EXECUTIVE SUMMARY... 3 II. INTRODUCTION... 4 III. PHYSICAL DESCRIPTION...

Transcription

1

2 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E CONTENT I. EXECUTIVE SUMMARY... 3 II. INTRODUCTION... 4 III. PHYSICAL DESCRIPTION... 5 EXTERNAL ARCHITECTURE:... 5 INTERNAL ARCHITECTURE:... 7 ELECTRONIC ARCHITECTURE:... 8 Overview... 8 Thrusters control... 9 Embedded computers Sensors Communication Power supply IV. AUTONOMY AND MISSION PLANNING OVERVIEW: PROCESSING ALGORITHMS: Overview Set-membership methods and outliers Interval arithmetic Localization Image processing CONTROL: Regulations Autonomy V. INNOVATION VI. FINANCIAL SUMMARY VII. RISK ASSESSMENT VIII. REFERENCES

3 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E I. Executive summary As last year and every year since 2007, ENSTA Bretagne (previously ENSIETA) will participate in the SAUC-E competition. Our AUVs SAUC ISSE and SARDINE will be returning with the new addition of a hovercraft. This year, we focused on the dynamic localization problem of a submarine in a pool using the sonar and compass (no Doppler Velocity Log or other direct speed sensor) and using a new algorithm based on interval arithmetic for its robustness even in the case of outliers. Additionally, we also developed a hovercraft (surface robot with 2 thrusters) that should help us do more tests and videos of our submarines. 3

4 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E II. Introduction For the past two years, ENSTA Bretagne has used two AUVs to compete: SAUC ISSE and SARDINE. The use of two submarines has proven to be useful for several reasons: The new robot, SARDINE passed the qualifications successfully in This year, we plan to use it to touch the midwater target orange ball (we did not have the time to really work on it last year). We might also install hydrophones on board as to follow the surface vehicle. Another test that could be done is to use a mirror in front of the front camera of the robot to see either the bottom (to follow the pipeline) or above (to follow the surface vehicle) since SARDINE only has a forward looking camera housed inside the vehicle and looking though a plexiglas window. SARDINE replaced SAUC ISSE when it had problems. As most of the work is done by several students, having several robots is very helpful to test and practice both new code and hardware. It was a good way to test new ideas without loosing in reliability (we kept SAUC ISSE with the old but reliable concepts). One of the missions of SARDINE was also to follow SAUC ISSE autonomously to take underwater videos of it. We did not really have the time to do it last year but we plan to work on it soon. It could be the makings of a collaborating swarm of robots in a submarine context, which would introduce something new and innovative. For example, instead of using a single complicated robot with several different sensors, we could use several different and more simple robots, each one designed for a dedicated task (a camera-robot, a sonar-robot ). Following last year s idea of using several robots, we decided this year to also build a surface vehicle, because it is easier to build and can be useful to carry out tests, especially to divide the work between people. The structure of this paper is broken down as follows. First, we will detail the physical architecture (mechanical and electronic) of our robots. Then, we will explain how we will handle the autonomy of the robots and the missions. A section detailing the innovations in our robots will follow. Finally, a financial summary and a risk assessment table will be provided. 4

5 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E III. Physical Description External architecture: SAUC ISSE and SARDINE share the same mechanical base (Figure 1). Indeed, an aluminum tube is an attractive choice for its resistance to pressure and corrosion as well as its amagnetism and other properties that make it perfectly suited for sea usage. Its size is no longer than 70 cm to ease its transportation and reduce its weight, and the space inside is sufficient to contain all the needed devices. The diameter of the tube for SARDINE was chosen to be 20 cm to allow for an EeePC as an embedded computer (contrary to SAUC ISSE, which has a tube of 17 cm which can only contain a PC/104). Figure 1 : SAUC ISSE (left) and SARDINE (right) The watertightness of each tube is made by two aluminum covers (Figure 2) with toric joints (O rings) and waterproof connectors (Switchcraft EN3 and Bulgin Buccaneer, these connectors limit the maximum operating depth of the submarines to 10 m) to connect the external sensors and actuators of the robot to the internal elements. Three stainless fastener screws maintain each cover on the tube and three extraction screws ease the opening of the tube. The front cover of SARDINE has a Plexiglas window to enable the use of a webcam directly inside the tube. All the waterproof connectors are on the rear cover. This cover should only be opened in case of a problem, as opening the front cover is sufficient to change the batteries and switch on the EeePC. Therefore, only the front plaque, that has no connectors on it, should be opened for normal operation. The principle is the same for SAUC ISSE (but it does not have a window on its front cover). 5

