Projects

Autonomous Wide Aperture Cluster for Surveillance (AWACS)

This project employs of a cluster of Autonomous Underwater Vehicles (AUVs) for adaptive Detection Classification Localization (DCL) of quiet targets. The concept for Autonomous Wide Aperture Cluster for Surveillance (AWACS) is comprised of (at least) three REMUS vehicles, each one sensing the local environment and a “master” (gateway) vehicle operating at the surface equipped with GPS and an RF link. The onboard sensors sample oceanographic, bottom and acoustic features and the three vehicles communicate via AComms LAN.

In the first stage of the proposed work, adaptive sampling and search algorithms will be developed to optimize the probability of detection (PD), minimize probability of false alarm (PFA) and provide glimpses of the target. Deployment of the system will be based on acoustic models of transmission loss and ambient noise derived by assimilating environmental data gathered using multiple AUV’s with numerical ocean models. Vehicle command and control algorithms will be developed to coordinate the individual and collective sampling of the AUVs within the cluster.

In the second stage of the proposed work, adaptive DCL concepts will be employed and tested against a surrogate target. Duke University’s concept for Dynamically Reconfigurable Sensor Arrays (DRSA) and other advanced adaptive signal processing will be explored.

The sensors to be installed on the REMUS vehicles will consist of conventional CTDs and ADCPs as currently deployed on REMUS for oceanographic measurements, and passive acoustic arrays for DCL. Two array types will be explored, a lower frequency towed array and a higher frequency hull-mounted array. The intent is to develop a fundamental understanding of the issues involved with using REMUS towed and hull-mounted arrays for the detection of quiet targets. Issues to be studied include drag and endurance as functions of array aperture and length, radiated noise effects on side lobes, vibration effects on hull-mounted arrays, array data acquisition and processing requirements, element location issues, handling, logistics, and durability, among others. Having developed this understanding, we will then be able to make important comparisons between each array configuration and assess their benefits and drawbacks. This experience will be used in selecting the acoustic array most appropriate for installation on future AUVs.

One aspect of this project will be the development of an autonomous surrogate target to both realistically simulate potential threats during DCL testing and serve as a mobile source for in situ TL measurement. The EMATT vehicle will be adapted to fill this role. Critical issues to be addressed with this source vehicle are source level versus size, tracking, command and control via the AComms LAN, waveform transmissions to simulate surrogate targets of interest for DCL testing, and recovering versus scuttling the vehicle.

A key issue is to optimally control the network of AUVs and their associated communication and sensor systems such that the Adaptive Search and Adaptive DCL algorithms can be pursued.

The proposed project will leverage a rigorous testing program in which AWACS components and algorithms will be tested, modified based on test results, and then retested in a systematic iterative development strategy.

ProSapien's contribution is to explore and develop vehicle command and control based on environmental and acoustics sensing

In order to function efficiently and effectively, the AWACS network of AUV's and their associated communication and sensor systems must be optimally managed and controlled in order to detect, classify and localize a quiet target in the complex shallow water environment. This optimization is a multiple variable time and resource constrained decision-making problem. Over the course of the project, ProSaipien will design, simulate, implement, integrate, and at-sea test algorithms that govern the time and space behavior of the vehicles and optimize their combined sensing capabilities. The initial baseline architecture will be derived from Dr. Smith’s work on weightless reconfigurable agents for intelligent distributed autonomy (WRADIA), utilizing pattern centric reasoning, and trajectory-based optimization, as well as data and task fusion. There has been extensive work done on multiple vehicle coordination and control to date, where the various strategies may be grouped into three categories: behavior-based methods, leader-following, and virtual structure approaches.

In a behavior-based approaches, several “behaviors” are prescribed for each vehicle and the resultant control is derived from a weighted combination of behaviors that can be varied over time. The drawback of behavior-based approaches is that they are difficult to analyze mathematically. Thus, it is difficult to guarantee a certain level of performance. This is especially problematic for ASW related tasks because signal processing gain is a critical parameter, and one that would be difficult to calculate under behavior-based control. The advantages of behavior-based approaches are that they are decentralized and make minimal demands on inter-agent communication.

In a leader-follower approach, the full state of the leader is communicated to all the followers who execute a “formation” relative to the leader. The advantage is that this allows both mathematical analysis and performance guarantees. The drawbacks are that the leader follower approach is less robust to failures in communications and loss of members and does not easily allow coordinated non-homogeneous behavior of the vehicles.

