Delft University of Technology
FACULTY MECHANICAL, MARITIME AND MATERIALS ENGINEERING
Department Marine and Transport Technology Mekelweg 2 2628 CD Delft the Netherlands Phone +31 (0)15-2782889 Fax +31 (0)15-2781397 www.mtt.tudelft.nl
This report consists of 75 pages and 1 appendices. It may only be reproduced literally and as a whole. For commercial purposes only with written authorization of Delft University of Technology. Requests for consult are only taken into consideration under the condition that the applicant denies all legal rights on liabilities concerning the contents of the advice.
Specialization: Transport Engineering and Logistics Report number: 2012.TEL.7728
Title: Proposal for an experimental setup with guided autonomous vessel formations. Author: J.J.M. de Kraker
Title (in Dutch) Voorstel voor een experimentele opstelling met autonome boten formaties.
Assignment: literature Confidential: no
Initiator (university): Dr. Rudy Negenborn Supervisor: Dr. Rudy Negenborn
Delft University of Technology MATERIALS ENGINEERING Department of Marine and Transport Technology Mekelweg 2 2628 CD Delft the Netherlands Phone +31 (0)15-2782889 Fax +31 (0)15-2781397 www.mtt.tudelft.nl
Student: J.J.M. de Kraker Assignment type: Literature Supervisor (TUD): Dr. Rudy Negenborn Creditpoints (EC): 10
Specialization: TEL
Report number: 2012.TEL.7728 Confidential: No
Subject: Towards a new experimental setup for navigation over water
In the Netherlands transport over water is already important now, and will become more important in the future as an increasing amount of containers and bulk materials will have to be transported over waterways. Currently, vessels that transport such materials do not take into account the presence of other vessel. They travel as quickly as possible from one location to another. However, in order to most effectively (safest, fastest, most energy efficient) use the existing infrastructure in the future, individual vessels will have to take into account the presence of the other vessels and adjust their own behavior(i.e., position, heading and speed) based on the behavior of the surrounding vessels. Therefore, currently, control algorithms and cooperation protocols are being developed aimed at achieving this. These algorithms are, however, being tested mostly in simulated and fully controlled computer environments. In order to bring such methods close to practice, an experimental setup has to be developed that enables assessing the performance of such control algorithms in a more realistic setting. Apart from the scientific value of such a setup, it will also have an important function for demonstration purposes.
As a first step in getting to a new experimental setup this literature study aims at investigating what and how existing experimental setups involving water systems and vessels have been developed around the world, and how improvement over these existing setups can be made.
Questions that you will address in this literature study are:
Why have the experimental setups currently existing been realized? What do these existing setups consists of? How have they been designed and implemented? What are the advantages and disadvantages of these setups?
What kind of experimental setups involving water and navigation could be developed in Delft? Why would such setups be interesting to have here? How could such experimental setups be realized? What would be the requirements and costs involved?
Based on your literature survey, it is expected that you conclude with a recommendation for future research opportunities and potential for more ideas and/or applications. The report must be written in English and must comply with the guidelines of the section. Details can be found on the website.
The professor,
Summary
A lot of research nowadays is done about autonomous vessels, because that decreases the workload of the crew of a ship and chances of collision. An addition to the autonomous vessels is the autonomous vessel formation, where multiple vessels move together in formation. This step is a logical continuation of the autonomous control of an individual boat, but still needs the research be started with a small and concise experiment. Therefore an experimental setup is designed to test and show the possibilities of the control of an autonomous vessel formation.
The structure of the currently existing experimental setups, realized for testing individual or formation controllers, use all the same control structure. Namely the boat’s position is determined and is used, together with the data of the user interface, to determine the heading angle to be followed. These are compared, together with the measured heading angle and possible input of the collision avoidance system, in the control system where the control signal is determined, which is sent to the boats.
The characteristics of a boat which can be changed are: the type, size, propulsion system, steering system and the communication between all different systems. All characteristics have different possibilities with corresponding advantages and disadvantage. The environment of the setup can limit the use of some boats and is therefore also analyzed. The controller consists of a control system fed by multiple inputs that are adapted by multiple systems, such the data becomes useful for the control system. The systems that adapt the input are: positioning system, measuring heading angle, filtering, guidance method and a collision avoidance method. For all systems in the controller and characteristics of the boat a possibility has to be chosen in order to create a good experimental setup.
The towing tank inside the faculty of 3mE is perfect to test and improve the controller with only one or a few boats, because the turning radius of a full formation is too large compared to the size of the tank. The pool in front the faculty is larger and a better place to show a larger boat formation, but unhandy situation arise in the tests in the beginning, because having the boats nearby is very useful. Using both places is therefore not avoidable, but has some limitations. So is the GPS signal too low at the tank and therefore positioning using a camera is the only remaining suitable option.
The proposed experimental setup is furthermore designed such that the best options are combined with the previous mentioned limitations. The most suitable option for the boat is a normal radio controlled boat, which is not too small and not special, driven by propellers, a bow thruster and differential thrust. The control system is placed on the boat and communicates with the other systems using a wireless network. The positioning and measurement of the heading is done with a camera and filtered with a Kalman filter. The leader boat is guided by the line of sight method and the other boats follow using the leader – follower formation distance – distance or distance – angle control. The line of sight guidance uses a PID controller and the formation control a sliding mode controller. For all systems the simplest, most precise and the best suitable option is chosen.
Summary (in Dutch)
Veel onderzoek is tegenwoordig gedaan over autonome schepen, omdat dat de werkdruk van de bemanning van een schip en de kans op botsingen verlaagd. Een toevoeging aan autonome schepen zijn autonome schepen formaties, waar meerdere schepen samen in formatie bewegen. Deze stap is aan logisch vervolg op de autonome controle van een individuele boot, maar toch moet er worden begonnen met een klein en compact experiment. Een experimentele opstelling is ontworpen om een autonome schepen formatie te testen en de mogelijkheden te laten zien
De structuur van reeds bestaande opstellingen, gerealiseerd voor het testen van individuele of formatie besturing, gebruiken allemaal dezelfde controle structuur. De positie van de boot is bepaald en wordt samen met de beschrijving van de gebruiker gebruikt om te volgen koers te bepalen. Deze worden vergeleken, met de gemeten koers en mogelijke input van het botsing ontwijk systeem, in de besturing en een controle signaal is bepaald, die naar de boot wordt gestuurd.
