Information Sciences 145 (2002) 6987 www.elsevier.

com/locate/ins

A behavior-based architecture for autonomous underwater exploration

Abstract We present a system for behavior-based control of an autonomous underwater vehicle for the purpose of inspection of coral reefs, a task currently performed by divers holding a video camera while following a rope. Using sonar and vision-based approaches, behaviors have been developed for guiding the robot along its intended course, for maintaining a constant height above the sea oor, and for avoiding obstacles. A task-level controller selects which behaviors should be active according to user-dened plans and in response to system failures. Behavior arbitration has been implemented using both fuzzy logic and utility fusion. Initial experiments have been conducted in a natural coastal inlet, and the system is to be soon demonstrated in the coral reef environment. 2002 Elsevier Science Inc. All rights reserved.

1. Introduction In unstructured, unknown, and dynamic environments, such as those encountered by outdoor mobile robots, an intelligent agent must address the issues of incomplete and inaccurate knowledge; it must be able to handle uncertainty in both its sensed and a priori information, in the current state of the agent itself, as well as in the eects of the agents actions. Autonomous underwater vehicles have the potential to play a substantial role in ecology management, geophysical surveying, oshore oil and gas exploration and well
Corresponding author. E-mail addresses: [email protected] (J. Rosenblatt), [email protected] (S. Williams), [email protected] (H. Durrant-Whyte). 0020-0255/02/$ - see front matter 2002 Elsevier Science Inc. All rights reserved. PII: S 0 0 2 0 - 0 2 5 5 ( 0 2 ) 0 0 2 2 4 - 4
*

70

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

maintenance, undersea mineral exploration and mining, and in surveillance and defence. However, a subsea environment is particularly unstructured and dynamic, the degrees of freedom in the control and estimation problem are greater than on land, and there exists no reliable positioning data. To function eectively in such conditions, an autonomous system must be responsive to its environment, proceeding in a data-driven manner, as well as goal-oriented, taking into account the higher-level goals of the system. When used appropriately, deliberative planning and reactive control complement each other and compensate for each others deciencies. In order to achieve this desired symbiosis of deliberative and reactive elements, the distributed architecture for mobile navigation (DAMN) consists of a group of distributed task-achieving modules, or behaviors, communicating with a centralized command arbiter [8]. Within this framework, we have developed an architecture for behaviorbased control of an autonomous underwater vehicle for the purpose of inspection of coral reefs. A survey project for reef management, carried out by the Australian Institute of Marine Sciences, is designed to provide long-term quantitative data about corals, algae and marine life over the extent of the Great Barrier Reef. This data are for studies of abundance and population change in selected organisms on a large geographic scale. Currently, visual transect information is recorded using underwater video cameras held by a diver following a rope, as shown in Fig. 1. The reef surveillance task, as it is currently dened, consists primarily of following an assigned path while maintaining a xed altitude above the reef and avoiding collisions [13]. Independent behaviors and arbiters, using decoupled controllers, have been developed as a modular means of accomplishing these various sub-tasks. For example, two behaviors have been developed for the path following aspect of the task; the rst behavior uses video input to track a rope laid out along the coral, while the second behavior uses sonar to detect passive beacons. The DAMN arbiters are then responsible for combining the behaviors votes to generate controller commands. Fuzzy logic arbiters are currently used for control of the vehicle; another set of behaviors and arbiters that perform utility fusion [9] are under development. In both cases, the distributed, asynchronous behaviors provide real-time responsiveness to the environment, while the centralized command arbitration provides a framework capable of producing coherent behavior. A task-level controller selects which behaviors are active at any given moment.

