Boundedness Conditions in Adaptive Control (2006-now):

Adaptive control systems pose a number of challenging problems that have not yet been completely addressed by the control systems community. For example, existing "stability" proofs typically show that in presence of modeling uncertainties the evolution of a system subject to direct adaptive control laws is Uniformly Ultimately Bounded (UUB) into a certain compact set (also known as the UUB set). However, these proofs often lack generality and/or constructiveness. 
Efforts are under way to express the UUB bounding set within the framework of Ellipsoidal Calculus framework.
This formulation will be much less conservative than the standard one, it will lead to bounding values for each state variable of the system (as opposite to a single limit on the entire norm of the state vector, as is currently the case), and it will allow using the Ellipsoidal Toolbox to develop a MATLAB based software for the calculation and visualization of the bounding set.

Visual Navigation via Optical Flow Techniques (2006-2007):

Optical flow sensing allows flying insects with compound eyes to perform quick and highly accurate navigation/avoidance maneuvers. Therefore, Optical flow based techniques could potentially be extremely useful within a variety of autonomous vehicles that typically have very limited on-board computational resources.
However, despite several ongoing efforts, several more years of research are likely needed before such algorithms could find use within UAVs by providing capabilities such as, for example, autonomous collision avoidance. 
This is mostly due to the fact that most optical flow algorithms do not provide yet the accuracy, the measurement density and the computational speed that are necessary for real world implementation.
To date, several Optical Flow algorithms are being extensively compared, the best of them will be implemented for real-time execution within an PC-104 computer.

Visual Assisted Autonomous Aerial Refueling (2004-2007):

One of the biggest current limitations of military UAVs is their limited range. An increase in fuel capacity would lead to larger and heavier UAVs along with deterioration in maneuverability and handling qualities. Therefore, the acquisition of autonomous aerial refueling (AAR) capabilities for UAVs is a critical goal. 
Currently, precision maneuvering in close proximity operations cannot rely on GPS alone, due to issues of inherent limited accuracy together with dropouts and/or distortions caused by shadowing effects between the tanker and the UAV. 
A solution can be provided by the integrating GPS sensing with Machine Vision (MV) based position estimation. Since both cameras and hardware for image processing are getting less expensive, it is appealing to conceive the notion of a MV-assisted AAR for UAV. Within this approach, several well-established image-processing techniques can be used to filter the image, detect the main features of the tanker, and isolate them from the rest of the image. Finally standard pose estimation algorithms can be used to estimate the UAV relative position and orientation. An AIR Force Sponsored project is currently ongoing at WVU to evaluate the feasibility of Machine vision based Autonomous aerial refueling for UAVs.

Real Time Operating Systems (2003-2007):

The discipline of control system design has undergone tremendous changes in the last two decades. High level visual tools for simulation and control system generation, together with low cost and high performance microprocessors, have been surely among the main actors of such transformation, since they allowed the control engineer to progressively reduce the overall development time, while gradually enabling newer and cheaper control architectures.
However, the deployment of a control system that uses a general purpose CPU is often held back by the lack of a low cost, general purpose, reliable, and standard-compliant real time operating system. Even among the top 30 most used real time operating systems, alternatives are either very expensive, or hard to interface to the most used control synthesis tools, or too hardware dependent, or too big to fit an embedded system, or sometimes they are just not reliable enough.
Our research group at WVU has tried several RTOS in the past few years. At this time (2008), RTAI (version 3.6 on a 2.6.23.13 Linux kernel, within a Busybox 1.01 based system) is used to deploy real time controllers generated from Simulink-RTW (Matlab R2008a). While RTAI currently presents a few drawbacks mainly due to the lack of comprehensive documentation, it definitely holds the promise of becoming an open source, standard compliant, highly flexible and very reliable RTOS, therefore placing itself in the position to be the RTOS of choice for a majority of research and commercial Hard Real Time applications.