De karakteristieken van een boot die kunnen verschillen zijn: het type, formaat, aandrijving, stuursysteem en de communicatie tussen alle verschillende system. Alle karakteristieken hebben verschillende opties met overeenkomstige voor- en nadelen. De omgeving van de opstelling limiteren het gebruik van sommige boten en is daarom ook geanalyseerd. De besturing bestaat uit een controle systeem die is gevoed met meerdere invoeren die herleidt zijn in verschillende systemen, zodat de informatie bruikbaar wordt. De herleidende systemen zijn: positie bepaling, meten koers, filteren, leiding methode en botsing ontwijk systeem. Voor alle systemen in de besturing en karakteristieken van de boot is een optie gekozen om een goede experimentele opstelling te creëren.
De sleep tank in de faculteit van 3mE is een perfecte plaats om te besturing te testen en verbeteren met één of enkele boten, omdat de draaicirkel van een hele formatie is te groot vergeleken met de tank. De poel voor de faculteit is groter en daarom een betere plek om een grotere boten formatie te laten zien, maar onhandige situaties ontstaan voor de testen in het begin, omdat het handig is dat de boten dichtbij zijn. Gebruik van beide plaatsen is onvermijdelijk, maar heeft enkele limieten. Het GPS signaal is te laag en positionering met een camera is enige overblijvende optie.
De voorgestelde experimentele opstelling is ontworpen zodat de beste opties zijn gecombineerd met de genoemde limieten. De beste optie voor de boot is een normale radiografisch bestuurbare boot, die niet te klein en speciaal is, aangedreven met propeller, een boegschroef en differentiële sturing. Het controle system is geplaatst op de boot en communiceert met de andere systemen via een draadloze netwerk. Het meten van de koers en het positioneren gebeurt met een camera en wordt gefilterd met een Kalman filter. De leider boot is geleidt met de LOS methode en de andere boten volgen de leider met formatie besturing gebaseerd op afstand – afstand of afstand - hoek. De LOS methode gebruikt een PID controller en de formatie besturing een sliding mode controller. Voor alle systemen is de simpelste, meest precieze en best passende optie gekozen.
List of abbreviations
COLREG “International regulations for preventing collisions at sea” DGPS Differential Global Positioning System
DVL Doppler Velocity Log GIB GPS Intelligent Buoy GPS Global Positioning System INS Inertial Navigation System LIDAR Light Detection And Ranging LOS Line Of Sight
RC Radio Controlled
WAAS Wide Area Augmentation Systems WLAN Wireless Local Area Network
Contents
Summary ... 3
Summary (in Dutch) ... 4
List of abbreviations ... 5
1. Introduction ... 7
2. Currently existing experimental setups ... 11
2.1 Formation control setups ... 11
2.2 Setups to test individual controllers ... 14
2.3 Positioning tracking setups ... 21
2.4 Improvements of autonomous water measurement... 24
2.5 Concluding remarks ... 27
3. Boat characteristics and availability ... 31
3.1 Different boat types ... 31
3.2 Dimensions ... 33 3.3 Propulsion systems ... 34 3.4 Steering systems ... 37 3.5 Communication schemes ... 38 3.6 Commercial availability ... 41 4. Environments ... 43
4.1 Type and surface ... 43
4.2 Obstacles ... 44
4.3 Disturbances ... 45
5. Positioning, guidance and control ... 47
5.1 Positioning system ... 47
5.2 Measuring heading angle ... 50
5.3 Filtering positioning data ... 51
5.4 Guidance methods ... 53
5.5 Obstacle and collision avoidance methods ... 57
5.6 Control algorithms ... 60
6. Analyzing environment for setup in Delft ... 63
6.1 Requirements ... 63
6.2 Situations analysis ... 63
7. Conclusions and Recommendations ... 67
7.1 Conclusions ... 67
7.2 Recommendations ... 70
References ... 71
1. Introduction
The introduction is divided in five parts. First the background information of the research is given, which is required to understand the problem statement arising from the background. The problem statement is expressed in the research questions who are the foundation of the chosen approach to find a solution to the problem statement. The approach is formed into a structure of the report, which addresses the sequence and chosen chapters.
Background
To prevent vessels from colliding, a lot of experiments are done to introduce new technique. These techniques should help the captain with maneuvering, by preventing collision and make the boats autonomous [1]. The autonomous control of a boat would result in less people required to control the boat and a decrease in workload of crew. Besides could money be saves on fuel, because the autonomous control is capable of find more optimal ship routes [2]. Even more environment friendly and cheaper are multiple vessels in vessel formation. The vessel formation can also be used in case of special or large transport, so the formation can carry the cargo together. About formation control is less known, because autonomous boats are required and these are not fully developed yet.
Problem statement
Different control algorithms and cooperation protocols are developed in order to effectively use the current water infrastructure, while taking the presence of other vessels and obstacles in account. Besides effectively use of the infrastructure, some algorithm are also engaged with formation moving. All these algorithms are already being tested in a wide range of existing experimental setups and computer simulation, but in order to bring a good algorithm to practice in the real world on real vessels in working conditions, an experimental setup has to be developed that enables assessing the performance of such control algorithms in a more realistic setting. Not only a good control algorithm, but also all details of the experimental setup has to be chosen, so the algorithm can completely be tested with good results at the end.
Research questions
The main question answered in the paper is:
Which experimental setup can make an autonomous boat formation following a path most smooth and precise ?
To the content of an experimental setup belongs a lot like the boats self, the controller and all parts the boat and controller exists of. The combination of the whole content form the setup together and should therefore also be judged as a whole, instead of taking the best of each part independently. Besides the part belonging to the setup, should also the external influencing factors of the boat, like the environment and the user interface, be considered.
In order to clarify what choices should be made and what external influencing factors should be kept in mind when designing the desired experimental setup are six sub questions defined.