2. Oberon submersible vehicle We have constructed a simple low-cost underwater robot named Oberon, shown in Fig. 2, as a test-bed for experimental work in autonomous undersea navigation. There are currently ve thrusters on the vehicle. Three of these are oriented in the vertical direction while the remaining two are di-

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

71

Fig. 1. Diver taking visual transect of the reef.

rected horizontally. This gives the vehicle the ability to move itself up and down, control its yaw, pitch and roll, and move forwards and backwards. This thruster conguration does not allow the vehicle to move sideways but this does not pose a problem for the missions envisaged for this vehicle. The submersible vehicle is connected by tether to a ground base station, which provides power to the robot. The tether also provides communications between the o-board computers at the surface and those on-board the vehicle, and includes a coaxial cable for transmitting video data, as well as serial lines. The on-board computer is directly interfaced to the vehicle hardware and is used to control the motion of the robot and to acquire sensor data. There is also a leak detection circuit that shuts o power to the vehicle if water is detected inside the pressure hulls using a pair of optical diodes. 2.1. Internal sensors Gyroscope. An Andrews Fiber Optic Gyro has been included in the Oberon robot to allow the robots orientation to be determined. This sensor provides the

72

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

Fig. 2. Oberon submersible vehicle is propelled by 5 thrusters and is equipped with a sonar on top and in front and a color CCD camera. The on-board processor communicates with o-board computers via a tethered cable, which also supplies power.

yaw rate and is used to control the heading of the submersible. The bias in the gyroscope is rst estimated while the vehicle is stationary. The bias compensated yaw rate is then integrated to provide an estimate of vehicle heading. Because the missions described here were less than 30 minutes, the yaw drift caused by the integration of a noisy signal did not pose a signicant problem. Compass. An integrated compass and tilt sensor has recently been added to the vehicle. The compass signal is ltered with the output of the gyroscope to estimate the yaw rate bias of the gyroscope on-line. Pressure sensor. A bathyometric pressure sensor measures the external pressure experienced by the vehicle and is used to measure depth. This sensor provides a voltage signal proportional to the pressure and is sampled by an analogue to digital converter on the embedded controller. Feedback from this sensor is used to control the depth and altitude of the submersible. 2.2. External sensors Sonars. There are two scanning low frequency terrain-aiding sonars that serve as the main sensors for the robot. The rst is a Tritech SeaKing imaging sonar, which has a dual frequency narrow beam sonar head. It is mounted on top of the submersible and is used to scan the environment in which the sub-

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

73

mersible is operating. It can achieve 360 scan rates on the order of 0.25 Hz using a pencil beam with a beam angle of 1.8. This narrow beam allows the sonar to accurately discriminate bearing returns to objects in the environment. It has a variable mechanical step size capable of positioning the sonar head to within 0.5 and can achieve range resolution on the order of 50 mm depending on the selected scanning range. It has an eective range of up to 300 m in the low frequency mode, thus allowing for long range target acquisition, but can also be used for high denition scanning at lower ranges. The information returned from this sonar is used to build and maintain a feature map of the environment. The second sonar is an Imagenex sonar unit operating at 675 kHz. It has been mounted at the front of the vehicle and positioned such that its scanning head can be used as a forward and downward looking beam (see Fig. 5). This enables the altitude above the sea oor as well as the proximity of obstacles to be determined using the wide angle beam of the sonar. The sonars beam width is 15:0 1:8, thus covering a broad swath with each ping, making it an ideal sensor for altitude estimation and obstacle avoidance. Because this sonar is used for multiple purposes, a scheduler is used to allocate this resource to the appropriate behaviors. Camera. A color CCD camera mounted in an underwater housing provides a PAL signal via a coaxial cable to an o-board computer, where a Matrox Meteor card is used to digitize the image for further processing.

2.3. Communications Inter-process communication occurs asynchronously via a TCP/IP socketbased interface using a distributed message passing system. Processes register as either producers or consumers of a particular message type. It transparently communicates with processes on the same processor or across the network, without requiring a xed network topology to be dened prior to operation. Messages are sent as a continuous stream of inputs to behaviors and outputs to controllers; a lost message may introduce a slight delay in the response of a behavior to a stimulus, but otherwise operation is unaected. The communications system also logs message data and provides a playback facility which can later be used to send and receive commands, thus providing a simulated vehicle for testing and development purposes. Fig. 3 shows a schematic diagram of the vehicle sensors and their connections. The Oberon robot uses a CompactPCI system running Windows NT that interfaces directly to the hardware and is used to control the motion of the robot and to acquire sensor data. While the Windows operating system does not support hard real-time performance, it is suitable for soft real-time applications and the wide range of development and debugging tools make it an appropriate environment in which to test new navigation algorithms.