UAV Formation Flight Control (2003-2005):

Development and application of unmanned air vehicles (UAV) is rapidly expanding as a result of evolving needs for more affordable and survivable systems. In the military environments, it has bees conjectured that by 20-30 years combat flight fleets will consist almost entirely of UAVs.
It is known that multiple aircrafts flying in a close formation can yield a substantial gain in efficiency, therefore saving fuel and/or increasing the operating range. This efficiency boost can be crucial when considering UAVs. The flight of multiple UAVs in close formation involves numerous technical challenges, just to name a few, aerodynamic modeling, modeling of the formation as a system, sensing and communication issues, distributed real-time control, fault tolerance, collision avoidance and formation engagement / disengagement. An AIR Force Sponsored project at WVU successfully demonstrated the actual flight in formation of three WVU-Built autonomous YF22 models. 

Aircraft Parameter Identification (PID) (2000-2003):

This is definitely a branch of System Identification. The problem is to identify (possibly on-line) the linear and nonlinear models of the aircraft. The most effective way to do that  consists in estimating the derivatives of the aircraft aerodynamic forces versus the state and input variables, therefore using the available knowledge about the general structure of an aircraft model. Multiple (Recursive) Regression methods are the inner engine of the algorithms that solve linear identification problems. Nonlinear identification problems are obviously more challenging and their solution usually relies on a good physical insight and on some form of optimization algorithm like Steepest Descent or Newton-Raphson.

Several Frequency Domain Based and Time Domain Based algorithms for PID have been proposed, evaluated and compared. A Simulink Library featuring the most effective on line PID algorithms (e.g FTR, LWR, RLS, LS) as well as several examples has been built and made available. Most of the current work on PID is funded by a research contract with ISR (SAVE).

SFDIA (2000-2003):

SFDIA stands for Sensor Failure Detection Identification and  Accommodation, it is closely related to the System Fault Diagnosis, which in turn can be seen as a branch of System Identification.

The most recent approaches to the problem involve 2 key ideas:

1. Make an extensive use of "analytical redundancy" between measurements, (in short use the available/retrievable relationships between measurements), to make decisions about the healthy/faulty state of a sensor.
2. Use (on-line) function approximators such as Neural Networks to approximate the above relationships between measurements.

On this subject, our group produced several journal and conference publications, as well as several research contract proposals. Most of the current work is funded by two research contracts with DERA and NASA.

Underwater Vehicle Modeling & Control (1996-2000)

Autonomous Underwater Vehicles (AUV’s) will play an important role in the oceans exploration. Both modeling and control of such vehicles are challenging due to a variety of reasons:

1. The nonlinear behavior of a vehicle subjected to hydrodynamic forces and moments.
2. The multivariable character of a 6dof vehicle, which often exhibits an high degree of coupling among different channels.
3. Uncertainties due to adverse environments, different load conditions or unknown sea currents.

Such vehicles (or their models, when sufficiently accurate) are therefore an excellent test platforms for different Robust Nonlinear Multivariable Control methods.

A detailed 6DOF Nonlinear model of both an Autonomous UV and a Towed UV have been developed, and the most rewarded control techniques have been applied to it and compared.

Adaptive Control via Neural Networks (1999-2001)

An adaptive controller is a nonlinear controller that "learns from experience" and adapts itself so as to improve closed-loop performance. This learning (optimization) process often involves on-line System Identification to some extent. The resulting closed loop system is a nonlinear system often having two time scales: a fast inner control loop and a slow outer "updating" loop.
Adaptive controllers are an effective way to cope with some  nonlinear system that exhibit two time scales, as well as slow time varying system and systems with some uncertain parameters.
A Neural Network based Adaptive Controller has been studied on a detailed Nonlinear 6DOF model of an Autonomous Underwater Vehicle.

Robust Linear Multivariable Control (1996-1999)

Multivariable Systems (MIMO) have coupling and directionality properties that render them harder than SISO systems from the analysis, simulation, and especially control synthesis points of view. The importance of multivariable systems comes up not only from the fact that they are more general than SISO systems, but it is also due to the fact that most complex systems are multivariable in nature.
In the eighties and early nineties, many control synthesis techniques for linear systems have been unified under the "modern" multivariable control paradigm, which is a natural evolution of the Hamiltonian approach.
The challenge of the current research is the extension of the paradigm to the Nonlinear and/or Hybrid systems.

A powerful Visual Matlab toolbox for Multivariable Systems Analysis and Robust Control Synthesis has been developed. Using this tool, several MIMO control systems have been designed for a variety of systems.