What can be learned of currently existing experimental setups ?
o Why have the setups been realized ?
o What do the setups consist of ?
o How have the controller been designed and implemented ?
o What are the advantages and disadvantages of the setup ?
What characteristics of boats can be distinguished and between which possibilities can be chosen considering a characteristic?
Are there any commercially boats available with promising combinations of characteristics or is building the boats self a better option ?
How does the chosen environment influence the boats’ or controllers’ choice?
Of which systems consists a controller so the boat can be controlled properly ?
o Which positioning system is most accurate and most reliable ?
o What is the most accurate way of measuring the heading angle if this is required ?
o Is filtering the input data necessary and which filter is best if one is required ?
o What guidance method or combination can provide smooth and precise guidance ?
o Which collision avoidance method is most reliable and has minimal path influence?
o Which control algorithm has the smallest tracking error with following the guidance
and is most resistant to disturbances ?
Are there any situations in Delft which give limitations to the setup in Delft ?
Approach
An appropriate approach is required in order to give an solid answer on the main question and give a well defined advise for the experimental setup in Delft. The sub questions are formulated such that answering all sub questions would give an appropriate answer to the main question. Each sub question is answered in the chapter with the corresponding topic.
The three general topics in the control of an autonomous boat are: the boat self, environment (and other boats) and positioning, guidance and control. All are given in Figure 1.1 which represents the general control scheme of the boats. The sub questions are focusing on one topic at a time and should clarify the content of that specific topic or part of the content. The last sub question is not part of one of the three topics, but forms the connection between the whole setup and the situation the setup should fit in. Some sub questions also have sub questions and those need also to be answered and form a deeper analysis of the corresponding topic.
Report structure
The structure of the report is based on the general control scheme of an experimental setup which is presented in Figure 1.1. The main systems in the general control scheme are: boat, environment (and other boats) and positioning, guidance and control, which is also called controller. In the scheme are those three main systems presented and the arrows between the systems represent the connections. The dashed arrows between the boat and environment represent a physical communication, like collision and interaction of disturbances, and the remaining solid arrows represent data flows. The user can only influence the setup by sending data, like trajectory or paths, to the controller. The data flows between the boat and controller are: positioning estimates to the controller and the control signal return. The last remaining data flow is from the environment to the controller, which represents the position of obstacle in the environment and other boats, so the controller can avoid collision.
Figure 1.1 General control scheme of the boats. Solid arrow are data flow and dashed ones are physical contacts.
The main systems in the general control scheme forms the basis of the partition of the chapters. Besides the discussion of the three main systems, are the currently existing experimental setups described in order to gain information to analyze the main systems. To answer the last sub question is an addition chapter is required to analyzed the connection between the whole setup and the situation the setup should fit in. Because general knowledge of setups must be gathered first are the currently existing experimental setups described in Chapter 2. The existing setups are ordered at their reason of realization. The first main system from the general control scheme in Figure 1.1 which is analyzed is the boat, because that is the center of the control scheme. Therefore is Chapter 3 about the characteristics and the commercially availability of boats. Chapter 4 is about the connection between the environment and the boats, and analyze the influence the chosen environment can have on the boat to use. The whole structure of the controller (positioning, guidance and control) is analyzed in Chapter 6 and refers also to the interaction the user has on the controller. Chapter 7 is about the influence the situation at the faculty at the TU Delft has on the choice of the three main systems. The conclusions and recommendations are done in Chapter 9.
User
Positioning, Guidance and Control
Boat Environment
2. Currently existing experimental setups
Different experimental setups have already been realized worldwide to investigate the behavior of autonomous boats. All these setups differ from design, implementation and reason for realization, making them all very interesting for comparison. The most interesting existing experimental setups are discussed in this chapter and a summary of the important principles is given in Table 2.1 at the end of this chapter. An individual name for each experiment is required, i.e. to refer to when explaining a specific principle, so each setup is named. When a setup do not have a specific name, the setup is named after the boat used or another specific detail. The varying and important things in the currently existing setups are: reason for realization, content of experiment setup, design and implementation of the controller relative to the boat, advantages and disadvantages. Therefore the explanation of the setups consists of the following content: Reason of realization
Contents
Design and implementation
Advantages
Disadvantages
There are a lot of currently existing experimental setups, but only the most varying and interesting setups are treated. The discussed setups are split into four categories based on the reason for realization of the experiments, which form also the structure of this chapter. The first two sections consists respectively of the experimental setups who are realized to test formation controllers and individual controller. The third section discusses setups where a positioning system is tested. Setups for who are improving the sampling of water measurements are treated in the fourth section.
2.1 Formation control setups
The currently existing experimental setups discussed in this section are realized to test the formation controller. Formation can be achieved with different types of control and both setups discussed rely on a different method. The first setup is the Silverlit setup and consists of multiple boats who are centrally controlled and some boat may be controlled individual. The Auton robot setup is discussed secondly and consists of one boat following two virtual leader using a special formation controller.
2.1.1 Silverlit
Reason of realization. Simple self propelled robots were the basis of the Silverlit experiment [3] where collective motion is observed, emerging as a result of inelastic collision only. Demonstrated in the experiment is the presents of the phases jamming, clustering, disordered and ordered motion. Also is shown that the noise level has a fundamental role in phase transition and generating collective dynamics. With introducing intelligent leaders, which can steer individually, is demonstrated that even a few guided leaders can determine group direction and enhance ordering through inelastic collisions.
Contents. The experiment setup consisted of up to 30 commercial RC boats, a plastic pool, an industrial camera and software. The simple Silverlit RC Hovercrafts (shown in Figure 2.1a) were used in the experiment and are only capable of doing three thing: going forward, going backward and turn in random direction, and float without propelling force. These boats were controlled completely passive and were “communicating” with each other through collision. In the second experiment a real time tracking system and a couple of steerable boats were used to introduce intelligent leader boats into the flock. The leader boats (shown in Figure 2.1b) were controlled individually to align their orientation clockwise or counter clockwise. The pool was formed annular with respectively an outer- and inner diameter of 180 and 97 cm, giving place for about 500 boats. The camera and all software were added to control the leader boats and to observe the flock behavior.
