Research Areas and Projects

Our projects are divided into five core specialized areas, detailed below.

Intelligent Robot Navigation and Control

There are two basic types of robot navigation planning: the conventional point-to-point path planning and the complete coverage path planning. The point-to-point path planning finds an optimal path from a start location to a goal location. However, the complete coverage path planning (CCPP) is a special type of trajectory generation, which requires the robot path to cover every part of the workspace. Many approaches have been proposed in the literature. One of them is to use random motion to scan the whole connected workspace in the case of an unknown map and limited sensory information. We proposed chaotic motion for CCPP. As in single-robot coverage the goal is to completely cover the whole work area in the shortest time. Cooperating robots have the capability of accomplishing the same coverage work faster than a single robot. Synchronization of chaotic systems, showing organized collective behavior out of cooperation of a large number of chaotic subsystems, can be viewed as a simplest kind of "useful" cooperation of chaotic robots. The performance of the synchronized chaotic system with the same coverage of the work area is greatly improved compared to the two random systems and non-synchronized chaotic systems. We optimize random robot collaboration based on chaos synchronization by using genetic algorithms (GA). The optimized cooperative robots improve the efficiency of the synchronized chaos robots by selecting the optimal coupling parameters. With the application to surveillance, chaotic robots can survey the obstacle/person more effectively and successfully compared to the periodic motion surveillance.

Adaptive Distributed Fusion

In order to accommodate multiple sensors over moving objects and their fusion, information from multiple sensors should be registered and aligned. Our newly developed multi-sensor registration and fusion algorithms based on expectation maximization (EM) have been successful in simultaneous registration and fusion of dissimilar sensors. The biases from these sensors are estimated using the EM algorithm for registration. Quite frequently, unstructured data in several forms including textual information will be available in addition to the surveillance information. The new data needs to be seamlessly fused with the surveillance information to provide improved performance. Hence, the problem of fusing structured information with ad-hoc unstructured data needs to be investigated. The unstructured data will be semantically classified and ordered based on the relevance to the structured data available on hand. The classified unstructured data will then be associated with the structured information fused from several sources. Advanced algorithms will be designed to perform contextual fusion of unstructured data with structured information. In case there is an information conflict, we use artificial intelligence (AI) based techniques to reduce the model uncertainty.

Analog Wideband Communications Based on Nonlinear Dynamics

The emergence and ever increasing technological advances in communication systems has been a direct consequence of the evolving consumer demand for low power, low cost and high data rate systems. In the early stage of telecommunications, analog narrowband technologies were developed primarily for long distance transmission of speech signals. Then came the digital narrowband technologies, followed by digital wideband technologies, each overcoming some of the shortfalls of its predecessor. However, digital wideband technologies today still face challenges in accommodating two contradictory requirements of high data rate and low power. The proposition of developing analog wideband systems, presents an exciting possibility in eliminating inherent drawbacks of power hungry digital wideband systems. However, they remain unexplored till date as viable options for wideband communications, resulting in a lack of mathematical foundations to support effective analog wideband communications. An added advantage of using analog communications is the richer signal quality at the receiver. It is this exciting possibility of a novel technique that has motivated us to consolidate our efforts on further pursuing the idea of analog wideband communication schemes. The definition of the ergodic chaotic parameter modulation (ECPM) scheme has been a significant contribution towards this end

Signal Processing for Sensor and Communication Systems

Through-the-wall-imaging: “Seeing” through obstacles such as walls and opaque material provides a powerful tool for surveillance. We proposed a novel chaos-based pulse amplitude modulated ultra-wideband (CPAM-UWB) waveform for through-the-wall radar imaging and detection, since UWB waveform offers good wall-penetration and range resolution. Using chaos transmission, an on-line expectation maximization (EM) based algorithm was also developed to perform semi-blind system identification and to equalize room reverberations in a real time fashion. The result is a chaos-based receiver achieving non-coherent reception. GPS/ INS Integration: Integration of global positioning system (GPS) and inertial navigation system (INS) provides continuous positioning information of high accuracy due to the synergistic effect of both systems. While a Kalman filter is usually employed to fuse the measurements, it requires priori knowledge about the stochastic and deterministic parameters of the systems. We developed an EM method to estimate the unknowns by employing a delta operator model to approximate the continuous-time system. The method achieves simultaneous positioning and unknown parameter estimation. Pre-distortion in HPA: The memory effect in a nonlinear high power amplifier (HPA) renders its output signal to be a function of its past input signals, thereby limiting the performance achievable by linearization. We developed a histogram-based pre-distortion method for compensating the nonlinearity in a communication system. Instead of using conventional parametric models, the proposed method exploits statistical information in the input and output to form an exact inverse of the HPA distribution for accurate compensation.

Goal Driven Situation Assessment

Situation assessment involves analyzing the evidence obtained from the sensors in contrast to prior knowledge to provide decision support. Complexity in situation assessment comes from a large number of system variables and an even larger number of relationships between the variables. In addition, the relationships are also characterized by feedback, non-linearity and conditionality. Another major issue is that the underlying phenomena scale over spatial and temporal boundaries and can potentially depict multi-dimensional asymmetric covariance. We developed a system-of-systems approach to decompose a complex process into a group of simpler tasks which interact and can be managed through the system architecture. Apart from interacting at the system level, the simpler tasks or the sub-systems are relatively independent and designed to represent various functional blocks of the complex process. The proposed systems approach has been developed using graphical models. Information from various sensor platforms in a surveillance network is combined at the algorithmic and signal levels, thereby producing state estimates with statistical information including model probabilities and likelihood functions. When used with dynamic belief update, goal driven recursive reasoning and an integrated database with prior knowledge, a graphical model can be implemented to integrate emergent information about the likelihoods and the uncertainties to approximate the situation assessment beliefs. In the multi-layered graphical structure, sub-assessment is initially performed and these sub-assessments are summarized to form an overall assessment. Depending upon the type of the application, the sub-assessments can either represent evidence required to establish a situation or sub-situations which are summarized to determine a global situation. Use of graphical models, firstly contributes to the enhancement of quantifiable system performance by processing statistical information and the second advantage is the ability to implement an adaptive framework for situation assessment.