Grid size . . . 71

6.3.2 Dimensions of the map . . . 71

6.3.3 Closed zones and open zones . . . 71

6.4 Global planning. . . 72

6.4.1 Comparison between real vessel track and simulation result with rules . . . 73

6.4.2 Comparison between real vessel track and simulation result without rules . . . 74

6.4.3 Interim conclusion. . . 75

6.5 Local planning . . . 75

6.5.1 Assumptions. . . 75

6.5.2 Comparison between real vessel track and simulation result with rules . . . 78

6.5.3 Comparison between real vessel track and simulation result with rules . . . 81

6.5.4 Interim conclusion. . . 85

6.6 Summary of case study . . . 85

7 Conclusions and future works 87 7.1 Conclusions. . . 87

7.1.1 Conclusion on the development of the guidance model . . . 87

7.1.2 Conclusion on the simulation of different encounter scenarios . . . 88

7.1.3 Conclusion on the comparison between simulation results and real dataset. . . 88

7.2 Future works . . . 89

7.2.1 Considering speed changes of the own ship . . . 89

7.2.2 Considering obstacles with more complex behaviours. . . 89

7.2.3 Discussion on the determination of dmi n . . . 89

7.2.4 Case study on other encountering situations. . . 90

7.2.5 Path smoothing techniques . . . 90

7.2.6 Influences of environment factors. . . 90 A Pictures of SPRANKY and ANDA 91 B AIS data of SPRANKY and ANDA 93

L

IST OF

F

IGURES

1.1 Overview of the research approach. . . 11

1.2 Research approach of model establishment and validation . . . 12

2.1 Illustration of different encounter situations according to COLREGS. . . 17

3.1 Illustration of ship model . . . 24

3.2 Illustration of the CPA approach . . . 25

3.3 Division of three typical encounter situations . . . 26

3.4 Flow chart of collision avoidance strategies based on navigation rules . . . 29

4.1 The combined path planning approach . . . 32

4.2 Example of cost determination of the A* algorithm . . . 34

4.3 Example of deadlock prevention of the A* algorithm . . . 36

4.4 2-geometry, 4-geometry and 8-geometry neighbourhood . . . 37

4.5 The turning rate constraint of a vessel . . . 38

4.6 Expansion in a 4-geometry (a) and an 8-geometry (b) neighbourhood under the turning con-straints . . . 39

4.7 Working mechanism of the global planner . . . 40

4.8 The turning rate constraint of a vessel . . . 42

4.9 Working mechanism of the local planner . . . 44

5.1 Map setups and the optimal paths when the ship travels in different directions . . . 47

5.2 Global paths when applying the 4-geometry (blue) and the 8-geometry (green) neighborhood . 48 5.3 Head-on Situation . . . 50

5.4 Head-on scenario: (a)dC P Avalue, (b)distance between the two ships, and (c)heading of the own ship at each time step . . . 51

5.5 Crossing situation. . . 53

5.6 Crossing scenario: (a)dC P Avalue, (b)distance between the two ships, and (c)heading of the own ship at each time step . . . 54

5.7 Overtaking Situation . . . 56

5.8 Overtaking scenario: (a)dC P Avalue, (b)distance between the two ships, and (c)heading of the own ship at each time step . . . 57

5.9 Multiple encounter scenario including the Head-on situation and the Crossing situation . . . . 58

5.10 Multiple encounter scenario: (a)dC P A value, (b) distance between the own ship and the two target ships, and (c) heading of the own ship at each time step . . . 59

5.11 Global path when (not ) incorporating navigation rules . . . 60

5.12 Global paths in opposite directions when no rules apply . . . 62

5.13 Head on situation with and without navigation rules applied . . . 63

5.14 Crossing situation with and without navigation rules applied . . . 64

5.15 Overtaking situation with and without navigation rules applied . . . 66

5.16 Multiple encounter situation with and without navigation rules applied . . . 67

6.1 Map of Maasvlakte and chosen reseach area (Sour ce : Goog l emap) . . . 70

6.2 Interpreting the map in to grids . . . 72

6.3 Global planning: real vessel track and simulated track with rules . . . 73

6.4 Global planning:real vessel track and simulated track with rules . . . 74

6.5 chronological positions of SPRANKER and ANDA . . . 76

6.6 Local planning of SPRANKER when rules are applied . . . 78

6.7 Local planning of SPRANKY with rules: (a)dC P Avalue, (b)distance between the two ships, and

(c)heading of the own ship . . . 79

6.8 Local planning:real vessel track and simulated track with rules . . . 80

6.9 Local planning of SPRANKER when rules are not applied . . . 82

6.10 Local planning of SPRANKY without rules: (a)dC P Avalue, (b)distance between the two ships, and (c)heading of the own ship . . . 83

6.11 Local planning:real vessel track and simulated track without rules . . . 84

A.1 Picture of SPRANKY . . . 91

L

IST OF

T

ABLES

1.1 Summary of the applied path planning algorithms in COLREGS-based ship navigation . . . 8

5.1 Computation time and path cost of the global planning . . . 48

5.2 Initial states of both ships in the Head-on situation . . . 49

5.3 Initial states of both ships in the Crossing situation . . . 52

5.4 Initial states of both ships in the Overtaking situation. . . 55

5.5 Initial states of both ships in the multiple encounter situation . . . 57

5.6 Summary of path costs in situations with and without rules incorporated. . . 65

6.1 Information of SPRANKY and ANDA (Sour ce : M ar i net r a f f i c.com) . . . 70

6.2 Positions of ANDA and SPRANKY at each time step in the grid map . . . 77

6.3 Local planning:real vessel track and simulated track with rules . . . 80

B.1 AIS data of SPRANKY (Sour ce : M ar i net r a f f i c.com) . . . 93

B.2 AIS data of ANDA (Sour ce : M ar i net r a f f i c.com). . . 93

1

I

NTRODUCTION

The introduction part will give an overview of the research background, literature review, research problems and scope. First, the research background of the topic area is introduced, followed by a literature review dis-cussing the existing research achievements and remaining questions. Research opportunities are identified based on the literature review. Accordingly, the research objectives and questions are defined. The research approaches are proposed to answer the questions in order to realise the objectives.

1.1.