In a virtual structure approach, the state of the system is communicated to each of the members in a peer-to-peer or ad-hoc manner, where each vehicle broadcasts its local state information, and relays its neighbors’ state as well. Because each vehicle eventually obtains a global view of the system state, virtual structure approaches enable mathematically analyzable performance. Unlike leader-follower, however, the vehicles can easily engage in coordinated non-homogeneous behavior. Moreover, unlike the leader-follower approach where communication tends to be top down, the peer-to-peer communication of system state and coordination variables is more robust against failures in communications or loss of members.

For the AWACS program, a hybrid virtual structure and behavior-based approach is contemplated. In this approach, virtual structure coordination is used when sufficient conditions are met for reliable vehicle communications, including proximity, and latency. When those conditions are not met, the system reverts to a behavior-based approach that would attempt to repair the insufficiency condition. Recent work at BYU with multiple unmanned air vehicles has demonstrated the efficacy of a hybrid behavioral and virtual structure approach .The AWACS program will extend and adapt the hybrid approach to the multiple underwater vehicles.

Multi-sensor and multi-task fusion are important aspects of the command and control task, and key factors in the success of the virtual structure approach. We propose to use a generalized approach that subsumes many of the common techniques used for task and sensor fusion, referred to as Symmetric Multi-objective decision making with Elastic Constraint propagation (SMEC). SMEC generalizes decision-making and naturally accommodates incomplete information and natural language expressions in the decision making process. When coupled with a learning mechanism such as genetic algorithms, neural networks, or case based reasoning, advanced deliberative mapping and planning tasks can be implemented and adapted through experience in a robust manner. It is a unifying simplifying technique that provides a meta-framework for intelligent decision-making and cognitive processing.

The core tasks for autonomy include both sensor and task fusion. These are a form of multi-objective decision-making. Different methods for decision making are based on different paradigms and methodologies for handling information including probabilistic methods such as Kalman filters and Bayesian estimators or learning methods, such as artificial neural networks. We believe the most effective approach will be to use a meta-framework upon which various approaches can be applied as the situation warrants. This meta-framework is based on the concept of elastic constraint propagation. All other forms of decision-making logic can be expressed as special cases where the degree of elasticity and the semantics of the constraints are varied.The overriding philosophy behind this approach is that the interaction between components of the intelligent agent sensing, command, and control system is performed through the propagation of elastic constraints. Inherent in each task is the satisfaction of event and time-based constraints on the behavior of the system. Natural constraints are imposed by the environment and platform. Artificial constraints are imposed by the goals and the missions to be performed. Communication between various components of the system can be represented as the transmission of constraints.

Conventional techniques use crisp or Boolean constraints. These can create problems such as rapid switching between decisions. The system may be brittle to changes in constraint boundaries, thereby making their precise specification critical. Moreover, the simultaneous satisfaction of multiple constraints may become problematic due to contradictions that arise as the number of constraints increases. Hard constraints are often specified arbitrarily.

Elastic constraints avoid these problems through partial satisfaction. The degree of satisfaction of an elastic constraint is an inverse function of how far the constraint must be stretched to accommodate the given situation. Although it may be impossible to simultaneously satisfy every constraint completely, given enough elasticity, it may be possible to simultaneously satisfy every constraint to some nonzero degree. Decision-making consists of ordering the space of decision alternatives based on the degree of constraint satisfaction. With elastic boundaries, the system is more robust to parameter variations. Elastic constraints better represent the real world where constraint boundaries are often flexible depending on the circumstances.

Given a state of nature, a set of decision alternatives, a set of constraints, and a set of objective functions, conventional decision-making consists of finding the alternative that maximizes the objective functions while satisfying the constraints. This model of decision-making is not symmetrical in the sense that constraints (expressed as crisp sets) and goals (expressed as objective functions) have different representations. The fuzzy model of decision-making represents both the constraints and goals as elastic constraints represented by fuzzy sets. Decision-making consists of finding the confluence of goals and constraints by aggregating the fuzzy sets.

