As robots become more complex it becomes necessary to create more effective methods of imparting them with knowledge. One area where complexities are apparent is in learning to manipulate. Vanderbilt Universitys Intelligent Soft Arm Control robot, ISAC, designed with two 6 degree-of-freedom arms, actuated by McKibben artificial muscles was designed for interaction with people. Programming specific movements for that purpose quickly becomes impractical. It would be more practical for ISAC to learn motions by observing them. The field of Imitation Learning is devoted to that problem. As a precursor to Imitation Learning and as a base for human-robot interaction I have designed a system by which ISAC recognizes human arm gestures and repeats them back. The system requires minimal equipment to be worn by the user. A simple pair of brightly colored gloves is all that is required. This is in contrast to other approaches that require the user to don a set of encoders on his, or her, arms to record precise joint angles. This is impractical in many real world settings. Here I present the motivations for the creation of the system and describe several tools that simplified its construction. The methods by which the system tracks, filters, compresses, and translates the motion of the human operators hands are also discussed.

Acoustic beamforming is described in general terms, and the problems of implementing online acoustic beamforming on a wireless sensor node are described. A low-power FPGA-based solution is presented, implemented on a sensor board for a MicaZ mote. An integrated FFT-based frequency filter is also presented. The results of tests performed in an anechoic chamber demonstrate that the system behaves as desired, performing beamforming in real time, filtering undesired frequencies, and transmitting the beamforms to another nearby node. Power consumption of the system remains within acceptable limits for wireless sensor deployment.

In low-end microcontroller systems task scheduling is often not done, performed using very ad-hoc methods, or using a high-overhead RTOS. In this paper we offer an alternative: the design-time generation of high-precision schedules which are then executed using a minimal scheduler on a microcontroller. The schedules are derived from a simple task modeling language that allows the specification of properties of tasks, constraints on task execution, as well as worst-case interrupt rates and interrupt service execution times. An off-line scheduling algorithm then computes a schedule for the tasks specified which is then compiled and linked with a simple execution kernel that performs the scheduling. The paper describes the modeling language, the schedule computation approach used, the run-time kernel, as well as experimental results.

Dynamic logic circuitry is generally faster and smaller than equivalent static logic circuits. The use of these circuits in space is desirable, but not much work has been performed in assessing their vulnerability to the ionizing particles present in the space environment. These particles can generate transient currents that can affect circuit performance. The effects of single event transients on dynamic logic circuitry are studied. Simulations are performed using the IBM 130nm and 90nm processes. Mechanisms of upset are determined for both precharge and evaluate phases of dynamic logic. Frequency effects, stage length dependency, keeper effects, and LET effects are all considered. An analysis by hit nodes on a dynamic full adder circuit is performed at varying frequencies. Results are compared with equivalent static logic circuits.

Alternative dielectrics are rapidly becoming necessary for the future of Metal-Oxide-Semiconductor (MOS) devices, as SiO2-based technology reaches scaling limits. In this thesis, the total ionizing dose irradiation effects on Si and Ge MOS capacitors with Hf-based gate dielectrics are investigated. The radiation response of the HfO2/Dy2O3 on Ge MOS devices exhibited no change in the capacitance-voltage characteristics. HfO2 has the tendency to trap large amounts of electrons relative to SiO2. However the high gate leakage current likely efficiently neutralized any radiation-induced charge build-up for these devices. We have also examined the radiation response and bias-temperature effects of Si MOS devices with HfSiON gate dielectrics. The results showed much improved radiation hardness relative to earlier Hf silicate devices (net oxide-trapped charge density ~ 16× less). The low-Si3N4 content film also displayed enhanced charge trapping relative to the high-Si3N4 content film. In addition, the HfSiON devices showed much reduced degradation from bias-temperature stress relative to SiO2 and HfO2 devices. However the improved bias-temperature stability likely results from difference in processing techniques, and comes at the cost of high density of process-induced interface traps. The material properties and processing techniques can significantly influence the reliability and radiation response of high-k MOS devices. The work in this thesis provides insights to the reliability and radiation degradation that are essential to the development of Hf-based MOS devices.