Figure 2.1 The simple RC boats used in the Silverlit setup. (a) The Silverlit RC Hovercraft. (b) Leader boat [3]
Design and implementation. A graphical representation of the control design used in the experiments is shown in Figure 2.2. The camera was placed above the centre of the pool and made 800x800 pixel images, resulting in a resolution of 2.5 mm/pixel. For easier observing, the hovercrafts were painted black and a retro reflexive elongated white marker was placed in the middle to aid the tracking algorithm. Missing data points were filled using linear interpolation from neighboring frames. The leader boats were identified using an unique colored barcode on top. The multi target tracking software was used to determine time, position and orientation of each boat in each frame of the camera. The orientations of the boats was defined using the local angle polar coordinates, making the orientation independent of the location. The phase control of the boats were computer driven with a fixed repeated sequence corresponding to the going experiment. The control signals were sent through a parallel port and relays connected to the common RC of all hovercrafts as can be seen in Figure 2.2. The only difference with the control of a leader boat was that the control signal of a leader boat was sent to that leader boat independently by the RC ID shown in Figure 2.2
In order to simulate different situations a couple of parameters were controlled. Firstly, a noise level can be generated and controlled precisely with the length of the backwards motion of the hovercrafts. Secondly, the number and length of the orientations phases were differed to investigate the collective flock behavior using a small number of controllable leader boats.
Figure 2.2 Design of control scheme used in the Silverlit experiments. The boats in the pool are observed by camera, the controller determines the control signal, using the observations, who are sent via the RC to the
boats. RC ID controls the boat with corresponding ID and RC controls all boat without ID [3].
Advantages. This experimental setup relies only on the simple interaction (collision), which seems to be enough for group navigation. While most experiments dealing with swarming robots use at least some kind of self organizing intelligence. The use of relatively few individual controlled boats to collectively move a whole flock, is another example of a simple interaction. Leaders also extend the noise range where an ordered state is present and increase the critical noise level of the transition into the disordered state. Using the local angle polar coordinate system in combination with the annular pool is another big advantage, because of the easiness to analyze the orientation of all boats together. The number is hovercrafts and leaders can easily be extended. The last advantage is the ability to tune the noise level of the system in a wide range.
Disadvantages. The need of a velocity vector to observe collectively moving and phase transition, because the orientation vector cannot give satisfying results and abrupt changes. For long term ordered motion some noise is need, which may cause the system to switch.
2.1.2 Auton robot
Reason of realization. Two formation controllers for marine unmanned surface vessels were tested in the Auton robot experimental setup [4]. The formation controllers were designed using the sliding mode approach. The underactuated boat should be able to operate in arbitrary formation configurations by using two leader virtual leaders and a leader - follower control scheme.
Contents. The boat used in the experiment (shown in Figure 2.3) has only two actuators and was underactuated since three degree of freedom are assumed for the design. The experiments were performed in a natural environment on a lake using a small test boat. The sensor unit and control computer both onboard the auton robot and with a RC servo switch it was possible to switch between automatic and manual control. Besides the sensor unit the controller was also fed with the position of the two leaders. The sensor unit consists of: three accelerometers, three rate gyroscopes, a magnetometer and a wide area augmentation system (WAAS) capable GPS receiver.
Design and implementation. In the first 30 second a PID controller was used to initialize the boat with the desired speed and the desired angle. Then the position of all vessels were reset to their initial conditions and the slide mode controller took over control to follow the simulated leaders. For the sliding mode controller both distance-distance and distance-angle control were used. The formation was tested with three maneuvers: a straight line, a circular motion and a zigzag maneuver.
Advantage. The setup with virtual leaders is advantageous, because the leaders follow a perfect course. The test have shown the effectiveness and robustness of the control laws in the presence of parameter uncertainty and high environmental disturbances.
Disadvantage. Replacing the virtual leaders with real boats, solves the disadvantage of the following boat suffering from disturbances and the leaders not, which will lead to better performance of the controller. Doing the experiment in still water with less wind and wave disturbances, will also result in better performance.
2.2 Setups to test individual controllers
Besides the formation controller are there also individual controllers, which can be of interest of formation control, because some techniques may be useful to use in formation control. The coming six setups all use a different individual controller and are called: internal model autopilot, Cybership II, Sea grant college, Springer, Atlantis, CENDAC.
2.2.1 Internal model autopilot
Reason of realization. In the interval model autopilot experimental setup [5] a real time kinematic GPS based track keeping control of a small boat was used to demonstrate the practical use of the controller. The internal model control method was adopted in the autopilot design and the controller was recast in the PID controller format that is characterized by its simple structure and relative ease of implementation.
Contents. A four meter fiberglass reinforced plastics boat (shown in Figure 2.4) was used in conducting track keeping experiments and was equipped with one outboard motor mounted at the stern to provide propulsion and steering. The position of the boat, needed for determination of the position and reference heading angle, was provided by the real time kinematic GPS. The actual
heading angle was provided by an electronic compass. The GPS and compass signal, being updated once per second, were sent to the onboard mounted PC were the guidance law and autopilot control algorithm are implemented, as input data to perform the control signal for the outboard motor.
Figure 2.4 Fiberglass reinforced plastics boat in a harbor used in the internal model autopilot setup [5].
Design and implementation. Computer simulations were carried out to find the feasible controller design parameters, who were applied in the algorithm to control the boat during the experiment. Two approaches were adopted in selecting the waypoints, who were required by the line of sight guidance method to compute the reference heading. First, the waypoints were selected arbitrary and the boat was required to track the preselected waypoints. Second, a path planning based on Bezier curves was adopted, which selected several waypoints on the curve generated to be tracked. The internal model autopilot is tested using both guidance approaches. Also the target zone radius around a waypoint was reduced to check the effect of different radiuses.
Advantage. Useful is the path planning method based on Bezier curves to achieve smooth trajectory around the obstacle. The internal model autopilot design method is easy to follow and is straightforward to recast the controller into the PID controller format, which has clear physical meaning and allows easy implementation.