Aggregation operator(s) are selected based on the type of decision being made. Different aggregation operators might be used for different constraints on the same criteria space and also for constraints on different criteria spaces. For example, goals and objectives might be aggregated in a different manner than other constraints. Many different aggregation operators have been developed that provide great flexibility in tailoring the decision-making process to a given problem. Moreover because the model is symmetric the same approach works for both sensor fusion and task fusion. Based on the fuzzy symmetric decision model a generic architecture for controlling agent tasks can be implemented as a tree of arbiters and behaviors, as highlighted in Figure 10. Coordination of controllers is equivalent to constraint aggregation. Depending on the relations between successive criteria/decision spaces competitive, cooperative, and hierarchical arrangements can be implemented.

References:

[1] Feijun Song, and Samuel M. Smith, “Autonomous Underwater Vehicle Control Using Fuzzy Logic” Invited Chapter, Intelligent Control Systems Using Soft Computing Methodologies. CRC Press 2001. pp. 243-260

[2] Feijun Song and Samuel M. Smith, "Cell-State-Space-Based Search," IEEE Control System Magazine. August 2002, pp 42-56.

[3] Smith, S.M.; An, P.E.; Holappa, K.; Whitney, J.; Burns, A.; Nelson, K.; Heatzig, E.; Kempfe, O.; Kronen, D.; Pantelakis, T.; Henderson, E.; Font, G.; Dunn, R.; Dunn, S.E, “The Morpheus Ultra Modular Autonomous Underwater Vehicle.” IEEE Journal of Ocean Engineering, October 2001 vol 26 # 4, Page(s): 453 -4651.

[4] Grenon, G.; An, P.E.; Smith, S.M.; Healey, A.J. “Enhancement of the inertial navigation system for the Morpheus autonomous underwater vehicles”, IEEE Journal of Ocean Engineering, October 2001 vol 26 # 4, Page(s): 548 -560

[5] Edgar An, Manhar Dhanak, Lynn Shay, Samuel Smith, John Van Leer, “Coastal Oceanography Using a Small AUV”, Transactions of the American Meteorological Society. February 2001 pp 215-234

[6] A. Neel, L. LeBlanc, J. Park, S.M. Smith, “Peer to Peer Communications Protocol,” Sea Technology, pp 10-15, May 1998

[7] S.M. Smith, K. Ganesan, P.E. An,S.E. Dunn, “Strategies for Simultaneous Multiple AUV Operation and Control,” International Journal of Systems Science, Vol. 29 No. 10, 1998, pp 1045 -1063

[8] WRADIA Next-Generation, “Unifying Architecture for Intelligent Agents in Automation”, Samuel M. Smith Ph.D, Adept Systems Incorporated White Paper 26 October, 2004.

[9] Walter Tucker, Brian Callahan, Samuel Smith, Adept Systems, Inc., “Application Of Network Fragment Healing Technology To A Reconfigurable Electrical Power System”; 13th SCSS (Ship Control Systems Symposium), Orlando Florida, April 7-9 2003.

[10] Kenneth Lively, SYNTEK Technologies, Inc., USA Donald Dalessandro, Naval Surface Warfare Center/ Carderock Division, USA Samuel Smith, PhD, Adept Systems, Inc., USA, “Complexity Management in Shipboard Automation Architectures Employing Component Level Intelligence”. 13th SCSS (Ship Control Systems Symposium), Orlando Florida, April 7-9 2003.

[11] “Trajectory tracking for unmanned air vehicles with velocity and heading rate constraints”, Wei Ren; Beard, R.W.; Control Systems Technology, IEEE Transactions on,Volume: 12 , Issue: 5 , Sept. 2004 Pages:706 - 716

[12] “Coordinated target assignment and intercept for unmanned air vehicles”, Beard, R.W.; McLain, T.W.; Goodrich, M.A.; Anderson, E.P.; Robotics and Automation, IEEE Transactions on ,Volume: 18 , Issue: 6 , Dec. 2002 Pages:911 - 922

[13] “A coordination architecture for spacecraft formation control”, Beard, R.W.; Lawton, J.; Hadaegh, F.Y.; Control Systems Technology, IEEE Transactions on ,Volume: 9 , Issue: 6 , Nov. 2001 Pages:777 - 790

[14] “A decentralized scheme for spacecraft formation flying via the virtual structure approach”, Wei Ren; Beard, R.W.; American Control Conference, 2003. Proceedings of the 2003, Volume: 2 , June 4-6, 2003 Pages:1746 - 1751

AWACS Team Members

ProSapien

OASIS

Woods Hole Oceanographic Institution

Boston University

Naval Postgraduate School

Harvard University

Duke University