74 J. Rosenblatt et al. / Information Sciences 145 (2002) 6987 Fig. 3. Sensor data is collected by the on-board processor and communicated to the surface computers via ethernet and coaxial cables in the tether.

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

75

Time-critical operations, such as sampling of the analog to digital converters, are performed on the hardware devices themselves and use manufacturer supplied device drivers to transfer the data to the appropriate processes. The low-level processes run on the embedded controller and are used to interface directly with the hardware. This allows the controllers to respond quickly to changes in the state of the submersible without being aected by delays due the data processing and high-level control algorithms running on the remote computers. Sensor data are collated by the on-board processor and sent to the surface via the tether using an ethernet connection. The base station contains a network of computers that are used for further data processing, data logging, and providing feedback to the user about the state of the submersible. Set points to the low-level controllers are provided by the behaviors and high-level controllers described in the following section. 3. Behavior-based control architecture In order to avoid the bottlenecks and brittleness of centralized systems, the DAMN consists of a group of distributed specialized task-achieving modules, called behaviors, communicating with centralized command arbiters, as shown in Fig. 4. A behavior encapsulates the perception, planning and task execution capabilities necessary to achieve one specic aspect of robot control, and receives only that data specically required for that task [2]. Each behavior

Fig. 4. DAMN consists of centralized arbitration of votes from distributed behaviors, activated by the task-level controller based on vehicle and behavior status.

76

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

operates independently and asynchronously from the others, and uses domain and procedural knowledge to vote for desirable actions and against objectionable ones, from the perspective of the task for which it is responsible. An arbiter is responsible for combining the behaviors votes to generate actions, which are then sent to the vehicle controller. One type of arbiter used for control of the vehicle employs fuzzy logic [15]; the behaviors send depth or heading commands as fuzzy sets. Another type of arbiter used performs utility fusion; instead of voting directly for actions, behaviors vote for the utility of possible outcomes and their associated uncertainty, and the arbiter determines which action will maximize expected utility. Each of these schemes has their own advantages and disadvantages, and both have been demonstrated to be eective for navigation of land-based vehicles [4,9]. A task-level controller is also employed to select which behaviors should be active according to user-dened plans and in response to system failures, based on knowledge of which behaviors are most relevant and reliable in a given situation. 3.1. Decoupled control Control of an mobile robot in six-dimensional space in an unstructured and dynamic environment would be a highly computationally intensive endeavor, even more so with the diculties in controlling an underwater robot. We instead chose to decouple the control of motion in the horizontal plane from motion along the vertical axis. Thus greatly simplifying the individual design of controllers, behaviors, and arbiters for the two subproblems. The dynamics of the Oberon vehicle are such that the vertical motion of the vehicle is largely decoupled from the lateral motion. The vehicle is very stable in the roll and pitch axes due to the large righting moment induced by the vertical conguration of the pressure vessels. A steel keel provides an added moment to maintain the vehicle in an upright pose. Furthermore, this decoupling of the control problem is reasonable for the types of task we anticipate performing with underwater vehicles. In the reef surveying task, for example, we need to maintain a xed height above the sea oor, independent of the path following behaviors that operate in the horizontal plane. The surveying task can then be made independent of maintaining the vehicle altitude. In addition, we are not interested in explicitly controlling the vehicles roll or pitch, merely minimizing them. Thus, there is one PID controller for the vertical thrusters which servos on depth, and another PID controller for the horizontal thrusters which governs yaw and speed. These controllers can then be designed, tuned, and operated independently of each other. There are also two arbiters in the system, each one corresponding to a separate controller. Likewise, behaviors are separated into those involved in vertical movement along the Z-axis and those involved in horizontal movement in the XY plane, as illustrated in Fig. 4.