This dissertation presents several innovations in the area of small animal image registration. We start with the problem of reconstructing 3D histological volumes. A new method is proposed to normalize intensity across slices. We then propose a solution that enhances the reconstructed histological volumes. Our results have shown that with as few as four histological volumes we can produce one virtual volume that is substantially better than any of the volumes used to create it. But, brain studies are only a small part of the studies being conducted with small animals. Registering the whole body volumes brings a new set of challenges. Hence, a fully automatic registration algorithm for whole body CT images is proposed in this dissertation. Qualitative and quantitative validation results show the algorithm can successfully register images even when these images are substantially different in shape and size. Furthermore, one constrained intensity based nonrigid registration algorithm is presented to permit the deformation of soft tissues while limiting the deformation of bony structures.

Silicon Germanium (SiGe) Heterojunction Bipolar Transistors (HBTs) are a new candidate technology for space applications due to their superior performance to traditional bipolar transistors and tolerance of total ionizing dose (TID) and displacement effects from space radiation. However, SiGe HBTS are susceptible to low energy ion inducing Single Event Upsets (SEUs), another form of space radiation. The high susceptibility of these devices to SEU is a reliability concern for the technologys use in space applications. The Rectangular Parallel-Piped (RPP) model fails to predict the results of broadbeam testing on these devices. The observations made through this work offer new insights into the mechanisms that must be treated to arrive at such a model. This work presents evidence through microbeam testing and Technology Computer Aided Design (TCAD) simulations that the isolation technology used in design of the SiGe HBT plays a unique role in the charge collection process of low energy deposition events in these devices. A interplay between the ions angle of incidence and its interaction with the deep trench isolation (DTI) account for its unexpected response. These results will be shown through microbeam testing of an IBM 7HP SiGe HBT and TCAD simulations on an IBM 5AM SiGe HBT.

Pulsed-laser irradiation is used to simulate dose-rate effects in discrete devices. A methodology is established to correlate laser pulse energy results to an equivalent dose rate value without LINAC validation. It is demonstrated that laser dose rate data can be converted to equivalent LINAC data and that the percent coverage of metallization has a linear relationship with the conversion factor of device families. By using the percent metal coverage, the conversion factor of an untested part may be determined from the linear extrapolation of a trend established by similar device types. This conversion factor may be applied to laser data to establish dose rate or may be applied directly to the slope of the laser data to establish the correlating Kpp for insertion into dose-rate models. This process can be performed using a common laser and may be quickly applied to produce models for the fluctuating discrete parts lists for weapons system development.

This project is concerned with adaptive image processing systems. In this dissertation I present the ground work for a generalized adaptive image processing (AIP) methodology / environment that can evolve to increasingly complex problems. The generalized image sequence processing environment defined in this research was evaluated by mapping example user image sequence problems through the AIP representation to an instantiation / implementation under the ISIS model integrated computing signal processing platform (SPP) environment. The AIP methodology breaks up adaptive image sequence processing problems into localized control loops. Complex adaptive image sequence processing problems can be broken down utilizing hierarchy or other combinations of the AIP elements. The AIP localized processing loop includes the following top level representations: monitoring point, preprocessor, evaluator, planner, reconfigurer, and user goals. The architecture of these basic building blocks allows for their growth/evolution to increasingly sophisticated implementations.

This work will focus on giving a perceptual system the ability to detect changes while maintaining its understanding of the environment. Most change detection systems can only perceive the change in the environment. They are not capable of processing the other objects in the environment. Nor are they capable of understanding what type of change they have just detected. This system aims to detect the difference between a novel object introduced to the environment and a known object moved within the environment while still segmenting the image. The image segmentation will use very high dimensional feature vectors. These will be obtained from multiple training images, and each percept will be given a specific label. The feature vectors will then be converted into sparse vectors and arranged in an approximate nearest neighbor (NN) search tree. The new images sparse vectors will scale the tree based on the Euclidian distances of the current sparse vector to the tree leaf nodes. The label from the leaf nodes will be selected as the representation of the percept in the new image. The novel objects will be detected based on a threshold distance from the leaf leave node. If this distance exceeds the threshold the object will be considered novel. The moved objects will be determined by a previously trained look up table (LUT). The LUT will hold a list of acceptable labels in for each pixel, and will be created from a series of training images. The results from the experiments show that this system is capable of learning the objects in an environment and understanding how the environment changes.