Disadvantage. The proposed system in the presented form cannot outperform existing track keeping systems. The simplicity of the controller requires a target zone radius of at least the turning radius of the boat to ensure the boat is capable of reaching a waypoint after at least a couple of times.
2.2.2 Cybership II
Reason of realization. In the Marie Curie EU training site at the Norwegian University of Science and Technology in Trondheim is an experiment done with the Cybership II, which is described in [6], [7] and [8]. An adaptive recursive design technique was developed for describing the dynamics of a boat. First the geometric part which forced the system output to continuously converge to a desired path. Second an update law was constructed that bridges the geometric design with the dynamics
task, which should satisfy a desired dynamic behavior along the path. The design procedure is performed and tested with several experiments for a model boat. Besides the described algorithm lot of other guidance and control algorithms are tested with the Cybership II in the Marie Curie EU training site (shown in Figure 2.5), but those are used to a less extent in the experimental setup.
Figure 2.5 The Marie Curie EU training site (MC lab) at the Norwegian University of Science and Technology [7].
Contents. The Cybership II is an 1:70 replica of a supply ship (shown in Figure 2.6) and was used in this experimental setup. Its mass was 23.8 kg, length and breadth were respectively 1.3 and 0.3 m. The Cybership II was fully actuated with two main propellers, two rudders and one bow thruster. Further, the boat was equipped with a real time operating system which controls the internal hardware and communicates with onshore computers through WLAN. To facilitate real time feedback control of the ship a camera and addition software programmed in MATLAB and SIMULINK were used.
Figure 2.6 Cybership II in the MC lab, replica of a supply ship with LEDs for positioning using camera above [6].
Design and implementation. The main difficulty was to find the coefficients in the damping matrix. By towing the ship at different speeds in different directions and measuring the corresponding drag forces, about half of the damping coefficients have been identified. The vector of the desired parameterized path to be followed in the geometric part, consisted of the x- and y- coordinate and the heading angle. To satisfy the second task, the desired dynamic behavior along the path, the previous mentioned vector was made time dependent. In the setup a path could also be defined using waypoint. Sub tracks between the waypoint were used in the polynomial interpolation method to create a smooth path through all waypoints.
The user interface gives the possibility to choose from seven different allocation algorithms, three different controllers and three different guidance methods. The most often used combination works as followed, first, the adaptive maneuvering system using the gradient update law was tested by following an ellipsoid with heading along the tangent vector. Second, the performance of a gradient update law implementation versus the tracking update law when the forward thrust is forced to saturate was compared. Other often used combination where the line of sight method, docking and trajectory tracking. These three were also tested in combination with waypoints, obstacles and docking possibilities.
Advantage. The gradient update law procedure has the ability to adjust online the forward speed along the track, so the ship will not get off track due to lack of power. So the ship is capable of following a continuously defined track at a speed which is close to the desired speed. One of the major strengths of using a graphical searching algorithm to search the map is the need of minimum of preparation. All that needs to be done is discretize the map and assign a value to each coordinate.
Disadvantage. The major weakness of the path generation algorithms is that the curvature radius of the path is not taken into account. Hence, infeasible paths may be experienced due to tight curves. The computing time is also a disadvantage, because the Cybership II has only one processor so all computing tasks share the same memory. Disadvantageous is that all parameters have to be updated before the procedure can be used for another vessel design. So when this procedure will be used for navigating another vessel, all parameters and matrices need to be redefined.
2.2.3 Sea grant college
Reason of realization. In the sea grant college experimental setup [9] a fuzzy controller was developed to provide waypoint tracking guidance control of an autonomous boat that will be used for oceanographic research. The membership functions and fuzzy rules were generated and after initialization, the vehicle was successfully tested in an environment with varying external disturbances.
Figure 2.7 Autonomous boat used in the sea grant college at the MIT Sailing Pavilion dock in Cambridge [9].
Contents. The boat was a fiberglass epoxy, 1:17 scale fishing trawler (shown in Figure 2.7) having a length and maximum width of respectively 1.4 m and 0.4 m and the water displacement is 23 kg. For
autonomous operation the boat was equipped with a computer to perform all computations, a rudder and a thruster motor. The fuzzy controller required as inputs the boat’s heading and position, which are given by a digital flux gate compass and a differentially corrected GPS receiver The vehicle’s speed was estimated by passing GPS position updates through a scalar Kalman filter. A simple algorithm was used to estimate the vehicle’s position between GPS position updates.
Design and implementation. The guidance problem was to navigate to and cross, on a specified heading, a series of waypoints defined by their position, crossing heading and arrival radius. The fuzzy sets and rules were designed such that waypoints are crossed with the read heading and a special action is enforced when the waypoint is missed. Different patterns with waypoints were tested.
Advantage. The model free nature of the fuzzy controller is advantageous, because the nature allows rapid design and implementation of the control laws without initially having to develop dynamic models or complex control system architectures. The heading constrained waypoints allow paths to be generated with arbitrary complexity. Close approaches to complicated shorelines and safe navigation down narrow waterways can be achieved with using the fuzzy controller.
Disadvantage. Fuzzy control does not allow analytic analysis, but this is not a serious limitation in the development of the waypoint following controller presented.
2.2.4 Springer
Reason of realization. An automatic collision avoidance technique for an autonomous and underactuated surface vehicle called Springer based on standardized rules is tested in [10] [11]. The approach was essentially a reactive path planning algorithm which provided feedback to the autopilot for steering the craft safely. The dynamics of the underactuated surface vehicle were also incorporated, providing realistic trajectories which are closely followed by the autopilot. Also an offline grid based path planning scheme was modified to produce COLREGs compliant paths.
Contents. The Springer was a catamaran shaped research vessel and is shown in Figure 2.8. Each floater of the Springer is divided into two watertight compartments containing some of the onboard sensors and electronics including battery pack. A GPS receiver and wireless router to link the onboard computer to the base onshore were also installed. The steering mechanism is based on differential thrust. Three different compasses were used to determine the current heading angle.