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

77

3.2. Behaviors In order to perform the task of surveying a coral reef transect, there are at least three behaviors that must be part of the system: following the transect path, maintaining a certain height above the coral, and avoiding collisions. Because there is no global positioning data and only highly inaccurate internal data available to the robot, we attempt only to maintain position relative to sensed features in the environment. Two methods currently being used are vision-based navigation using a rope laid out along the coral and sonar-based navigation using passive beacons. They each have dierent failure modes and can be used in a complementary manner. Research into the simultaneous localization and mapping is also being conducted [13]. This approach allows the Imagenex sonar to be shared by both the task that maintains the altitude of the sub and by tasks interested in obstacle avoidance. Since the resources of any robot will be limited, it important to share these resources as necessary. A resource scheduler has been implemented that allows resources to be shared between various tasks based on their relative priority and their urgency. Maintain Altitude. In the context of this project, the robot must be able to detect the coral reef and maintain a relatively constant height above it in order to videotape the coral specimens and of course to avoid colliding with them. This behavior provides the ability to maintain the robot at a certain altitude above the sea oor. A sonar is used to periodically determine the altitude of the robot; several pings are taken directly downward and at slight angles in either direction, as shown in Fig. 5(a), and the minimum distance measured is used as a conservative estimate of altitude. The Imagenex sonar is pointed downward to measure height above the bottom. However, because this sensor must also be used for other purposes such as obstacle avoidance, and because of the sonars slow scanning rate, it is not possible to respond to disturbances and maintain stable control using this sensor alone. Thus, the task is accomplished using the sonar in combination with the depth sensor, which is continuously sampling the depth. The dierence between the desired and actual altitudes is used to set a new desired depth for this behavior, as illustrated in Fig. 5(b). Avoid Collisions. When the sonar is pointed forward, its returns are processed to determine if there are any obstacles that may lie ahead, as shown in Fig. 5(b). For the fuzzy arbiter, the behavior would vote against the heading in which the obstacle lies; for the utility fusion arbiter it would indicate a negative utility at the computed obstacle location, along with uncertainty estimates. Follow Rope. The vision-based approach uses camera images in order to detect the survey line by color. A target color is specied, and a distance metric for image pixels is dened as the sum of the dierences between a given pixels RGB values and the target RGB values, normalized by pixel intensity:

78

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

Fig. 5. Sonar-based behaviors: (a) Several altitude measurements are taken; (b) Maintain Altitude behavior then uses minimum altitude measurement to set desired depth, and Avoid Collisions behavior makes forward measurements to avoid obstacles.

Dp

jRt Rp j jGt Gp j jBt Bp j : R p G p Bp

A random sample of pixels are then chosen from the image, and a dynamic programming search [5] is begun from each pixel to nd a least cost path from the top to the bottom of the image, using the color distance metric dened above, as shown in Fig. 6(a).

Fig. 6. Path following sensors: (a) Follow Rope behavior performs visual tracking; (b) Follow Targets behavior uses sonar to detect two targets in natural terrain.

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

79

Fig. 7. Follow Targets behavior calculates desired heading wd based on distance d and heading wd from sub to line between targets.