6 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Figure 2: The rear cover of SAUC ISSE with its waterproof connectors (interior side) and the front cover of SARDINE, with a window A special structure (Figure 3) was made to carry the horizontal thrusters. Figure 3: Structure that carries the horizontal thrusters of SAUC ISSE The roll and pitch are not controlled but remain stable thanks to a weighted keel, which is also a support for the sonar and the vertical thruster (Figure 4). The keel is cut to put the vertical thruster in the center of the submarine, in order to keep symmetry. Only SAUC ISSE has a sonar. Figure 4: Vertical thruster centered in the keel of SAUC ISSE The weight in water of a submarine to reach the limit of buoyancy (which changes depending on several conditions) is adjusted by adding pieces of lead on threaded rods placed in the 4 corners of the submarine. As a result we just need a very weak vertical propelling force to make the submarine go deeper, and when the vertical engine is turned off, it returns to the surface by itself. 6

7 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Internal architecture: Rails (Figure 5) placed inside the aluminum tube support the insertion of a Plexiglas plaque of 6 mm. This is the main support base for the internal electronic devices of SAUC ISSE (Figure 5). Figure 5: Rails inside the tube of SAUC ISSE Below the plaque, another sliding support (Figure 6) contains the batteries. Thus, we can readily access the batteries without having to move any of the other electronic devices, which are fixed above the main Plexiglas plaque using Velcro. Figure 6: Sliding support and internal devices in SAUC ISSE The internal architecture of SARDINE is almost the same (Figure 7). Figure 7: Internal devices of SARDINE 7

8 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Electronic architecture: Overview The electronic parts of the two submarines are similar (Figure 8 and Figure 9). Figure 8: Electronic architecture of SAUC ISSE 8

9 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Figure 9: Electronic architecture of SARDINE Thrusters control Three thrusters DC BTD150 from SEABOTIX (Figure 10), an American manufacturer specialized in ROVs (Remote Operated Vehicles) are used to make the robot move: 1 vertical thruster to adjust the depth of the submarine. 2 horizontal thrusters to control the speed and the direction. These thrusters are waterproof up to 150 m. Figure 10: SEABOTIX thruster Speed controllers Robbe Rokraft (Figure 11) are used to control the thrusters with electronic signals. 9

10 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Battery with Tamiya connector PWM signals, servomotor connector Thrusters Figure 11: Thruster controller The power sent to the thrusters (and therefore their speed) depends on the PWM (Pulse Width Modulation) signal received by the speed controllers. U : Voltage of the PWM (5 V) t : Pulse width (between 1 and 2 ms) T : Period (20 ms) Motor state Motor stopped Turn in a direction Turn in the other direction Pulse width 1.5 ms 1.5 to 2.0 ms 1.0 to 1.5 ms Stopped Figure 12: PWM signals To generate these PWM signals from computer programs, we need an interface module between the computer and the speed controllers: the Labjack UE9 (Figure 13). It is a USB (or Ethernet) device that provides several IO pins to connect to electronic devices. 10

11 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Figure 13: Labjack UE9 generating PWM signals Embedded computers The SAUC ISSE embedded computer is a PC/104 from EUROTECH with a Pentium M 1.4 GHz CPU and 512 MB of RAM (Figure 14). The operating system and the programs are stored on a hard drive 2.5 of 320 GB. There are 8 USB, 1 Ethernet, 2 RS232 and 1 VGA ports that connect several devices and communicate with the embedded computer as with any desktop or laptop computer. Figure 14: PC/104 CPU module It is powered directly from 12 or 24 V batteries thanks to a power supply module compliant with the PC/104 standard that provides regulated +5, +12 and -12 V (Figure 15). Figure 15: Power supply module for the PC/104 SARDINE s embedded computer is an ASUS EeePC T91MT (Figure 16). It is larger than a PC/104 (it could not go inside the tube of SAUC ISSE) but a little bit slimmer and has an integrated battery allowing it to run up to 5 hours (in reality it can be used from 2 to 3 hours in the submarine, with all the devices connected and several programs running), with almost the same technical characteristics (CPU, RAM ). It just needs a USB hub to connect with all needed devices. 11