R

ESEARCH BACKGROUND

The booming of international trade is resulting in rapidly increasing volumes of maritime transport flows between countries and continents. As the busiest port of import and export in Europe, the Port of Rotterdam is visited by tens of thousands different types of ships annually. Large volumes of goods are therefore waiting to be transported to the inland area of the continent from the seaside. In 2012, there were 32,057 arrivals of sea-going vessels reaching the port, and the total throughput of the port was 441.5 million metric ton [1]. Now, these numbers are growing. According to the port of Rotterdam authority, they forecast the throughput in 2030 to reach 650 million metric tons in the scenario of moderate economic growth, and 750 million metric tons if there is further globalization [2].Apart from the growing transport volume from the waterside, the port itself is expanding with the ongoing construction of Maasvlakte 2. As a consequence, an increase of the demand of transport between terminals within the Port of Rotterdam is also expected. The exchange of goods between different terminals at the same port is defined as inter terminal transportation (ITT) [3]. If all the ITT activities are assigned to trucks and trains, it may lead to the concern of higher density and possible congestion of traffic in the coming years. Even though road transportation still dominates over other transport means in the ITT system at the Port of Rotterdam, road infrastructure in and to the port are most likely reaching their capacity limits [4]. On the other hand, the port is also faced with different restrictions regarding capacities and operational issues in railway services in the hinterland, and investments in a long-term process are needed [5]. Whilst road and rail transport on land encounter so many constraints, the idea

of transferring part of the current on-land ITT demand to the waterside is considered as a way to address the issue. Considering the well-developed waterway network at the Port of Rotterdam, the Port Authority is now firmly focusing on increasing the share on inland activities carried by inland waterways.

The concept of waterborne AGVs (Automatic Guided Vessels) with advanced technologies is proposed as a possible alternative to spare the burden of the landside traffic, aiming to improve the efficiency of port operations and constitute a complete automated transportation cycle at the Port of Rotterdam. Waterborne AGVs are unmanned surface vehicles which can follow pre-scheduled path with the capabilities to avoid col-lision and follow required navigation rules, such as International Regulations for Avoiding Colcol-lisions at Sea (COLREGS) defined by the International Maritime Organization (IMO) and local regulations. They are used to transport goods between terminals in relatively short distance within the area of the Port of Rotterdam. With automated technologies, waterborne AGVs have the benefits of low operating costs, higher controllabil-ity, reduced human labours and errors, increased safety, and environmentally clean operations. Such a new transport approach is expected to contribute to more efficient terminal operations and distribution of goods in the port area.

As automatic transport vessels, waterborne AGVs are supposed to have the abilities of self-guidance, au-tomatically interacting with other moving or static objects, and more importantly avoiding collision so as to ensure safety and security. To realize the automation of the waterborne AGV at the Port of Rotterdam, the pre-liminary step is to find the control mechanism of its behaviours in water under the constraints that should be followed at its operating area. One of the most important behaviours of it is its way to generate a path between its origin and destination when interacting with its environment and other waterway traffic along the way. When generating such a path, the vessel is also obligated to obey the rules stated in COLREGS and local regulations so as to avoid collision and ensure safety during the trip. These tasks should be fulfilled by the guidance system of waterborne AGVs.

This research is focusing on developing such a dynamic model for the guidance system of waterborne AGVs in accordance with the applicable navigation rules. Combined functions of both offline path planning and online collision avoidance are to be realized to complete this guidance system. To assess the performance of the model, its simulation results are to be compared to the data representing the actual behaviour of a manned vessel operating under the same or similar circumstance. By comparing the model and real data indications, similarities and differences can be found, thus limits and improvements for the model can be analysed. The design of a reliable self-guidance system with the above functions is therefore essential to the development of waterborne AGVs.

1.2.

L

ITERATURE REVIEW

Nowadays, much research has been conducted regarding path generation algorithms of surface vehicles and marine vessels, or how an unmanned vessel react to avoid collision according to COLREGS. The following paragraphs will give an overall of previous research on relevant topics on path planning and marine vessel navigation.

1.2.LITERATURE REVIEW 3

1.2.1.

R

-

As a major function of all automatic guided vehicles and devices, path-planning methods have always been the research focus in this area. There are many guidance and path planning algorithms that have been devel-oped regarding the navigation of unmanned surface vehicles (USVs) as well as robots.

LINE OFSIGHT

The Line-of-sight (LOS) guidance is the simplest and most popular guidance law in use today, particularly in missile guidance technology. Its idea is that if the vessel converges to a constant LOS heading angle directly between the vessel and target (which may be stationary or moving at a constant heading), then it shall even-tually converge to the target position [6]. The drawback of LOS guidance is potential overshoot caused by reducing the cross track error due to environmental disturbances [7]. Furthermore, the final path it generates is not optimal since cost of the path is not considered during the guidance.

ARTIFICIALPOTENTIALFIELD

Artificial Potential Field (APF) methods were contrived originally by Khatib, taking known obstacles into con-sideration by building a representation of the environment by potential gradients [8]. The advantages of such a method are effective use in real-time with low computational requirements. However, it has several limi-tations under certain conditions. The primary disadvantage is the potential to get trapped in local minima, resulting in this method being used solely for local path planning in literature. Oscillations in narrow passages is also a major deficiency [9]. Furthermore, optimization of the path is again not guaranteed.

EVOLUTIONARYALGORITHMS

Evolutionary Algorithms (EAs) are applied in path planning scenarios to address the multi-objective opti-mization problems. Among all the EAs, Genetic Algorithms (GAs) are the most popular ones used for multi-objective optimization. In 2007, Kanakakis and Tsourveloudis proposed GAs for generating waypoints to plan a path [10]. Such a waypoint path planning technique used splines to provide polynomial fitting, applied as smoothing functions. The objective of the GA is to evolve a population of possible solutions from which the individual (solution) with optimal characteristics will be selected [11]. Despite their advantages in solving multi-objective problems, EAs have the pitfalls of the potential of getting trapped in local minima, finding at best a near optimal solution (as the global optimum is never guaranteed) or even failing to find a solution at all in some instances. In addition, these methods are generally restricted to offline implementation [6].