Once the line has been found in the image, a pursuit algorithm is used to determine which orientation would best bring the robot onto the line and sends that as a vote to the fuzzy arbiter. The location of the line itself can simply be sent to the utility fusion arbiter as a positive utility. Follow Targets. At present, sonar targets are introduced into the environment at the two ends of the transect to be followed. These targets present large, easily detectable features for the sonar, as indicated in Fig. 6(b). By maintaining a x on two of these targets, the robot can nd its current position and orientation relative to the line joining the two targets. Multiple targets can be used, depending on the roughness of the terrain and the length of the transect. The redundant information will also allow a better estimate of the lines position and orientation, thus improving line tracking. Using the range and bearing information to both of the targets to dene the line to be followed, the perpendicular distance from the sub to the line (d) as well as the dierence in orientation between the principle axis of the sub and the line (w) are calculated, as shown in Fig. 7. These two values can then be used to compute a new desired heading wd that will keep the sub on its desired path [14]; this is used to dene the vote sent to the fuzzy arbiter. Again, for the utility fusion arbiter, the line is sent as a positive utility. 3.3. Fuzzy logic arbiter Fuzzy logic has found many uses for mobile robot control, including command fusion systems (see [11] for a survey). For the DAMN fuzzy logic arbiter, each behavior must determine the desired action, in this case either a heading or a depth, and vote for a fuzzy membership set with that action at the peak. Command defuzzication is achieved using the maximum criterion to take the peak value; other defuzzication strategies, such as center of mass, assume a unimodal function and in general this averaging of inputs will select

80

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

Fig. 8. Fuzzy Arbiter combines votes from two behaviors by computing their sum, shown as a solid gray line, and selecting the maximum, shown as a dashed black vertical line. The current direction is shown as a solid black vertical line.

inappropriate commands. A fuzzy vote is specied by four points that may take the form of a triangle, rectangle, or trapezoid. Fig. 8 shows votes from two behaviors. The rst is the Move To behavior, which simply steers the vehicle to a specic set of coordinates; it is voting with a triangle whose apex is at 55. The second behavior is Avoid Collisions, which has a weight of 5; it has detected an obstacle ahead and is voting equally against all headings between 50 and 70. The fuzzy sum of all votes is shown as a solid gray line, and the output direction determined by the arbiter (40) is shown as a dashed black vertical line. The current direction (60) is shown in the gure as a solid black vertical line. 3.4. Utility fusion arbiter Utility fusion is a novel means of action selection in which behaviors determine the utility of possible world states and their associated probabilities, and the arbiter evaluates candidate actions based on that information [9]. Utility theory provides a framework for dening votes and dealing with uncertainty in an explicit fashion using Bayesian probabilities [1]. If assign a utility measure U c for each possible consequence of an action a, then the expected utility U a is X U a U c P c j a; e; 2
c

where P c j a; e is the probability that consequence c will occur given that we observe evidence e and take action a [6]. Thus, if behaviors provide these

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

81

utilities and probabilities to an arbiter, it can then apply the Maximum Expected Utility (MEU) criterion to select a Pareto-optimal action based on all current information [7]. Utility fusion does not create a world model as sensor fusion systems do. The information combined and stored by the utility fusion arbiter does not represent sensed features of the world, but rather the desirability of being in a particular state according to some criterion dened within the behavior. The processing of sensory data is still distributed among behaviors, so the bottlenecks and brittleness associated with sensor fusion are avoided. Unlike command fusion systems, the utility fusion arbiter does not simply select among or combine actions proposed by behaviors. Instead, behaviors indicate the utility of possible world states, together with estimates of uncertainty; the arbiter maintains a map of these utilities, and evaluates candidate actions within it. A more complete overview of various types of architectures and their corresponding advantages and disadvantages can be found in [10]. Another advantage of map-based utility fusion over command fusion is that the dynamics of the system being controlled can be fully modeled and accounted for by the arbiter. The utility arbiter maintains a local map of the utilities and associated uncertainties sent to it by the behaviors. The utilities are represented by geometric constructs (points, lines, and polygons) with associated two-dimensional Gaussian distributions. For the sake of eciency, the command choices to be evaluated are discretized; the results are smoothed and interpolated to avoid problems with bangbang control, i.e., switching between discrete choices to achieve an intermediate value. In the example shown in Fig. 9, the two negative utilities correspond to obstacles detected by the Avoid Obstacles behavior; a line of positive utility is placed between the two sonar targets by the Follow Targets behavior, and the Follow Rope behavior has submitted another line of positive utility corresponding to the rope found in the video image. The arrows emanating from the vehicle represent the candidate actions being evaluated by the arbiter. The one which has maximum expected utility is selected as the current commanded action.