12 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Figure 16: ASUS EeePC T91MT used in SARDINE Sensors To detect the objects in the basin (especially the mid-water target, the pipeline and enable the autonomous following of SAUC ISSE), we have a standard webcam Logitech Quickcam/Webcam Pro 9000 placed directly in the tube of SARDINE, behind the front window (Figure 17). This type of webcam can get pictures with a very high resolution (up to 1600x1200 pixels and even more with its integrated interpolation algorithms) and has a horizontal view angle of 75. Moreover, common defaults in webcam pictures such as distortions, light or color problems, focus, zoom are automatically handled by its integrated filter and can also be configured using the driver and software provided on Windows. Their integrated microphone could also be used to communicate with the robot by voice recognition (for example by telling it to start, stop...). Figure 17: Webcam Some webcams were also made waterproof by putting them in standard household water system PVC tubes with a Plexiglas window or PVC waterproof boxes with transparent covers (Figure 18). These could be used on SARDINE to detect the pipeline and the surface vehicle (the front webcam of SARDINE is not be suitable for these tasks). Figure 18: Home-made waterproof webcam For SAUC ISSE, analog waterproof (uo to 50 m) cameras ALLWAN AL-2121 are connected to the embedded computer via waterproof connectors (they are outside the tube, one looking to the front and the other to the bottom) and Grabbino analog audio-video to USB converters (Figure 19). 12

13 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Figure 19: Analog waterproof camera and analog audio-video to USB converter from Grabbino To get the depth of the robots, a pressure sensor Keller PAA33X connected to the embedded computer with a RS485 to USB converter (Figure 20) is fixed on the rear cover of the submarines. Figure 20: Pressure sensor An IMU (Inertial Measurement Unit) Xsens MTi (Figure 21) lent by the GESMA (Groupe d Etudes Sous Marines de l Atlantique) is used to get the orientation of SAUC ISSE with respect to the North (the single angle used is the heading since the others should remain stable thanks to the keel of the submarine) accurately and robustly thanks to its built-in fusion filter that uses magnetometers (3D compass) and gyrometers (inertial sensors) to provide a correct orientation even in case of magnetic disturbances (however corrections of the values obtained can be done to take into account the magnetic deviations due to the parts of the submarine itself). It is connected to the embedded computer via a RS422 to USB converter. SARDINE has a MTi-G (the same, but with a GPS and a RS232 to USB converter). This one seems to be less accurate than the MTi when the GPS is not available (which is the case when the submarine is below surface). Figure 21: The MTi and MTi-G To obtain its position with respect to the basin borders in which it moves and detect objects, SAUC ISSE has a Tritech MiniKing sonar (lent by other people in our school) connected to the embedded computer in RS232 via a waterproof connector. It can only be mounted on SAUC ISSE (Figure 22). This is a rotating sonar: it sends ultrasonic waves at different angles, with a rangescale of up to 100 m (and waterproof up to 1000 m). 13

14 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Figure 22: Sonar This device can provide data as images that can be interpreted more easily when the submarine does not move: the center of the image corresponds to the position of the sonar (and therefore the submarine) and we can see the basin borders as well as objects in the water (Figure 23). Figure 23: Sonar image of the SAUC-E 2009 competition area SAUC ISSE has 2 hydrophones Aquarian Audio H2a-XLR to detect the pinger of the surface vehicle (Figure 24). They can be also installed on our surface vehicle or SARDINE for tests. Figure 24: Hydrophone Communication A Wifi router D-Link DIR-600 (Figure 25) in combination with an external antenna of 1 m enables us to communicate with SAUC ISSE via a laptop when it is near the surface of the water (i.e. when the antenna is outside the water). 14

15 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Figure 25: Wifi router of SAUC ISSE and access point of SARDINE If the robot needs to be controlled at greater depths, the antenna is put on a buoy connected to the submarine via a cable (Figure 26) of 5 m (using SMB Bulgin Buccaneer waterproof connectors with a RG174 cable). SARDINE has the same communication system (but with a Wifi access point D-Link DAP-1160 with 2 Ethernet ports). Figure 26: Wifi communication at high depth. Additionally, SAUC ISSE has a waterproof Ethernet port on its rear cover (using a Switchcraft 8 pins waterproof connector) which enables to connect an Ethernet cable of 30 m to a laptop near the pool. The Remote Desktop Connection program integrated in Windows XP is used to access to the desktop of the embedded computer via Wifi or Ethernet (Figure 27). When connected, it is possible to launch any program on the embedded computer, check if all the devices are connected and interface completely with the computer onboard. The advantage is that the network does not have to be handled by the embedded program made for the robots, it is already handled by Windows (this could also be done on Linux, using RDP, VNC or X server via SSH ). Figure 27: Remote Desktop Connection client of Windows 15