HEURISTIC SEARCH ALGORITHMS

Heuristic search algorithms are a group of grid-based path-planning algorithms which use grid-searching techniques with associated heuristic cost functions. The most commonly-used algorithms from this fam-ily are the A* Search Algorithm, D* Search Algorithm, and Heuristic Anytime Algorithms. The A* algorithm performs a Best-first search to find the least-cost paths from the initial point leading to the goal [12]. It is useful to expand only the fewest number of nodes to quickly find an optimal solution. But it requires large computational memory, and is difficult to deal with unknown environmental dynamics [6]. To reduce the computational cost of A*, Yang et al. developed a modified version of A* called Direction Priority Sequential Selection (DPSS). DPSS consists of two functions of Node Sorting Search (NSS) and Path Fin Cutting (PFC),

using a direction priority sorting technique to search for a path with minimum steps and then optimize the raw path. Results show that DPSS manages to reduce approximately 50% of the search time than A* and pro-duces smoother paths with less jagged segments [13]. However, both the A* and DPSS algorithms are limited to offline path planning, which means they are not able to deal with moving obstacles without further modifi-cation. In order to handle real-time dynamics, the D* algorithm is contrived and known as the dynamic A*. It plans optimal traverses in real-time by incrementally repairing paths to the robot’s state as new information is discovered [14]. However, the large amount of renewing information due to system dynamics results in long computation time, which is not efficient in real implementation. Other derivative algorithms such as Fussed D and D* Lite were proposed which re-use previously known cost information of unchanged nodes, so as to simplify the computation process [15]. When the time available to solve a search problem is limited or un-certain, heuristic anytime algorithms should be applied which allows a flexible trade-off between search time and solution quality [16]. They are developed to achieve solutions which are as optimal as possible within the time frame given as time progresses [17].

RANDOMIZED APPROACHES

Randomized approaches were proposed and applied to the general problem of path planning in a high-dimensional configuration space. This group of approaches include the randomized potential field algorithm [18], the probabilistic roadmap algorithm (PRM) [19] and the Rapidly-Exploring Random Tree (RRT) algo-rithm [20]. The first two suffered from their inadequate efficiency to solve non-holonomic planning problems with kinodynamic constraints. The RRT algorithm was therefore proposed for obstacle-based path planning and is now most popularly used in robot path and motion planning. RRTs have their unique advantages to handle non-holonomic constraints (including dynamics), especially in high degrees of freedom. Meanwhile, they also shows higher efficiency in holonomic path planning than a basic PRM algorithm. An RRT is iter-atively expanded by applying control inputs that drive the system slightly toward randomly-selected points, as opposed to requiring point-to-point convergence, as in the probabilistic roadmap approach [20]. This algorithm is able to predict distributions over future states given a candidate nominal trajectory and gen-erate an optimal path under dynamics. General RRT can easily and quickly gengen-erate a collision-free path in known environment, both in 2D or 3D dimensions. In 2008, Fulgenzi combined RRT algorithm with Gaussian processes to carry out probabilistic navigation in dynamic environment. Simulation results show the perfor-mance of the navigation algorithm for a car-like robot moving among dynamic obstacles with probabilistic trajectory prediction [21]. The path that RRT generates is feasible but probably not the optimal. To achieve an optimal or near optimal solution, RRT* algorithm and many of its extensions were gradually developed, e.g. RRT*-Smart which accelerates the rate of convergence and reduces search time [22], and CC-RRT* which can generate the optimal path under uncertain dynamics [23]. The family of RRT algorithms are most suitable in quick searching of a feasible path in a high-dimensional spaces, but can also be used in a variety of path planning applications.

1.2.2.

R

Marine autonomous navigation is one of the research focuses of general navigation autonomy. Many re-searchers have dedicated to the development of autonomous navigation for marine vessels, with the main focus on collision avoidance. Many existing path planning algorithms and methods were used to search for a

1.2.LITERATURE REVIEW 5

collision-free path which can avoid both static and dynamic obstacles.

On the early stage, researchers only dedicated to design a guidance system to avoid collision regardless of navigation rules. In 2000, a new version of the Evolutionary Planner/Navigator (EP/N) namedϑEP/N++ was introduced as a decision support system for selection of ship trajectory in collision situations [24]. As an EA,ϑEP/N++ supposed to choose the optimal path with minimal total cost which included safety and economy factors. Both online and offline planning were considered in the integrated system. The method succeeded in generating an optimal path within an average computational time of 55 seconds. However, navigation rules such as COLREGS were not considered. In later research, researchers gradually put their focus on the marine navigation under international navigation rules, in most cases, incorporating COLREGS rules into path planning methods to realize collision avoidance during ship encounters. Among all the rules in COLREGS, the Steering and Sailing Rules with particular respect to the conduct of vessels in sight of another are most important to implement full autonomy. Three typical encounter scenarios are specified, including ’Overtaking’, ’Head-on Situation’, and ’Crossing Situation’. Two other rules regarding ’Action by Give-way Vessel’ and ’Action by Stand-on Vessel’ are also combined in the above situations to indicate appropriate operations of each vessel [25]. Based on these rules, researchers dedicated to find the way and algorithm that allows a USV to generate its maneuvers accordingly.

APPLICATION OFLOS

The LOS technique was commonly used to guide a vessel to its destination and desired waypoints along the path. A biasing scheme was introduced to be coupled with LOS technique so that the vessel could generate a COLREGS-compliant path [7]. The idea of the method was that when the distance between the own ship and the obstacle was too short to ensure safety, the guidance system would add a bias angle to the ship’s heading to avoid the possible collision. The bias angle applied in the simulation was given as -75 to realize starboard manoeuvers. The value of the bias angle was based on assumption and practical experience, which was not proven to be an optimal strategy under different circumstances. The simulation results showed that the bias-LOS method was effective in avoiding collisions in static and dynamic environments, but the final path it generated consisted of several obvious detours thus was not most economic. Besides, at some waypoints, the heading of the ship changed dramatically due to collision avoidance behaviors, which is not desired in real operation.

APPLICATION OFAPF

Modifications have also been done to APF approaches to incorporate COLREGS rules. A Modified Virtual Field Forces (MVFFs) method was developed for autonomous ship navigation by combining a fuzzy logic with the concept of potential field method [26]. However, it contains 210 fussy rules and only considers two-ship encounters. A more dedicated research on dynamic marine vessel navigation using APF was later con-ducted, which considered multiple-ship encounters in accordance with COLREGS in a congested water area [27]. The simulation results proved the program to be effective to deal with complex encounter situations. Nevertheless, the optimality of the path was still not taken into consideration, as well as velocity change of all the vessels.