Design and implementation. The proposed strategy consists of waypoint guidance by line of sight coupled with a manual biasing scheme. Where the heading angle is increased by the bias angle if an obstacle comes to close to the boat. The other avoidance technique is based on a general avoidance direction vector, so the COLREGs are neglected. A simple PID autopilot is incorporated to ensure that the boat adheres to the generated seaway. Simulation experiments had been carried out both for static and dynamic obstacles.
Advantage. In this setup the results of both avoidance techniques can easily be compared. Manned vessels could also benefit from these autonomous path planning, thus helping to eliminate the subjective nature of human decision making and safeguarding the onboard personnel.
Disadvantage. More advanced motion planning strategies can be investigated for COLREGs compliance using evolutionary algorithms such as genetic algorithms and particle swarm optimization.
2.2.5 Altantis
Reason of realization. An autonomous catamaran (called Atlantis [12]) fitted with sensors and actuators was built to test the viability of GPS based system identification for precision control. The identification process used a special method for identifying a linear time invariant model and associated pseudo Kalman filter. System identification input was generated using a human pilot driving the catamaran on roughly straight line passes. A fourth order discrete time model was generated from the data and was shown to be excellent in prediction results. Using these models controllers were designed and tested.
Contents. The boat is based on a modified catamaran and used wing sail propulsion at the end, but an electric trolling for propulsion first because the wing sail was under construction. A rigid wing sail was used (shown in Figure 2.9) and is capable of self trimming. The boat is equipped with GPS for localization, distributed sensors sampling at 100 Hz and actuators connected with a high speed digital serial bus and the central computer performs estimations and control tasks at 5 Hz.
Design and implementation. The system identification algorithm uses input/output data from a rich stream to produce a differential or difference equation of the system control. The identifying method formulation minimizes the error in the observer, which will converge to the true Kalman filter for the data set used if the true world process is corrupted by zero mean white noise. From the identifying method methodology, the linear system matrices and the estimator G are extracted. The control gain matrix K is extracted from the control design, after while a controller is generated and uploaded to the Atlantis computer. Runs along a straight line are done with the controller during changing currents, wind and waves.
Advantage. The experimental approach to generate a model that does not require the tedious and often imperfect modeling of the physical model. Using the controller has several advantages, first, the controller is the best mathematical representation of the system possible. Second, the identifying method method provides the Kalman gain directly, lastly the controller can only act on the measured states and thus forces more sensors onto the system to adequately measure all of the states. Great is to see that large gusts simply caused the wing to quickly stall and with only a slight shudder reposition at the new angle of attack
Disadvantage. The controller does not have any of the guaranteed stability margins, robustness nor simplicity that the controller possesses. The Atlantis also requires wind to move, because the propulsion is based on the wind. Too much wind also not work, because the Atlantis then cannot handle the stability anymore and will fall over.
2.2.6 CENDAC
Reason of realization. A sliding mode control law was presented and experimentally implemented for trajectory tracking of underactuated autonomous surface vessels in the CENDAC setup [13]. The control law was developed by introducing a first order sliding surface in terms of lateral motion tracking errors. The resulting sliding mode control law guaranteed position tracking while the rotational motion remains bounded.
Contents. The CENDAC setup consisted of a small boat with two propellers (shown in Figure 2.10) in an indoor pool. The position and orientation of the boat were measured using a camera that detects two infrared diodes attached near the front and back ends of the boat. A computer with a capture card processed the camera image to determine the position, calculated the control forces and their corresponding input voltages and sent the control signals to wireless receivers on the vessel using a wireless transmitter.
Figure 2.10 Boat used for experimentally testing the sliding mode control in the CENDAC setup [13].
Design and implementation. In the sliding model controller asymptotically stable surfaces were defined such that all system trajectories converge to these surfaces in finite time and slide along them until the desired destination at their intersection is reached. Multiple series of straight line trajectory control experiments were conducted at different heading angles. Also a couple of experiments with a circular trajectory were done.
Advantage. A good tracking control is obtained due to the robustness of the sliding mode control approach. Another advantage of the setup is the calibration of the camera image, which works good.
Disadvantage. There are a couple of sources of error, like: minor camera calibration errors, lack of speed control at low voltages due to the dead band of the motors and motor calibration errors. To provide good performance the two propellers must rotate in opposite direction, but the motor speed control hardware does not perform the same in each direction.
2.3 Positioning tracking setups
Two existing setups are specially realized to test a specific positioning system. Other positioning systems are used in other existing setups (like the camera), but those setups are realized with another reason and are therefore discussed in another section. GPS and two relative positioning systems onboard are analyzed in the Songhua lake setup. Secondly is the MEDIRES setup analyzed which has a GPS intelligent buoy positioning system.
2.3.1 Songhua Lake
Reason of realization. In Songhua lake near Harbin Engineering University an experiment [14] was carried out to test a proposed Kalman filter method working in the GPS / inertial navigation system (INS) / Doppler velocity log (DVL) integrated mode which combines outputs of all three system.
Normally submarines use only a combination of INS and DVL to determine their position, but the precision and reliability of this combination is severely limiting the performance of fine maneuvering, since the position error accumulates with time. GPS can provide superior three dimensional navigation capability, but the signal cannot be directly received by all submarines. Therefore a Kalman filter method, combining INS and DVL information and the GPS data when available, is proposed and tested.
Contents. Two boats are were attached to each other as shown in Figure 2.11 and the DVL was put in between at a depth of 1.5 meter. The antenna of the GPS receiver was put on the roof of one the boats. The boats was also equipped with INS and was controlled manually.
Figure 2.11 The two boats, attached together, which are used in the Soghoa lake experiment [14]
Design and implementation. Three phases can be distinguished in the experiment: preparation, performance test and long distance test. During all three phases the DGPS data was used as reference to determine the position and velocity error. In the first two phases the INS and DVL combination was tested and a control experiment with the DGPS as reference was done. A long distance voyage was carried out in the third phase to test the proposed Kalman filter method in different combination of the three navigation system.
Advantage. Using multiple navigation system gives the ability to combine the systems and get a higher precision and reliability. Position tracking is also still possible during connection failure with one or more navigation system and reconnection can give an correction on the position estimate. All due to the proposed extended Kalman filter, which is effective on limiting the position error and velocity errors of INS and DVL.