3.5. Task-level controller A task-level controller selects which behaviors should be active according to user-dened plans and in response to system failures, based on knowledge of which behaviors are most appropriate in a given situation. The system programmer denes schemas, parameterized planning constructs that indicate under which circumstances a behavior should run and how various success and failure signals sent by the behavior should be responded to.

82

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

Fig. 9. Utility Fusion Arbiter combines positive utilities from the Follow Rope behavior (shown as a wavy solid line) and the Follow Targets behavior (shown as a straight dashed line) with negative utilities from the Avoid Obstacles behavior (shown as two dark solid octagons); each of the candidate actions (shown as arrows pointing away from the vehicle) is evaluated and the one with the highest expected utility is selected.

Schemas may be invoked within other schemas, thus creating a task tree whose root is the overall mission plan and whose leaves are the behaviors to be executed. Each schema and behavior also has a set of preconditions which must be satised before it becomes active. In Fig. 10, the task-level controller is executing a user-dened mission for surveying a coral reef. The root-level schema, SU R V E Y TR A N S E C T , enables the AV O I D CO L L I S I O N S behavior, as well as the FI N D TR A N S E C T and FO L L O W RE E F sub-schemas; the FO L L O W RE E F schema enables the MA I N T A I N AL T I T U D E behavior and the FO L L O W LI N E sub-schema, which in turn enables the FO L L O W TA R G E T S and the FO L L O W RO P E behaviors. These last two behaviors, when enabled by their parent schema, are active when their respective conditions are true; if the targets are detected by the sonar, then FO L L O W TA R G E T S is active, and if the rope is visible in the video image, then FO L L O W RO P E is active. In the example shown in the gure, FO L L O W TA R G E T S is active and FO L L O W RO P E is inactive, but the reverse could be true, or it could be the case that both behaviors are active at once, and the outputs from the two behaviors would be combined by the arbiter. When neither the targets nor the rope is visible, these behaviors are quiescent, but the conditions for FI N D TR A N S E C T are satised and that schema and its descendants (not shown) become active. At any given moment, the systems state of execution is dened by a hierarchy of active schemas and behaviors, the set of goals that they are trying to satisfy, and the values of various state variables. Thus, unlike TCA [12], the hierarchy is not strictly top-down and the behaviors do not operate in a xed

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

83

Fig. 10. Task Tree: the SU R V E Y TR A N S E C T schema has activated the AV O I D CO L L I S I O N S behavior and the FO L L O W RE E F sub-schema, which enables the MA I N T A I N AL T I T U D E behavior and the FO L L O W LI N E sub-schema, which in turn enables the FO L L O W TA R G E T S and the FO L L O W RO P E behaviors.

sequence, but rather are active only when all pre-conditions are satised. In addition, if one behavior or schema fails to produce the desired result, an alternative method will automatically be selected if available. This task-level controller is somewhat similar to he proposed continuous process extension to RAPS [3], with further capabilities and a more structured syntax. A behavior or schema signals success when it has successfully achieved its goal, otherwise it throws an exception back up to the calling schema when it has failed. This signal is caught by the parent, which can then assert or retract a predicates truth value, eectively invoking some children and causing others to halt. If this schema is unable to handle the signal, it can instead pass the signal up to its parent, which then handles the exception in the same manner. A schema can indicate whether its children should be run in sequence or in parallel. If they are in sequence, then the behaviors and schemas will wait for the previous one to be successful before executing. Those in parallel will execute simultaneously until one or more of the children (as specied) signal completion. 3.6. Complete software system The complete software system for the task of reef surveillance is shown in Fig. 11 below.

84 J. Rosenblatt et al. / Information Sciences 145 (2002) 6987 Fig. 11. Complete software system: raw sensor data is processed by virtual sensors, which provide input to task-specic behaviors, which in turn send votes to control arbiters. Sets of behaviors are activated by the task-level planner as appropriate.

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

85

Fig. 12. Results from Maintain Altitude behavior: (a) plot of altitude vs. time shows that the desired altitude was maintained; (b) depth vs. time as vehicle changed depth to match the prole of the sea oor.