APPLICATION OFEAS

The navigation of marine vessels is also regarded as a multi-objective optimization problem where security, economic and smoothness factors are considered. The goal is to minimize the sailing distance and change of course, while pursuing maximum safety. According, EAs and GAs are used to solve such a non-linear op-timization problem. Ant Colony Opop-timization approach was used to support the decision-making system of COLREGS-compliant collision avoidance in ship navigation [28]. This method showed its effectiveness in simulations, but its computational time was up to 1 minute for local planning and several minutes for global planning. The NSGA-II algorithm is another EA worth mentioning, which uses non-dominated sort-ing genetic algorithm to search for the optimal strategy based on Pareto solutions [29]. It was applied in the navigation of marine vessels in the present of potential collisions, and COLREGS rules were incorporated at the meantime [30]. Simulation results exhibited the validity of the method, with the assumption that the velocity of the vessel was constant and no wind and current were considered. However, in real practice, the sea condition can be more complex. The author also did not give detailed discussion on how the own ship identified the encounter situations it was faced with. Furthermore, the population size of NSGA-II is very important, which can directly affect the computational time of the process and the optimality of the final solution. A population size of 30 was chosen in the simulation. With other sizes, the simulation result, and the effectiveness and efficiency of this method remained to be examined.

APPLICATION OF HEURISTIC SEARCH ALGORITHMS

Among all the heuristic search algorithms, the A* algorithm is the most popular one that have been applied to search for an optimal collision-free ship trajectory. It has been chosen by many researchers as the basic path planning algorithm in ship navigation [31] [32]. Due to its own limitation, it is usually used for offline path planning in known environment, while combining with other methods which can deal with dynamic obstacles, e.g. the Velocity Obstacle (VO) method [33]. As a better version of A* with reduced processing time, the DPSS algorithm was also applied to generate a set of offline waypoints, while the LOS technique was used for online guidance towards these waypoints [7]. In conclusion, the basic A* algorithms are able to generate a collision-free path and ensure its optimality when the information of static obstacles is known. But they have high requirements of memory space and computational time. So they are more suitable for offline path planning or global planning. Local dynamic planners are needed in order to avoid moving obstacles.

To simultaneously integrate dynamic path planning and COLREGS rules, many modifications have been made to standard A*. Fast grid based collision avoidance (FGBCA) algorithm is one of the dynamic versions based on A* [34]. This method added a T-neighborhood area in terms of the ship’s physical constraints when conducting the A* search algorithm. Running time of this improved algorithm was highly reduced, with an average of 0.1 second for local re-planning. Though COLREGS-compliant and fast, the final path it generated presented rapid turns which are not desired in ship navigation. Additionally, situations with velocity changes were not discussed. Another new extension of standard A* named Rule-based Repairing A* (R-RA*) was con-trived in order to deal with dynamic obstacles and generate COLREGS-based collision avoidance strategies [35]. According to the author, R-RA* showed its computational efficiency in simulations of the three classic COLREGS encounter situations (Head-on, Crossing, and Overtaking), with a subsequent preplanning time of 0.2 seconds while a single A* iteration required more than 2 seconds. The short processing time allowed the R-RA* algorithm to be applicable and feasible in practical real-time operation. However, the simulation

1.2.LITERATURE REVIEW 7

again assumed the velocity of the own ship to be constant, and the motion of the target ship was simple and predictable, e.g. in straight lines with no speed change. Extensions of this method are needed to include more complexity of ship behaviours and encounters.

APPLICATION OF RANDOMIZED APPROACHES

Randomized approaches are mostly applied in robotic motion planning in high-dimensional space, and not much research on USV navigation has used this group of algorithms. However, Øivind Loe developed a COLREGS-compliant collision avoidance system for USVs that is able to avoid collisions with both static and dynamic objects using a combine A* and RRT algorithm [36]. The author used A to ensure the optimal-ity of the path RRT generated. Since both of the two basic algorithms are limited to offline application, a Dynamic Window approach was also included to avoid dynamic obstacles. In order to combine these three approaches into one integrated system, a hybrid structure was established with three layers of operation, which were global planning, local planning, and low-level control. According to the simulation results, the overall system was effective to guide the ship in accordance with navigation rules. But it also suffered from long computational time due to the complexity of the control architecture. The processing time to generate a collision-free path was approximately 400 seconds, which was not practical in real implementation.

OTHER APPROACHES

Except for all the above mentioned algorithms, some researchers also successfully constructed COLREGS-compliant collision avoidance systems for marine vessels with other types of approaches. The VO method was one typical example. Yoshiaki et al. presented a COLREGS navigation algorithm with the application of traditional VO method, and specified pre-collision check using closest point of approach (CPA) calculations and scenario-based rule selections [37]. The application was limited to local path planning, with the com-bination of global planning needed. Meanwhile, the VO method only took velocity of the ship into account, and more features are expected to be considered to ensure feasible maneuvers and operations. In addition to the EAs, another multi-objective optimization approach named Interval Programming (IvP) was also tested by Benjamin et al. in USV sea trials in several COLREGS encounter scenarios with two SCOUT vessels ([38]. The approach was proven to be useful in terms of the experiment results, however, the robustness remained to be checked and only local planning was discussed.

1.2.3.

S

According to the above discussion, it can be concluded that different types of algorithms have been explored to realize COLREGS-compliant marine vessel navigation, and many of them were proven to be effective in simulations. The algorithms that have been used to date and reviewed are listed in Table1.1, together with their advantages and disadvantages.

Simple existing approaches such as LOS and APF are effective in dynamic path planning with low com-putational cost. But they do not consider the cost of the path, so that the optimality of the final path they generate are not guaranteed. The same problem goes to the RRT algorithm, and its modified version of RRT* requires infinite time to reach an optimal solution. Standard algorithms in the groups of EAs and heuristic search algorithms (e.g. A*) are able to generate an optimal path. But they have their problems on dynamic path planning due to long computational time and high requirements on memory space. These drawbacks

Algorithms

Applications in COLREGS-based

ship navigation

Advantages Disadvantages

Line of Sight Bias-LOS 1. Simple and effective 2. Low computational time

1. Lack of flexibility with bias angle fixed

2. Optimality not guaranteed 3. Undesired detours

4. Pre-known waypoints needed

Artificial Potential Field MVFFs; APF modified by Xue et al. (2011) 1. Dynamic

2. Low computational time

1. Local minima

2. Oscillations in narrow passages 3. Optimality not guaranteed 4. Velocity change not considered

Evolutionary Algorithms Ant Colony Optimization; NSGA-II 1. Multi-objective optimization 2. Optimality guaranteed

1. Long computational time 2. Limited efficiency on dynamic planning

3. Velocity change not considered Heuristic Search

Algorithms

DPSS (combined with bias-LOS); FGBCA; R-RA*

1. Fast and effective 2. Low computational time 3. Dynamic

1. Lack of smoothness

2. Velocity change not considered

Randomized approaches RRT (combined with A* and Dynamic Window) 1. Hybrid structure of motion control 2. Optimality guaranteed 3. Dynamic

1. Optimality not guaranteed if not combined with other methods 2. Long computational time 3. Velocity change not considered

Others VO; 1. Dynamic 2. Easy to incorporate COLREGS rules 3. Velocity change considered

1. Limited to local planning 2. Pre-known waypoints needed

IvP

1. Multi-objective optimization 2. Dynamic

1. Limited to local planning 2. Robustness examination needed

Table 1.1: Summary of the applied path planning algorithms in COLREGS-based ship navigation

of them lead to the concern of their efficiency when they are used in situations involving complex dynamics. However, several researchers endeavoured to improve their speed of processing, and managed to develop fast and dynamic versions. The authors claim that the prospect of these new algorithms are promising for further modifications in order to be effective when there are more complex dynamics as well as environmen-tal uncertainties. One of them worth mentioning is the R-RA* algorithm, while the other is FGBCA, whose computation is even less then 0,1 s. Both of these two methods have high potential to be extended to deal with more complex situations of vessel encounters.