16 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Power supply The power supply of SAUC ISSE is divided into 2 parts (Figure 28): The engines are powered by a 12 V battery. The PC/104, the sonar, the hydrophones (via a small XLR compatible power supply card, Figure 29), the Wifi router (via the 5 V output provided by the power supply PC/104 module) and the analog cameras (via the 12 V output provided by the power supply PC/104 module) are powered by a 24 V battery (or 2 serial 12 V batteries). Figure 28 : Ni-MH batteries Figure 29 : Phantom power supply for a hydrophone All the other devices (pressure sensor, IMU, Labjack, webcams ) are powered via the 5 V from the USB ports of the computer. SARDINE has only one 12 V battery to power its thrusters, all the other devices are powered by the integrated battery of the EeePC, via the 5 V of its USB ports. Note that contrary to what could be thought, it is the embedded computer and its devices that are the most power-consuming and not the actuators (thrusters). 16

17 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E IV. Autonomy and mission planning Overview: We have several different approaches to try and accomplish the competition tasks: some are interesting because they are simple to find and implement, some are a good compromise between simplicity, reliability and accuracy, some are useful in case a sensor is not available for any reason (hardware failure, perturbations...), other are challenging and have an academic research interest. From the point of view of the control part of the submarine, we can consider all the competition tasks as a succession and combination between depth, orientation and distance regulations problems that change over time. Depending on the task, these regulations will be used with respect to a coordinate space (waypoints) or to an object (mid-water target detection ). As the roll and the pitch of the submarine should always remain stable by design, the localization problem of the submarine in the basin can be considered as planar. The studied coordinate space is defined by the South (y, vertical axis) and East (x, horizontal axis) walls of the basin. The robot can: Be teleoperated using different regulations or direct control algorithms: no regulation (open loop), depth, heading, speed, distance regulations Make sequences of predefined missions from scripts (text files containing keywords with parameters): successions and combinations of regulations, call to particular functions such as activation of the waypoint following with dynamic localization, wall, acoustic pinger, yellow pipeline or orange ball following In some cases, the input of the regulations can be directly the output of sensors, in other cases algorithms must be used to process the sensors data and provide a right input. Here is what our sensors provide (Figure 30): Pressure sensor: the depth with respect to the surface. 360 degrees rotating sonar: an image of all that is around the robot, up to a range of 100 m. With this device, we easily get the distance to the first obstacle at a specific angle. IMU: the 3D angles of the robot, the rotating speeds (using inertial sensors), the accelerations and the magnetic field (that should indicate the direction to the North if there are no perturbations) in 3D. As the basin does not move, we easily get the orientation of the submarine with respect to the basin (and therefore the orientation of our submarine in the coordinate space we defined). But the accelerations are not really usable in our case (the submarines have weak accelerations). Bottom and front cameras: pictures of the sea floor or in front of the robot. Front-right and front-left hydrophones: distance and angle from the pinger. 17

18 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Figure 30: Localization and detection Additional sensors could have been useful: Doppler Velocity Loch (DVL): to get the speed and altitude with respect to the sea floor. Drawbacks: too expensive, might be too big and heavy. Front sonar: to get an image of the sea floor in front of the robot, to follow the pipeline for example. Drawbacks: too expensive, might be too big and heavy. Lateral sonar: to get an image of the sea floor in the left or right of the robot. Drawbacks: too expensive, might be too big and heavy. 18

