Research
Autonomous Systems
The main goals are to contribute to the understanding of human intelligence and to participate in the development of intelligent machine systems, including autonomous bio-, software and hardware systems. We are currently exploring the idea that symmetry theory provides a key organizing principle for cognitive architectures. Operationally, the hypothesis can be formulated as follows: (1) symmetry theories for various essential aspects of the world are formulated and based on these, we seek: (2) mechanisms to organize percepts into instances of the vocabulary of the theories (e.g., sets, operators); (3) mechanisms to determine whether the perceived sets and operators form a valid model of the theory; and (4) mechanisms to exploit the model (e.g., theorems, policies, etc.). These theories are being tested in cognitive sensor networks, vehicles, and buildings.
Machine Learning
Machine learning technology aims to solve problems of inference and prediction: based on past data, we desire algorithms that can reliably forecast the future. Machine learning techniques have led to significant advances in the fields of natural language processing, computational biology, robotics and medicine. The machine learning group at Utah works in several areas, ranging from basic technology building to application development and from mathematical modeling to algorithmic implementation. These areas include structured prediction, domain adaptation, semi-supervised learning, bootstrapping and Bayesian statistics. Structured prediction aims at developing algorithms that can predict complex outputs, such as those found in natural language or biology. Domain adaptation, semi-supervised learning and bootstrapping address the frequently occurring problems of mismatches between past data and anticipated future data. Our work has led to significant advances in a variety of application domains, including natural language processing and data mining.
Natural Language Processing
The goal of natural language processing (NLP) research is to create computational models for understanding natural languages, such as English. The natural language processing group at Utah works in several application areas, including Information Extraction, Opinion Analysis, and Summarization. Information Extraction systems identify important facts associated with events, such as the perpetrators and targets of a terrorist attack or the disease names and victims associated with an infectious disease outbreak. Opinion Analysis systems determine whether a sentence or document is expressing an opinion or judgment, which is useful for classifying reviews, analyzing product reputations, and answering questions. Summarization systems automatically identify and condense the aspects of collections of source materials that are relevant to a user into a summary. The NLP group at Utah also specializes in techniques that can automatically learn dictionaries and linguistic knowledge from raw text. Our group has developed a variety of learning techniques and bootstrapping algorithms for learning extraction patterns, semantic dictionaries, opinion clues, and knowledge for coreference resolution.
Robotics
The robotics group conducts research on a wide variety of topics, particularly mobile robots, haptic interfaces, novel sensor and actuator systems, and intelligent sensor networks.
A variety of move mobility platforms are being developed for traversing varied terrain. For rough terrain, approaches include bipedal and quadrupedal legged robots, compliant framed wheeled robots, and hybrid robots with legs that tuck into a ball for rolling downhill. Climbing robots include insect-like robots that utilize claws and spines to adhere to small features even in nearly smooth walls, and robot snakes that can crawl through pipes. Ornathopters (flying robots) are also being developed using flapping wings for lift.
Haptic interfaces are robot devices that physically interact with humans, and include both manual interfaces and locomotion interfaces. The virtual prototyping project seeks to add a sense of touch to the mechanical design process. Aside from feeling the force of contact, the tactile feel of contact is also being provided by pressing on the fingertip with a moving indenter that simulates the point of contact. A method of measuring human grasp force is being developed, utilizing imaging coloration changes in a fingernail with a camera. The Sarcos Treadport Locomotion Interface seeks to provide a multi-sensory experience of walking, including visual, mechanical, auditory, and wind displays.
Wet robots are being developed that embed Shape Memory Alloy "muscles" within a network of biologically inspired "robotic blood vessels" that fluidically distributes thermal energy to and from any actuators in the array using only a small number of valves. Smart sensor networks are being developed that are capable of computation, communication and sensing for many distributed sensors.