Even though all the discussed algorithms performed well in terms of simulation results, their efficiency in practice of real ship navigation remains to be discussed. Such concerns are mainly derived from two aspect of limitations in today’s simulation:

1. the over-simplification of the model;

2. the insufficient consideration of environmental factors.

The over-simplification of the simulation model appears in different aspects. First of all, the dynamics now modelled in current guidance models are mostly ideal dynamics, such as non-reactive moving objects with

1.3.RESEARCH FOCUS AND CONTRIBUTION 9

constant velocity and direction whose motion can be easily predicted, and single-ship encounters. Further-more, in most cases, the velocity of the own ship is presumed to be constant and it only alters its course to avoid collision. However, in real implementation, there are surely situations in which simple course alteration cannot succeed to avoid the obstacles and a change in velocity is therefore needed. Such situations have not been discussed in current research. Last but not the least, only single COLREGS encounter scenarios have been modelled respectively, for example, simple overtaking, simple head-on, or simple crossing. Neverthe-less, multiple encounter scenarios might happen at the same time in clustered sea area. Thus, the current model and algorithm need to be improved to be fully compliant with COLREGS, as well as applicable local navigation rules.

As for environmental factors, such as wind, current and weather conditions, no simulation has been con-ducted when designing the COLREGS-based guidance system of a marine vessel. The influence of such fac-tors on ship navigation in accordance with navigation rules are to be explored in future researches

1.3.

R

ESEARCH FOCUS AND CONTRIBUTION

As can be seen from the literature review, there is currently much experience on the path planning and col-lision avoidance of autonomous vessel navigation under COLREGS. However, existing algorithms have their pitfalls and insufficient efficiency. Also, existing models are simply tested on simulation platforms or via scale ship experiments. Few research has done the work to compare the theoretical model with vessel behaviour in real practice. Therefore, the performance of the existing models is not solidly efficient in real implementation. Full autonomous marine navigation with the compliance of navigation rules is still a challenge.

To develop a more effective and efficient guidance system that is able to cope with more complex dynam-ics and in higher compliance with navigation rules, research opportunities on the following aspects can be identified:

• Simulation of more complex encounter situations (e.g. multiple-ship encounters)

• Simulation of obstacles with unpredictable (e.g. obstacles with changing speed and/or direction)

• Simulation of emergencies (e.g. when changing course, or the action of the give-way vessel could not avoid the collision)

• Consideration of environmental factors (e.g. wind, current, weather, etc.)

• Validation of the model using data of real vessel operations

Based on the above research opportunities, this focus of this research lies on:

1. discussion on the collision avoidance strategy in more complex encounter situations

2. comparison of the simulation results with real vessel dataset

Not only the singular encounter scenarios will be discussed, but also multiple encounter situations with more than one ship. Meanwhile, a very large data set of actual vessel movements can be accessed in the area of Port of Rotterdam, it is possible to examine the behaviours of manned vessels and compare them with the

simulation results between one specific pair of origin and destination. Hopefully, this may help to improve the model and realize higher compliance of COLREGS as well as local navigation rules.

Additional contribution of this research includes the improvement of a A*-based algorithm using a higher level of geometry setup while ensuring short computation time. The effect of geometry used in a grid-based algorithm has been discussed according to simulation results. The improved path planning algorithm aims to have higher efficiency in real marine navigation implementation.

Furthermore, the necessity of incorporating navigation rules has also been discussed via a serious of sim-ulations.

1.4.

R

ESEARCH OBJECTIVE

This research is conducted with the objective to build a dynamic guidance model of waterborne AGVs in the Port of Rotterdam area according to applicable navigation rules, and compare the result of this model with the data of real vessel paths in practice.

The guidance model is to realise the behaviour control mechanism of the navigation of waterborne AGVs. The behaviours of the waterborne AGVs include how it should be guided following a pre-designed path given the known environment, and how it should act to avoid collisions according to navigation rules (e.g. COL-REGS and local rules). The model is supposed to have the ability to deal with dynamic ship encounter situa-tions in which obstacles perform complex behaviours, and lead the ship to correspond according to the rules. Also, the path-planning and obstacle avoidance algorithms behind the model should be short and meet the requirement of computation time in real-time operations.

To achieve this general objective, the research is generally divided into two parts, model establishment and comparison between simulation results with real vessel behaviours, with their respective goals.

In the part of model establishment, the research intends to build a model in mathematical way which enables the waterborne AGV to generate a collision-free path, recognizing its real-time status and react in terms of navigation rules to avoid dynamic and static obstacles. After the establishment of the model, simu-lations are conducted to assess the effectiveness and efficiency of the model. The simulation results of model are compared with real vessel behaviours. By comparing the differences and similarities of the simulation results and real data based on selected criteria, opportunities can be identified for future improvements of the proposed model.

1.5.

R

ESEARCH QUESTIONS

Based on the research objective, the main research question is formulated in order to build a dynamic guid-ance model for waterborne AGVs, stated as follows:

What should the guidance model and the algorithm behind it be in order to generate a collision-free path in dynamic situations according to navigation rules at the Port of Rotterdam?

In order to thoroughly answer this question, several sub-questions are proposed step by step. They are listed as follows:

1.6.RESEARCH APPROACHES 11

to represent their features mathematically?

2. What path-planning algorithm is most applicable to the waterborne AGVs operating at the Port of Rot-terdam regarding the environmental features?

3. How to incorporate the path-planning algorithm with navigation rules in a mathematical way?

4. Which types of input and output data are needed to build and verify the model?

5. What aspects and indexes should be chosen and evaluated to verify the model using real data in the case study?

6. How the model can be improved according to the comparison between simulated results and real data and what further information are needed?