Fig. 13. Path of robot executing Follow Targets behavior is shown as series of ellipses connected by a line, within a map created from sonar data.

86

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

4. Experimental results This section presents some preliminary results from deployment of the vehicle in a natural terrain environment along Sydneys coastline. The rst behavior developed was Maintain Altitude, which keeps the vehicle at a xed stando distance from the ocean oor. In the experimental results showing altitude and depth in Fig. 12, the desired altitude was 1.5 m, which was maintained within a standard deviation of 0.2 m, as can be seen in the rst plot if altitude vs. time. This is despite a rapidly changing bottom prole, as can be seen in the second plot of depth vs. time. Executing the Follow Targets behavior resulted in the path shown in Fig. 13 as a series of covariance ellipses representing 95% certainty in vehicle location. The vehicle was deployed in a natural coastal inlet with sonar targets placed in a line at 10 m intervals. The areas on either side of the vehicle are reef walls detected and mapped from sonar data. 5. Conclusion We have developed a behavior-based system for control of an autonomous underwater vehicle performing a survey of coral reefs. Implemented behaviors provide the ability to avoid collisions, maintain a proper stando distance, and following the transect either using a rope with video or targets with sonar. Command fusion is performed using a fuzzy logic arbiter, and a utility fusion system is under development. A task-level controller selects which behaviors should be active according to user-dened plans and current environmental conditions. Initial experiments conducted in a natural coastal inlet have yielded promising early results, and the system is to be soon demonstrated in the coral reef environment. Acknowledgements The authors would like to thank the other members of the AUV project team at the ACFR, as well as Dr. Hugues Talbot of the CMIS CSIRO, who assisted in developing the rope detection algorithm.

References
[1] J. Berger, Statistical Decision Theory and Bayesian Analysis, second ed., Springer, New York, 1985. [2] R. Brooks, A robust layered control system for a mobile robot, Journal of Robotics and Automation RA-2 (1) (1986) 1423.

J. Rosenblatt et al. / Information Sciences 145 (2002) 6987

87

[3] J. Firby, Task networks for controlling continuous processes, in: Proceedings of the Second International Conference on AI Planning Systems, Chicago, IL, June 1994. [4] D. Langer, J. Rosenblatt, M. Hebert, A behavior-based system for o-road navigation, Journal of Robotics and Automation 10 (6) (1994) 776782. [5] N. Nilsson, Principles of Articial Intelligence, Tioga Publication, Palo Alto, CA, 1980. [6] J. Pearl, Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference, Morgan Kaufmann, Los Altos, CA, 1988. [7] P. Pirjanian, The notion of optimality in behavior-based robotics, Journal of Robotics and Autonomous Systems, 2000 (to appear). [8] J. Rosenblatt, The distributed architecture for mobile navigation, Journal of Experimental and Theoretical Articial Intelligence 9 (2/3) (1997) 339360. [9] J. Rosenblatt, Utility fusion: map-based planning in a behavior-based system, Field and Service Robotics, 1998. [10] J. Rosenblatt, J. Hendler, Architectures for mobile robot control, in: M. Zelkowitz (Ed.), Advances in Computers, vol. 48, Academic Press, London, 1999. [11] A. Saotti, The uses of fuzzy logic in autonomous robotics: a catalogue raisonne, Soft Computing 1 (4) (1997) 180197. [12] R. Simmons, Structured control for autonomous robots, Transactions on Robotics and Automation 10 (1994) 1. [13] S. Williams, D. Dissanayake, H. Durrant-Whyte, Towards terrain-aided navigation for underwater robotics, Advanced Robotics 15 (5) (2001) 533550. [14] J. Yuh, Modeling and control of underwater robotic vehicles, Transactions on Systems, Man and Cybernetics 20 (6) (1990) 14751482. [15] L. Zadeh, Outline of a new approach to the analysis of complex systems and decision processes, Transactions on Systems, Man and Cybernetics 3 (1) (1973).