Disadvantage. The methods is specially designed for submarines where position tracking should be as precise as possible and the systems used nowadays can be improved with partial GPS data. But when a GPS data is constantly available the other systems and the extended Kalman filter could slow down the process or introduce errors with are both unwanted.
2.3.2 MEDIRES
Reason of realization. In the MEDIRES [15] experimental setup is the GPS intelligent buoy (GIB) system tested, which is capable of estimating position of an underwater target. Acoustically measured boat to buoy ranges are fused by resorting to an extended Kalman filter structure which forms the basis of the GIB system. Due to finite speed of propagation of sound in water, a certain amount of time is required before the buoy receive the emitted signal. By dealing directly with each buoy measurement as it becomes available, a system is obtained that exhibits far better performance than that achieved with classical triangulation schemes, where all buoy measurements are collected before an estimate of the target’s position can be computed. The new extended Kalman filter structure together with the coded signals, making validation and depth determination possible, are simulated in the MEDIRES experimental setup.
Contents. The experimental setup consisted of a system of four buoys receiving a periodically emitted acoustic sent by a pinger, which is graphical represented in Figure 2.12. The pinger in the figure is placed on a submarine, but in the MEDIRES experimental setup the pinger hung beneath a boat. The four buoys were placed at the corners of a 500 meter square and were equipped with hydrophones to receive the signal and DGPS to determine the position of the buoy. The underwater pinger carried a high precision clock that is synchronized with those of the buoys and emitted an acoustic signal, at precisely known instants of time. The buoys recorded the time for each signal is received and sent the required time using RC to a central station (typically onboard a support vessel), where the position of the underwater target was computed. A pulse consisted of two successive acoustic pulses to determine depth, which is proportional to the time delay between the two pulses.
Figure 2.12 GIB system graphically represented. The submarine sends a signal at known time intervals to all buoy (position determination with DGPS) to determine to position of the submarine using time differences [15].
Design and implementation. The boat was also equipped with two GPS receivers and in combination with a GPS receiver installed inshore to provide corrections in the post processing phase. With these receivers an accuracy better than 10 cm was obtained, which was used as reference. The
boat was steered manually between the four buoys and sent a signal every second. The extended Kalman filter had a sampling time of 0.1 second, so the position could be updated every 0.1 second.
Advantage. Even in the presence of dramatic failure of a buoy, the positioning system can still provide good estimates. Another advantage of the setup with the extended Kalman filter is the relatively good position estimate, with use of the GIB system. So the extended Kalman filter is handling the measurements delays very well in comparison with the classically triangular approach. The low velocity of sound in water is not only a disadvantage, because a higher precision can be reached due to the larger differences in time caused by the low velocity.
Disadvantage. A disadvantage of an implementation using acoustic positioning system is the requirement of mechanisms being developed to deal with dropouts and outliers that arise due to acoustic path screening, partial system failure and multipath effects. Obviously is the extreme importance to have an accurate estimate of the velocity of sound since that is used in the process of transforming the times a signal requires to reach the buoy. The low velocity of sound in water result in a longer time before the signal reach the buoy.
2.4 Improvements of autonomous water measurement
Some existing experimental setups are not realized to test autonomous boats, but to improve water measurements. In most cases that result in the use of an autonomous boats, taking those water measurements. In these setups the boats has to collect water samples as quick as possible, while the position of the sample has to be exact. The water sampling setups discussed are: Roboduck II, Airboat and DELFIM.
2.4.1 Roboduck II
Reason of realization. The NAMOS team consists of biologists and engineers who believe that the use of Roboduck II [16] to perform autonomous sampling can improve the efficiency of data collection at high quality, since robots can be programmed to perform sampling tasks with higher repeatability without appreciably losing efficiency over time. Roboduck II is designed, made and used by this team to collect water samples.
Contents. The Roboduck II (shown in Figure 2.13) was equipped with: equipment to take water measurements at different depths, GPS receivers, an RC receiver/transceiver and a three degree of freedom inertial measurement unit consisting of gyroscopes, accelometers and an integrated magnetometer compass. The motor controller onboard was capable of speed control of the two DC motors driving the boat’s propeller shafts. Differential thrusters were used to perform steering The user interface was placed onshore and sent mission files to the computer onboard and receives data. The boat could also be controlled manually via the user and otherwise waypoints and “Keep out” area’s could be created in the user interface to control the boat automatically.
Figure 2.13 Roboduck II of the harbor patrol dock at Redondo Beach used for water measurements [16].
Design and implementation. The heading autopilot used a PID controller to control the input signal of both motors, to provide trust and steering action. An outer loop around the inner heading autopilot loop was used to compute a desired heading angle to reach a next waypoint. Stereo vision avoidance technique used two cameras and could determine distances to an object, so the obstacle could be avoided. In the experiment the boat was sent along a trajectory of waypoint, but the automatic collision prevention was not used because the system was not working properly.
Advantage. The user interface used in this setup is very useful, because easily a mission file can be created. Also calibrating the map to the GPS being used on the boat reduces the chances of the boat running into obstacles that are indicated to be water bodies by its own erroneous GPS reading.
Disadvantage. The differential thrusters used for steering works quite well at low speeds, but is not efficient at higher speeds due to the efficiency drop at a higher rpm. Another drawback is that the LOS guidance setup is susceptible to currents. The collision avoidance is not working well, because recognizing obstacles from the images coming from the cameras is very complicated.
2.4.2 Airboat
Reason of realization. An unmanned airboat for mapping the water quality of shallow mire pools where aquatic weeds flourish was developed in [17]. The boat was designed for automatic operation and was capable of measuring: temperature, pH, dissolved oxygen, electrical conductivity, turbidity and chlorophyll-a. In a 26 ha mire pool in Hokkaido, Japan the water quality was measured in a grid size of 10, 20 and 40 m.
Contents. The airboat was a float boat as shown in Figure 2.14 and had two floaters on each side and a platform in between. The boat was equipped with four gasoline engines, GPS compass including two DGPS receivers, equipment to provide the water measurements, a single board computer, a camera to observe the surroundings and a WLAN antenna to allow observing the status of the boat from the shore. The airboat was propelled by onboard fans and steered by differential steering.