19 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Processing algorithms: Overview Several processing algorithms were also made available as inputs for the regulations: Camera image processing: o Color selection [1]. o Simple shapes detection (lines, circles, rectangles...). o Movement detection (if the submarine does not move and we know the target object changes over time) by comparing successive pictures. o Interest point detection (can be a line, an object, a part of object or anything that could be singular and detectable in several pictures). For example, we can get the depth if we see the pipeline with the bottom camera and know its size and the characteristics of the camera. We could also try to keep a specific height, angle or distance with respect to the mid-water target by trying to keep it in the center of the front webcam picture and measure its width on the picture. Sonar image processing: o Ball, pipeline, wall and corner detection. For example, we can get the depth if we know that the size, the presence of objects or the distance between objects depends on the depth of the submarine (for example the pipeline should be visible with our sonar only when the submarine is close to the sea floor). We could also try to keep a specific angle and distance with respect to the mid-water target. Static localization: o Robust static localization using interval arithmetic ([2], [3], [4]): it takes the dimension, the shape of the basin and any other singular feature (as a list of segments and circles) as input and uses the sonar, and optionally the IMU, to get the position of the submarine in the basin. Any object detected by the sonar that does not correspond to the input is considered as outlier. Dynamic localization: o Open loop on the control of the thrusters: taking into account experiments done on acceleration time at start, average speed and deceleration time, we can evaluate the distance covered by the robot from one point to another. o Robust dynamic localization using interval arithmetic: it takes the dimension, the shape of the basin and any other singular feature (as a list of segments and circles) as well as a dynamics model of the movement of the submarine as input and uses the sonar and optionally the IMU to get the position of the submarine in the basin. Any object detected by the sonar that does not correspond to the input is considered as outlier. A sonar image processing algorithm (Figure 31) launched in the beginning when the submarine is stopped could automatically discover the environment of the robot (and therefore the shape and dimensions of the basin), assuming that the sonar can see some interest points (like the walls of the basin). This would be a kind of SLAM (Simultaneous 19

20 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Localization and Mapping) if it is used as input of the localization algorithms (that needs the dimension, the shape of the basin and any other singular feature) moreover if the environment discovery step is repeated several times to improve the trajectory estimation. Figure 31: Matching between sonar pings and the basin borders Some algorithms can provide different values for the same data (depth, orientation in the coordinate space, angle with respect to the mid-water target, distance to a waypoint in the coordinate space, distance to the mid-water target...) at the same time. For example, the camera image processing and sonar image processing algorithms can return different distances to the mid-water target. To handle that, all of the results of the sensors and algorithms are returned as intervals ([5]). A simple intersection between the results that evaluate the same data leads to 1 interval (more precise) for each data. The center of the interval is taken as input for the corresponding regulation. Set-membership methods and outliers Consider the following equations: Solving this system of equations translates to find the set of all points that satisfies all the equations i.e.: Set-membership methods such as interval analysis enable to manipulate sets ([5], [6]). For example, these methods enable to compute intersection of sets, union, make set inversion, compute the image set of a function To solve our system of equations, we will compute solutions subsets for each equation. For each equation we have: 20

21 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E be: Because we want all the points that are in the solutions subsets, the final solution will If we have outliers (erroneous data), some equations of the system of equation we want to solve will not be satisfied because they use false data. Therefore, we need a way to avoid using these false equations, but the problem is that we do not know which of them are false. The normal resolution which would compute the intersection between all solutions subsets will get an empty set of solutions. This would be normal since there are no points that satisfy all the equations because some of them are false. Set-membership methods give a solution with another type of intersection called the q- relaxed intersection defined by: Figure 32 shows an example with 5 sets. In this example we have: Figure 32 : Example of relaxed intersection with 5 sets 21

22 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E The dark gray part Note that other methods exist also to handle outliers (least squares ). Interval arithmetic An interval is a closed and connected subset of R. For example, [-1,4], [-oo,2],[- oo,oo] (Figure 33) are considered as intervals but ]1,3[ or [1,3] union [5,8] are not. In a robotic context, if a compass is known (according to its documentation for example) to have an accuracy of 2 degrees, when it indicates 45 degrees we will consider in all the computations the interval [45-2,45+2]. Figure 33: Examples of intervals An arithmetic can be made with intervals: Operations {+,-,*,/}: [x-,x+] op [y-,y+] = smallest interval containing the set of all possible values [-1,4]+[2,3]=[1,7] [-1,4]*[2,3]=[-3,12] [-1,4]/[2,3]=[-1/2,2] Multiplication by a real number, intersection, union 2[-1,4]=[-2,8] [-1,3] inter [2,4]=[2,3] [-1,2] union [3,4]=[-1,4] Image by a function sin([0,pi])=[0,1] 22