1.6.

R

ESEARCH APPROACHES

According to the research objectives and research questions formulated above, the overall research approach can be constructed to answer the questions and achieve the objectives. The overview of the research ap-proach is represented in Figure1.1.

Figure 1.1: Overview of the research approach

1.6.1.

T

As discussed above, the research mainly consists of two parts: model establishment and comparison with real dataset. They are the focus of the research work. The steps to be taken in order to finish these parts of work is illustrated in Figure1.2.

The establishment of the model includes the development of a proper path generation algorithm and the navigation method to avoid collisions according to applicable rules and regulations. The model is finally con-structed combining these two functions. The model is supposed to be concon-structed using MATLAB as a tool. The research focus of the construction part is to develop the mathematical approach to model the behaviour of a vessel when travelling from one place to another with the possibility to encounter potential collisions with other obstacles and maritime traffic. The navigation rules and regulations are the major concerns dur-ing the construction as constraints to regulate the behaviour of the vessel.

When the model is developed, the next step goes to the comparison between the simulation results and real dataset. The focus of the is part is to identify the differences and similarities between the model and real

Figure 1.2: Research approach of model establishment and validation

practice, and thus further improve the model. In this part, an OD pair or several OD pairs are to be selected for a case study in which the model should simulate the behaviour of the waterborne AGV. Criteria are set up based on selected parameters to check how the result of the model differs from the real data. The real data used here is empirical data of actual vessels operating between the selected origin and destination terminals. Based on the gap between the simulation result and the real data, the performance of the model is evaluated. Direction of future improvements are therefore identified to realise more effective and efficient autonomous marine navigation.

In the end, suggestions for improvement of the model, as well as limitations of the research due to current knowledge and research scope will be discussed.

1.6.2.

M

During the phase of model establishment, the challenge should be the combination of path planning method and the behaviour control of collision avoidance under certain scenarios according to navigation rules. To develop such a combined model, two sub-models of path planning and collision avoidance are first to be constructed, and later combined into one. As for path planning, there are now many algorithms available in different aspects of applications. The main task in this research to find a path planning algorithm is to compare and select appropriate algorithm that corresponds to the research context best, and make neces-sary modification based on particular features with respect to the research scope and the requirements of

1.7.SCOPE OF THE RESEARCH 13

within the scope. The way to do this is mostly theoretical work of literature review. Regarding collision avoid-ance, applicable navigation rules, including international and local ones at the Port of Rotterdam, should be reviewed in the first place. According the rules, an decision-making architecture of how the waterborne AGV should react when faced with different situations is to be constructed. In this architecture, different scenar-ios representing different about-to-collide situations are to be categorized. Actions of a waterborne AGV to take in these scenarios are compliant with the regulations declaimed in applicable navigation rules. Statistics of parameters that formulate the behaviours of vessels are necessary in the model establishment part, such as velocity, acceleration, distance a ship should keep to another, etc.. These data can be gathered from the navigation rules, or from empirical statistics in real practice.

Data needed in the comparison part are the simulation results of the model and the real data gathered in practice. The set of real data should include several aspects. First, the information of the selected OD pairs should be collected, including geographic position, environmental context, and terminal configuration, etc. Second, the real data of vessels operating between the terminals are also required. Such data should be able to represent the behaviour and status of a vessel at certain time steps, such as position, velocity, and direction of a vessel. The specific data needed is also related to how the model is formulated and the output of the model which gives information of the simulated vessel behaviours. These two sets of data should correspond to each other in order the complete the validation process. In the end, data of chosen parameters or standards to compare the simulation results and the real practice should be gathered. The values of the evaluation parameters can be derived from the raw data from both the simulation results and real vessel information.

1.7.

S

COPE OF THE RESEARCH

The concept of waterborne AGVs in this research is proposed to deal with growing inter terminal transport demand at the Port of Rotterdam, and to be implemented primarily in the Rotterdam Port area. Therefore, the scope of the research is specifically restricted to the path planning and collision avoidance method of waterborne AGVs that are to be implemented in the area of Port of Rotterdam.

As a result, the guidance system will be designed in the context of the Port of Rotterdam, according to practical geographic features, port regulations and applicable navigation rules in this area (COLREGS and local rules in the Port of Rotterdam area). The algorithms behind the guidance model is consequently aiming to ensure optimal performance of waterborne AGVs operating in this area. In the meantime, the real data to be used for model validation are also from historical statistics of manned vessels sailing between terminals at the Port of Rotterdam.

However, the method used to build and validate the model should be applicable for waterborne AGVs under other environmental circumstances, if parameters and their values are changed correspondingly.

1.8.

T

HESIS OUTLINE

The structure of the thesis is organized as follows. In Chapter 2, the navigation rules in the Port of Rotterdam area that are related to the research are reviewed. In Chapter 3, those rules that are identified to be applica-ble in Chapter 2 are interpreted mathematically in order to be incorporated into the guidance model. The collision avoidance strategies are therefore formulated. Chapter 4 proposes the new combine path planning

algorithm, and demonstrates the constraints to be added to the algorithm. In Chapter 5, a serious of simula-tion results are presented to show the performance of the proposed model when there are or no navigasimula-tion rules applied. After that, Chapter 6 compares the simulation results of the model with the vessel behaviours in term of real AIS data. In the end, conclusions and discussions of future works are presented in Chapter 7.

2

N

AVIGATION RULES AND REGUL ATIONS

This chapter reviews the navigation rules and regulations which are applicable within the Port of Rotterdam area, including international ones, and local ones at the port. Typical vessel encounter protocols which are stated in COLREGS and local rules are also identified and categorized. Specific collision avoidance policies defined in the rules under these encounter circumstances are taken as requirements of vessel’s behaviour when making collision avoidance decisions.

2.1.

A

PPLICABLE RULES AT THE

P

ORT OF

R

OTTERDAM

As a vessel travelling within the area of Port of Rotterdam, a waterborne AGV must obey the rules and regu-lations in order to be lawfully operational on water. The Authority of Port of Rotterdam applies international rules, Dutch national rules, and "house rules" of the port itself to ensure safe, efficient and environmentally responsible handling of shipping traffic [39].

2.1.1.

I

The international rules of the IMO that are applied includes COLREGS as the Rules of Road, and other rules as safety treaties such as the International Convention for the Safety of Life at Sea (SOLAS) and its amendments (e.g. the IMDG code and IBC). Besides, on a lower level, European rules including the recommendations of the European Community, are also in force at the port.

2.1.2.