Figure 2.14 Airboat driven by fans used for mapping water quality of shallow water [17].
Design and implementation. The position of the boat was translated in a special coordinate system, where the direction of the target point is important. The rotational speeds of both fans could easily be calculated from parameters derived from the coordinate system. With the rotational speeds the boat was controlled.
Advantage. The approach to reach target points is useful even when the wind is blowing, but another boat would increase the performance. A rudder is ideal to drive the boat in a stable straight line, but a rudder would in this setup get stuck in the weed.
Disadvantage. A big disadvantage of using fans is a deviation which will occur in most of all cases, because the boat suffer from every wind blow and unbalanced mass. Another problem is that the boat occasionally swayed from side to side, which can be eliminated by increasing engine control frequency and the gain of the controller.
2.4.3 DELFIM
Reason of realization. DELFIM was an autonomous boat [18] [19] developed for automatic marine data acquisition and to serve as an acoustic relay between a submerged craft and a support vessel. DELFIM was instrumental in enabling the transmission of sonar and video images through a specially developed acoustic communication channel that was optimized to transmit in the vertical water column. The DELFIM could also be used as a standalone unit, capable of maneuvering autonomously and performing precise path following. Practical results were obtained during sea test in the Atlantic.
Contents. The DELFIM was a small catamaran and was driven by two propellers with differential steering as shown in Figure 2.15. Systems for navigation, guidance and control and mission control as well were mounted onboard and control the DELFIM. Navigation was done by integrating motion sensor data obtained from an attitude reference unit, a Doppler unit and a DGPS. RC with a range of
80 km was used for the transmissions between vehicle, the support vessel and the control centre. At the bottom a low drag body was installed that could communicate with a submerged craft.
Figure 2.15 Shaded view from beneath the DELFIM’s 3D model. Yellow part houses all acoustic transducers [19].
Design and implementation. A model was derived from basic principles of physics that borrow from the theory of rigid body motion dynamics and kinematics. Heading and path following controllers were used to control the underactuated boat at constant speed and under influence of strong wave action. Simple yaw changing maneuvers were run and a path following algorithm along the longer transects was ran while fighting waves and the ocean current.
Advantage. The new class of trajectory tracking controllers exhibit two major advantages over classical ones: stability of the combined guidance and control system is guaranteed and zero steady state error is achieved about any trajectory. The DELFIM has proven to be a reliable and stable platform, capable of sailing in the open sea under not too adverse conditions. The vehicle’s main characteristic, namely the presence of a wing carrying a torpedo shaped body proved an interesting solution to the problem of housing acoustic systems below the vehicle’s waterline. The last advantage is that the vehicle’s modular organization is easy to integrate sensors and actuators seamlessly.
2.5 Concluding remarks
A lot of experimental setup are mentioned and each setup differ from the other, making them all very interesting. All mentioned setups, except one, consist of only one boat, but some algorithms are still useful for formation control. With the knowledge of the formation control algorithms other algorithms could be judged at their possibilities of adjustment to a formation controllers. The information is summarized in Table 2.1, where already a certain distinction is made. The distinction is made such that the same ordering as the coming three chapters are used, namely: boats, environment, positioning guidance and control. The possibilities used in existing experimental setups forms the basis of the coming chapters, where all information is ordered and the principles are discussed in detail.
Table 2.1 Summary of proposed existing experimental setups (setup with formation and individual controllers)
Name experiment
Silverlit
Auton robot
internal model
autopilot
Cybership II
references
[3]
[4]
[5]
[6] [7] [8]
realization reason
collective motion
with a few leaders
test formation
controller
test internal
model autopilot
test different
control algorithm
B
o
at
description
miniature
hovercraft
2 virtual leaders
and 1 real boat
fiberglass
reinforced plastics
boat
1:70 scale replica
of supply ship
type
hovercraft
normal boat
normal boat
replica
dimensions
15 cm x 10 cm
0.8 m x 7.8 kg
± 1m x 4m
1.255 m x 0.29m,
23.8 kg
number
27
1
1
1
propulsion
propeller
two propellers
outboard motor
at the stern
2 propellers, 1
bow thruster
steering
leaders use a
rudder, rest only
collision
rotate the
propellers
the same stern
outboard motor
two rudder
communication
RC
RC
positional data
link (PDL)
WLAN
En
vi
ro
n
men
t
type
annular pool
inside
freshwater lake
harbor of NTOU
Marie Curie EU
training site
surface
for 500 boats
very large
large
6,45 m x 40 m
obstacles
only the walls
none
known obstacles
virtual
circumstances
still water
waves and wind
waves
controllable
Po
si
ti
o
n
t
ra
ck
in
g
system
camera
WAAS capable
GPS
real kinematic
GPS
camera
filtering
-
-
low pass filter
-
update
frequency
10 FPS
continually
GPS and compass
at 1Hz
50 Hz
precision
2.5 mm/pixel
resolution
?
up to 2 cm
?
G
u
id
an
ce
measuring
heading angle
elongated marker
on top of each
boat
3 accelero-,
magnetometer, 3
gyroscopes
TCM2 electronic
compass
2 light bulbs are
mounted on back
and front of boat
coordinate
system
local angle polar
relative to leaders
global
fixed in basin
guidance
description
pre described
orientation
follow leaders
waypoint,
arbitrary or Bezier
curve
pre described
path or waypoints
guidance
method
only leaders keep
orientated as
described
distance –
distance or
distance – angle
LOS
Karl-Petter (2),
Eva, N3, invers,
Thomas, LOS
C
o
n
tr
o
l
control
algorithm
only steer leaders,
rest are controlled
due to collision
slide model
controller
internal model
gradient update
law or LOS
collision
avoidance
it is the purpose
to collide
-
path is generated
around obstacle
with Bezier curves
A* algorithm
forbidden areas,
time independent
The meaning of - is inapplicable (or not given but also not of interest) and ? means unknown.Continuation of Table 2.2 Summary of proposed existing experimental setups (individual controllers)