23 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Figure 34: Box representing sin([0,pi])=[0,1] There are algorithms and techniques that enable to solve equations using intervals: Contraction and propagation: If we know that z=x+y and z=[-1,1], x=[0,1], y=[0,1], therefore z=x+y=[0,1]+[0,1]=[0,2]. Because z=[-1,1] at first, we have z=([-1,1] inter [0,2])=[0,1]. Then, y=z-x Additional techniques are also available (bisections with the SIVIA algorithm for the set inversion [7] ) Localization The evolution equations of the submarine in the plane can be the following: x y v cos v sin u 2 u 1 v u 1 u 2 v where x,y are the coordinates of the robot, its orientation and v its speed. The inputs u 1 and u 2 are the accelerations (or forces) provided to the right and left thrusters. This normalized model (i.e. with all coefficients equal to 1) corresponds to a submarine robot at constant depth (depth regulation is independent from the movement in the plane) without roll or pitch. The system can be discretized as follows: xk 1 f k xk, uk where is the sampling time, x x,y,, v is the state vector, u u 1, u 2 are the inputs and 23

24 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E x 1 x 1. x 4. cosx 3 f k x 2 x 3, u 1 u 2 x 2. x 4.sinx 3 x 3.u 2 k.u 1 k x 4 x 4.u 1 k. u 2 k. x 4 We suppose that the robot moves in a basin of known shape (localization). The sonar enables to measure the horizontal distance between the robot and the basin borders in the direction pointed by the current sonar angle (this direction changes over the time because it is a rotating sonar). If the pool is composed of vertical borders, the observation equation of the system is yk g k xk, u with y d,, and g k x 1 x 2 x 3 x 4, u 1 u 2 g 1,k x where g 1,k is a function given by a simple algorithm of distance to segments computations. One of the most difficult tasks for the submarine is to localize in the basin while moving. Most of the existing solutions to solve that problem are based on probabilistic techniques (Kalman filters, particles filters [12][13] ). In our submarine, a state observer using interval arithmetic and taking into account outliers from the sonar enables the robot to estimate its position. Moreover, an original method using an image contractor (see [8] for more information on contractors and [9] for the image contractor) and a notion of accumulators which are represented by set valued polynomials (polynomials with intervals coefficients, see [10]) was elaborated and tested on SAUC ISSE in the ENSTA Bretagne pool. 24

25 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Figure 35: Dynamic localization with outliers using accumulators: the darkest box corresponds to the one that is compatible with the most of sonar measurements Image processing The cameras and the sonar can provide images that can be processed to detect objects, get movement or position information Several algorithms were made: Object detection using color [1] (Figure 36). Some objects can be characterized by their color. However, colors are altered in water: red is more absorbed than the blue when the distance to the object in the water increases. These modifications can be described by a simple formula, which gives the color that should be viewed in the water depending on the distance between the camera and the object in the water, knowing its color in the air. Some coefficients in the formula (red, green and blue absorption coefficients) must be measured by experiments because they depend on the water, current light... Simple shapes detections (lines, circles, rectangles ) with for example the normal Hough transform and its interval equivalent ([11]). Movement detection: if the submarine does not move and we know that the target object moves, it can be possible to compare successive images to see what changed. Interest points detection (lines, objects, parts of object or any singular shape that can be detected in consecutive images) to obtain the robot speed by looking at the floor movement on the pictures (optical flow) with a bottom camera and knowing its altitude. 25

26 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Figure 36 : Object detection using color selection taking into account color absorption in water depending on the distance to the object 26

27 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E Control: Regulations Due to the very slow dynamics of the robot, the depth regulation algorithm is very simple: it is a three state controller. If the submarine is below the desired depth, the vertical thruster is turned on at its maximal speed. If the submarine is near the desired depth, it is off. If the submarine is above, the thruster is turned on in the other direction. The vertical thruster control is independent from the 2 horizontal thrusters. The orientation and speed regulation is based on a PID (Proportional Integral Derivate) controller that uses the heading to the North and the rotation speed obtained by the IMU, and the distance regulation is based on time. We used experimental methods to set the coefficients of theses regulations. The principle of the waypoint following is described in the following figure (Figure 37): Figure 37 : Control of the submarine for the waypoint following First, the mission planner sends to the controller a waypoint x w, y w to reach. When the current waypoint is reached with a predefined precision (i.e. x x w 2 ŷy w 2 ), the mission planner goes to the next waypoint. The controller chosen is given by the following expression: u 1 1, where sign det cos x w x sin y w ŷ The direction that the robot should follow is given by the vector e x w x, y w ŷ T. The estimated orientation of the robot is given by the vector v cos, sin T. If v is on the right of e (i.e. detv,e 0 ), the robot turns right ( 1 ), otherwise it turns left 1 ). Autonomy High level actions such as moving to a waypoint, following a trajectory, searching for an object, following the borders of the basin can be done by executing the processing algorithms followed by the basic movements algorithms depending on results. These are 27

