Military systems are the motivational driver for much of the technology development conducted at applied research laboratories around the world. As the needs of the world's militaries change, so does the focus of this research and development. In this paper, we discuss how the fundamental characteristics of military operations in urban terrain (MOUT) impose requirements and constraints on sensing and reconnaissance. We highlight the importance of a new class of small unmanned aerial vehicles (UAVs) for network-centric military urban operations. We review some of the UAVs that have been developed in recent years, and that are under development, with particular attention to their endurance, portability, performance, payload, and communication capabilities. Selected university testbeds are also briefly noted. Over the last few years there has been considerable research focused on how these small UAVs, both individually and collectively, can operate autonomously in urban environments and help capture and communicate needed information. We discuss some of this research; specific topics covered include guidance and control for autonomous operation, multi-UAV coordination and route optimization, and ad-hoc networking with UAV nodes. A new concept of operations is described that relies on coordination and control of a heterogeneous suite of small UAVs for surveillance and reconnaissance operations in urban terrain

The paper proposes a structural Bayesian network learning in a biological decision-theoretic intelligent agent model to solve a herding problem. The proposed structural learning methods show that an agent can update its world model by changing the structure of its Bayesian network with the data gathered by experience. The structural learning of the Bayesian network is accomplished by implementing a score based greedy search algorithm. The search algorithm is designed heuristically and exhaustively. A complexity analysis for the search algorithms is performed. Intelligent agent software, IntelliAgent, is written to simulate the herding problem with one sheep and one dog.

A herding dog and sheep problem is studied where the agent "dog" is considered the control action for moving the agent "sheep" to a fixed location using the dynamics of their interaction. The problem is solved for the deterministic case using dynamic programming. Proofs are provided for the correctness of the algorithms. The algorithm is analyzed for its complexity. A software package developed for experimentation is described.

Proposes a biological decision-theoretic intelligent agent model to solve a herding problem. The proposed intelligent agent model is designed by combining Bayesian networks and influence diagrams. In our agent design, we used Y. Shoham's (1993) agent-oriented programming paradigm that defines an intelligent agent by its belief, preference and capabilities. Intelligent agent software is written to realize the proposed intelligent agent model. The same software is then used to simulate the herding problem with one sheep and one dog. Simulation results show that the proposed intelligent agent is successful in establishing a goal (herding) and learning other agents' behaviors.

This paper suggests an intelligent controller for an automated vehicle planning its own trajectory based on sensor and communication data. The intelligent controller is designed using the learning stochastic automata theory. Using the data received from on-board sensors, two automata (one for lateral actions, one for longitudinal actions) can learn the best possible action to avoid collisions. The system has the advantage of being able to work in unmodeled stochastic environments, unlike adaptive control methods or expert systems. Simulations for simultaneous lateral and longitudinal control of a vehicle provide encouraging results.

This paper presents a novel approach for coordinating a homogeneous system of mobile robots using implicit communication in the form of broadcasts. The broadcast-based coordination scheme was developed for the Army Ant swarm-a system of small, relatively inexpensive mobile robots that can accomplish complex tasks by cooperating as a team. The primary drawback, however, of the Army Ant system is that the absence of a central supervisor poses difficulty in the coordination and control of the agents. Our coordination scheme provides a global "group dynamic" that controls the actions of each robot using only local interactions. Coordination of the swarm is achieved with signals we call "heartbeats". Each agent broadcasts a unique heartbeat and responds to the collective behavior of all other heartbeats. We generate heartbeats with van der Pol oscillators. In this application, we use the known properties of coupled van der Pol oscillators to create predictable group behavior. Some of the properties and behaviors of coupled van der Pol oscillators are discussed in detail. We emphasize the use of this scheme to allow agents to simultaneously perform an action such as lifting, steering, or changing speed. The results of experiments performed on three actual heartbeat circuits are presented and the behavior of the realized system is compared to simulated results. We also demonstrate the application of the coordination scheme to global speed control.

We suggest an intelligent controller for an automated vehicle to plan its own trajectory based on sensor and communication data received. Our intelligent controller is based on an artificial intelligence technique called learning stochastic automata. The automaton can learn the best possible action to avoid collisions using the data received from on-board sensors. The system has the advantage of being able to work in unmodeled stochastic environments. Simulations for the lateral control of a vehicle using this AI method provides encouraging results.

Coordinated motion of a group of mobile robots bearing a common load is investigated. By extending the results of research with similar multi-robot material handling systems, a supervisory control is derived that provides an optimal distribution of work among the robots while accounting for the robots nonholonomic constraints.>

A technique is developed for the automatic, online generation of trajectories for robotic exploration of unknown objects. For manipulators and end effectors with an arbitrary assortment of sensing apparatus, especially tactile, the Kalman filtering approach is given for the estimation of unknown surface parameters. A cost function is found based on the inverse error covariance matrix for the parameter estimates. It is shown that this cost function can be written in terms of tangent vectors on the surface of the object. The cost can therefore be minimized over possible end effector velocities, resulting in optimally efficient exploration of the surface. Given an estimator and system model, the exploration path automatically generated in this way endows the robot with the ability to seek out geometric surface features that provide the most utility for the improved convergence of the parameter estimate. Computer simulations for a nine-dimensional surface parameter vector show that the algorithm is fast, accurate, and completely independent of the parametrization chosen for the analytic surface models.>

A Kalman-filter-based sensor fusion procedure is proposed for robotic manipulators on unknown curved surfaces. Models of ideal end-effector-surface contact properties are formulated in terms of a surface parameter vector. This vector becomes the state of an extended Kalman filter and completely defines the model of the surface. Filter input measurements can include, but are not limited to, force and joint kinematic data. It is assumed that control can be accomplished using existing techniques if accurate estimates of the surface normals (and hence, the tangent planes) are found. Therefore, a manipulator using incremental control based on local measurements might benefit from the online filter states. Filter covariance can then be considered an indicator for the point at which the unknown surface can be considered known.>

For mechanical hands, control of contact points is facilitated by fingertip force sensors which are capable of measuring the local surface normal of an unknown object. In addition to this information, joint angle sensors and knowledge of the kinematic parameters of the hand commonly provide the physical location of the surface contact point. This enables a surface recovery scheme to utilize both kinds of contact information. Here, a simple multifingered surface-exploration procedure is proposed which allows surface data to be obtained from a kinematic simulation. Using a linear formulation of the surface normal data, the shape estimator can be a simple sequential least-squares routine or the more versatile Kalman filter. Estimator performance is then evaluated for hands with no more than five fingers, with and without the surface normal data. >

A strategy for motion of a planar manipulator on an unknown surface is presented. To estimate a parameter vector for the surface, least-squares estimators are formulated that use both the kinematic and dynamic data available online. The authors begin with the presentation of a control strategy that uses local sensory data to guide the end effector incrementally over the surface. Force feedback is given, which is later shown to be successful in the uninterrupted maintenance of contact. Motion over the unknown shape is justified by demonstrating that a simple shape estimation algorithm can converge to give a parametric description of the surface. Geometric and differential models of the surface are presented and the estimation procedure is specified. Also presented are the results of a number of computer simulations for a three-link planar manipulator moving over an ellipse whose parameters are unknown to the controller. It is found that, in the presence of noise, dynamic data degrade the estimates.>