N

Since the port area also includes inland waterways which connect different terminals within the port, national inland waterway regulations are also applied at the Port of Rotterdam. The most important one is named Inland Navigation Police Regulations, known as Binnenvaartpolitiereglement (BPR) in Dutch. It specifies the traffic policies for Dutch inland waterways.

2.1.3.

P

Furthermore, the Port Bye-laws are the "house rules" of the port. Based on the Rotterdam Port Bye-laws, the Port Rules on dangerous substances contain additional, specific regulations for ships carrying dangerous cargoes in the port.

A list of applicable rules and regulations at the Port of Rotterdam are shown below [39]:

• International Regulations for Preventing Collisions at Sea (COLREGS)

• International Convention for the Safety of Life at Sea (SOLAS, with the IMDG code and IBC)

• Shipping Traffic Act

• Inland Navigation Police Regulations (BPR)

• Shipping Regulations for Territorial Waters (STZ)

• Compulsory Pilotage Decree 1995

• Decree on Pilot Exemption Certificate Holders Shipping Traffic Act

• Regulation for Licensed (Maritime) Pilots

• Regulation for the Prevention of Pollution from Ships

• Regulation on the Transportation of Dangerous Substances, 2007

• Port Bye-laws Rotterdam 2004 – updated for 2007

• Port Bye-laws Schiedam and Vlaardingen

• Regulation for Communication and Pilot requests sea shipping

• Port Security Law (ISPS)

Among all these rules and regulations, two of them are most essential in marine ship navigation: COL-REGS and BPR. Because they clearly specify the collision avoidance actions a vessel should take in front of potential collisions. COLREGS are applicable in international marine traffic practice while BPR are specially regulating the behaviours of ships in Dutch inland waterways.

2.2.

COLREGS

AS THE

’R

ULES OF THE

R

OAD

’

COLREGS, as international navigation guidelines, were designed for humans to obey when operating vessels at sea, and published by IMO in 1972. IMO set out it as the ’Rules of the Road’ to prevent collisions between vessels travelling on water. Apart from collision avoidance policies, COLREGS also cover a lot of detailed regulations during marine traffic operation, such as the responsibilities of vessels and even the use of lights, signals and sound.

Even though drafted to be followed by humans in the first place, COLREGS keep evolving to be applicable in modern marine traffic practice. Up to now, radar has been recognized as an important aid in collision avoidance and the use of it has been included in the regulations (Loe, 2008). More and more autonomous

2.2.COLREGSAS THE’RULES OF THEROAD’ 17

elements are being added into the rules, while COLREGS remain the lawful effect on all types of vessels at sea. Therefore, COLREGS must also be obeyed by automatic guided waterborne vessels so that such vessels can be lawfully operational at sea. On the other hand, a waterborne AGV should behave in term of the rules in order that its manoeuvres can be compatible with the manoeuvres of other vessels in the traffic.

Five parts are included in COLREGS [25]:

1. Part A - General: outlining the applicability and responsibilities of the regulations, and general defini-tions used in the regulation;

2. Part B - Steering and Sailing Rules: including three sections, respectively regarding the conduct of ves-sels in any condition of visibility, in sight of another and in restricted visibility;

3. Part C - Lights and Shapes : stating the guidelines for the use of lighting;

4. Part D - Sound and Light Signals: addressing the issue of various signals;

5. Part E - Exemptions: classifying the situations when exemptions are possible.

All the regulations in COLREGS are applicable in the navigation of waterborne AGVs. Among them, the second section in Part B – Steering and Sailing Rules which refers to the conduct of vessels in sight of another appears to be most important and challenging to realize in autonomous marine navigation.

Three vessel encounter protocols (Overtaking, Head-on and Crossing) are categorized in this section, and actions to be taken by each vessel to prevent collision are demonstrated. A summary of how a vessel should act in the three typical protocols are illustrated in Figure2.1. The rules included in this section are to be incorporated to the path-planning algorithm to search for a collision-free path. The own ship the vessel that is now being operated, and the target ship is the one operated by others and regarded as a moving obstacle. The detailed rules are reviewed and discussed in the following sub-sections.

Figure 2.1: Illustration of different encounter situations according to COLREGS

2.2.1.

A

In the first place, COLREGS the general principles for vessels while taking actions to avoid collision. Rule 8 gives a list of instructions regarding how a vessel should act to prevent collision effectively.

Rule 8(b)

Any alteration of course and/or speed to avoid collision shall, if the circumstances of the case admit be large enough to be readily apparent to another vessel observing visually or by radar; succession of small alterations of course and/or speed shall be avoided.

Rule 8(c)

If there is sufficient sea room, alteration of course alone may be the most effective action to avoid a close-quarters situation provided that it is made in good time, is substantial and does not result in another close-quarters situation.

Rule 8(d)

Action taken to avoid collision with another vessel shall be such as to result in passing at a safe distance. The effectiveness of the action shall be carefully checked until the other vessel is finally past and clear.

Rule 8(e)

If necessary to avoid collision or allow more time to assess the situation, a vessel may slacken her speed or take all way off by stopping or reversing her means of propulsion.

As it can been seen in these rules, the change of course is the prior choice in front of a possible collision. Speed alterations are not preferred in the first place unless the course alteration is not able to prevent the collision to happen. The reason is that the change of course is of higher manoeuvrability than the change of speed of a marine vessel, especially for larger vessels which have greater inertia. This is why in lots of simu-lations the own ship is assumed to move with a constant speed and only alter the course to avoid collision. This rule is also applicable in the simulation of this research.

Besides, according to Rule 8(d), a safe passing distance is required so that the motion of one vessel will not affect the motion of the other. If two vessel are too closed, the suction effect may pull the two ships to collision. Even though not collided, the waves each of them brings will possibly cause fluctuation and danger to another. Thus a safe passing distance is much needed.

2.2.2.

R

Rule 9 in COLREGS specifies the actions to be taken for a vessel when proceeding along a narrow channel or fairway.

Rule 9(a)

A vessel proceeding along the course of a narrow channel or fairway shall keep as near to the outer limit of the channel or fairway which lies on her starboard side as is safe and practicable.

This rule is given to reduce the chance of collision when there is limited navigable space for the marine vessels, i.e. in narrow channels. The term ’narrow channel’ here is not precisely defined, thus it is difficult to decide whether a channel is ’narrow’ or not. However, according to the Courts, passages of approximately 2 miles (about 3218.69 meters) wide are often considered as narrow channels [40]. The Nieuwe Waterweg as the main waterway connecting the North Sea to the hinterlands in the Port of Rotterdam area is currently

2.2.COLREGSAS THE’RULES OF THEROAD’ 19

between 480 and 675 meters [40]. In term of the regulation, it can be considered as a narrow channel, in which vessels should keep to their right in order to give sufficient space to other vessels approaching from the other direction.