28 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E implemented in the embedded program in C/C++. However, the sequence of these actions is specified by a script (Figure 38). This enables to quickly and easily change the missions that the robot must do and moreover, it can be understood and modified easily by anyone. Figure 38 : Script of autonomous missions 28

29 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E V. Innovation This year, the main innovation we would like to show during the SAUC-E competition is the idea of using several collaborating submarines as well as a surface vehicle to accomplish tasks. Additionally, if we manage to successfully use our localization algorithm (static and dynamic) that uses interval methods to compute the position of the submarine in a robust manner during the competition, it would be a good contribution to show that interval arithmetic can be successfully used in real autonomous submarine robots. 29

30 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E VI. Financial summary We received this year for the project. Products Prices ( ) Purpose PC/104 Eurotech ISIS 2000 Spare embedded computer 2 Analog webcams ALLWAN AL- 900 Spares waterproof cameras 2121 Hydrophones 1000 SEABOTIX thrusters BTD Spares in case of problem IP68 waterproof connectors 1000 AHRS CH Robotics 200 For tests D-Link wireless routers 500 Batteries 1000 Servo controllers Robbe Rokraft 600 Spares in case of problem Other 5000 Trip

31 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E VII. Risk assessment Risk Loss of control Recovery needed Sharp edges Frontal collision with a wall Electric shock Thrusters hazard Precaution Power switch, positively buoyant Central lifting cord Visible colors and diving gloves Bumper in front of the frontal webcam Only low voltages and intensities Propeller protected 31

32 SAUC ISSE and SARDINE, 2 AUVs for SAUC-E VIII. References [1] S. Bazeille, Vision sous-marine monoculaire pour la reconnaissance d'objets, Ph.D. Thesis, Université de Bretagne Occidentale, 2008, and other related work, [2] L. Jaulin, Robust set membership state estimation, Automatica, Volume 45, Issue 1, January 2009, Pages , [3] J. Sliwka, F. Le Bars, O. Reynet, L. Jaulin, Reconnaissance de forme pour la localisation de robots, submitted to RFIA 2010, 2009, France, [4] J. Sliwka, F. Le Bars and L. Jaulin, Calcul ensembliste pour la localisation et la cartographie robustes, JD-JN-MACS, Angers, [5] L. Jaulin, M. Kieffer, O. Didrit and E. Walter, Applied Interval Analysis with Examples in Parameter and State Estimation, Robust Control and Robotics, Springer-Verlag, 2001, ISBN: , [6] R. E. Moore. Methods and Applications of Interval Analysis. SIAM, Philadelphia, PA, [7] L. Jaulin and E. Walter. Set inversion via interval analysis for nonlinear boundederror estiamation. Automatica, 29(4): , [8] G. Chabert and L. Jaulin, QUIMPER, A Language for Quick Interval Modelling and Programming in a Bounded-Error Context, Artificial Intelligence, 173: , [9] J. Sliwka, F. Le Bars, O. Reynet, and L. Jaulin. Using interval methods in the context of robust localization of underwater robots. In NAFIPS 2011, El Paso, Texas, USA, [10] J. Sliwka. Set valued polynomials and their application to robust localization of an underwater robot. In SWIM 2011, Bourges, France, [11] L. Jaulin and S. Bazeille. Image shape extraction using interval methods. In Sysid 2009, [12] S.Thrun, W. Bugard and D. Fox, Probabilistic Robotics, MIT Press, Cambridge, M.A., United Kingdom, [13] D. Ribas, P. Ridao, J.D. Tardós, and J. Neira. Underwater SLAM in man made structured environments. Journal of Field Robotics, Accepted for publication,