2.2.3.

O

Rule 13 defines the actions of each vessel when they are in the overtaking situation.

Rule 13(a)

Notwithstanding anything contained in the Rules of Part B, Sections I and II, any vessel overtaking any other shall keep out of the way of the vessel being overtaken.

Rule 13(b)

A vessel shall be deemed to be overtaking when coming up with another vessel from a direction more than 22.5◦_{abaft her beam, that is, in such a position with reference to the vessel she is}

overtaking, that at night she would be able to see only the stern light of that vessel but neither of her sidelights.

Rule 13b deems overtaking to occur then another vessel is coming up from a direction more than 22.5◦_of

the ship’s beam. The overtaking vessel shall ensure enough clearance and keep out of the way of the overtaken vessel.

COLREGS do not specify which side the overtaking ship should go during overtaking. According to com-mon practice, a turn-to-starboard maneuver should be taken by the overtaking vessel so that it can pass on the right side of the overtaken one, see Figure2.1(a). Such a maneuver is more favorable in case that the over-taking vessel may go into other types of encounters situations in which it most possibly need to turn to its starboard side for collision avoidance. If it is the case, the vessel will have less alteration of speed and course after the overtaking is finished.

2.2.4.

H

-

The head-on situation is defined in Rule 14 in COLREGS.

Rule 14(a)

When two power driven vessels are meeting on reciprocal or nearly reciprocal courses so as to involve risk of collision each shall alter her course to starboard so that each shall pass on the port side of the other.

Rule 14(b)

Such a situation shall be deemed to exist when a vessel sees the other ahead or nearly ahead and by night she could see the masthead lights in line or nearly in line and/or both sidelights and by day she observes the corresponding aspect of the other vessel.

No specific standard is given to define when a head-on situation happens. However, the two vessels are required to turn to starboard and pass on the left side of each other, see Figure2.1(b).

2.2.5.

C

The crossing situation is defined in Rule 15.

Rule 15

When two power driven vessels are crossing so as to involve risk of collision, the vessel which has the other on her own starboard side shall keep out of the way and shall, if the circumstances of the case admit, avoid crossing ahead of the other vessel.

Again, no specific standard is given to define when a crossing situation starts. The rule specifies that the vessel which has the other on her own starboard side shall keep out of the way. This means the vessel which had the other on her port side has the right-of-way and is not required to give immediate collision avoidance actions unless there is an emergency. The illustrations of how two vessels should act in a crossing situation are in Figure2.1(c) and2.1(d).

2.2.6.

A

-

The give-way vessel is the vessel that is required to keep out of the way of the other in one encounter situation, as defined in Rule 16 in COLREGS.

Rule 16

Every vessel which is directed to keep out of the way of another vessel shall, so far as possible, take early and substantial action to keep well clear.

The give-way vessel is required to alter her path while the other vessel which has the right-of-way can remain on her original path. Unless the action by the give-way vessel cannot prevent the collision, the other vessel shall take other actions.

The give-way vessel typically exists in the crossing situation, and is the one on the port side of the other. Despite of that, a power driven vessel shall always give way to one which is not. Rule 18 defines the respon-sibilities between vessels when different types of vessels encounter each other. The responsibility of a power driven is clearly explained. According to this rule, a waterborne AGV as a power driven vessel shall give way to the other which belongs to one of the above vessel types, even in a crossing-from-left situation.

Rule 18(a)

A power driven vessel underway shall keep out of the way of:

• a vessel not under command;

• a vessel restricted in her ability to manoeuvre;

• a vessel engaged in fishing;

• a sailing vessel;

In addition, a small vessel should keep clear of a larger one, since the smaller one is more manoeuvrable and forcing a larger vessel to take evasive manoeuvres is inefficient and dangerous [36]. In order to ensure safety, our waterborne AGV should always give way unless it encounters a power driven vessel of a smaller or equal size.

2.3.THEBPRASINLANDNAVIGATIONPOLICEREGULATIONS 21

2.2.7.

A

-

Action by stand-on vessel is given in Rule 17. A stand-on vessel is the one which has the right-of-way in the encounter of the other. The stand-on vessel is forced to keep her course and speed when an applicable encounter situation is determined. However, the stand-on vessel must also take actions if the collision cannot be avoid by the action of the give-way vessel.

Rule 17(a)

• (i) Where one of two vessels is to keep out of the way of the other shall keep her course and speed.

• (ii) The latter vessel may however take action to avoid collision by her manoeuvre alone, as soon as it becomes apparent to her that the vessel required to keep out of the way is not taking appropriate action in accordance with these Rules.

Rule 17(b)

When, from any cause, the vessel required to keep her course and speed finds herself so close that collision cannot be avoided by the action of the give-way vessel alone, she shall take such action as will best aid to avoid collision.

Rule 17(c)

A power driven vessel which takes action in a crossing situation in accordance with sub-paragraph (a)(ii) of this Rule to avoid collision with another power driven vessel shall, if the circumstances of the case admit, not alter course to port for a vessel on her own port side.

Rule 17(d)

This Rule does not relieve the give-way vessel of her obligation to keep out of the way.

According to Rule 17(a)(i), the stand-on vessel shall keep her course and speed during encounters. This is also the reason why existing research assumed the target ship in a crossing situation to move in constant velocity and direction in simulations. Nevertheless, there are still moving obstacles with unpredictable move-ments which are hostile to the own ship. In such cases, the own ship should be able to take manoeuvres to avoid such obstacles to ensure safety.

2.3.

T

HE

BPR

AS

I

NLAND

N

AVIGATION

P

OLICE

R

EGULATIONS

While COLREGs is taken as a guidebook of collision avoidance at sea, the BPR offer guidelines to prevent collision for vessels sailing in inland waterways in the Netherlands. The BPR was first adopted on October 26, 1983, and updated to its latest version in 2012 [41]. The BPR is applicable to both commercial shipping and recreational vessels which are sailing in the Dutch inland waterways. Since the Port of Rotterdam area includes many inland waterways and barge terminals, the BPR is also to be considered in the navigation of waterborne AGVs which are to carry out ITT activities.

The BPR consists of two parts. Part I include detailed rules regarding general provisions, use of lights signals, use of radio and radar, traffic rules of sailing and berthing. Part II are appendix of illustrations of regulations in Part I.