Design and Implementation of Assistive Robotic Residence Home (DIARRH) (Funded by NSF-BEACON CENTER, )

RFID Tagging and Indoor Navigation: We are designing an RFID fully augmented home (rooms are tagged and most of the items in the house are tagged with RFIDs) setting with assistive robot which interacts meaningfully with that environment and residents through its sensors and RFID antennae. The robot is required to improve the quality of life of the cognitively impaired living in the house by running errands. It does this by learning to decide on and execute some daily activities such as searching for items e.g. gadgets, drugs, food items, literature, etc.; giving indoor directions, etc. The knowledge acquired from the RFID fully augmented environment will then be diploid in a RFID partially augmented home (Only house compartments are tagged to aid navigation). The robot has full or partial knowledge of targets and partial knowledge of target location. Thus, learning and decision making are the integral parts of this project. We are investigating evolutionary learning algorithm and developing a hybrid evolutionary learning algorithm capable of robust learning of common household items features and decision making routines. Consequently, the assistive robot should be equipped with a vision system to aid maneuverability and recognition. Thus, we are developing a hybrid hierarchical blend of low level and high level object segmentation and recognition techniques capable of robust and near real-time recognition of some selected household items and actions.

Learning: In a strength based learning classifier systems (LCS), auctioning among classifiers that matches to an environmental message has been used as a means to identify winner classifiers. All classifiers participating in an auction issue a bid proportional to their strength and a winner classifier is allowed to fire and receive a reward or punishment from its environment as a consequence of its action. In this kind of bidding strategy, good classifiers with low strength and little experience have to wait until the strength of useless classifiers has come down through continuous taxation. This slows down the convergence of the learning system to the optimal solution sets. In addition, offspring classifiers that come from weak parents as a result of randomness in the selection process may inherit a small strength as compared to experienced classifiers in the population. Our work introduced a decentralized loaning approach to mitigate the above shortcomings of the bidding strategy in traditional LCS. Loaning among classifiers in the population is allowed. The average bid history parameter gives general information about the bid market (potential of competent classifiers) and determines the amount of loan a classifier should ask. The results obtained show a significant improvement on the performance of the system.

Coordination of Micro-Satellites

We investigating the applicability of heterogeneous systems of micro-satellites in a mission to replace a big monolithic spacecraft, to minimize the cost, enable responsive operations, enable flexibility, and improve survivability. First, we investigated different types of formation-flying for micro-satellites. Then we are developing a coordination approach based on a “Market-Based” algorithm, by using the “Auction Technique”, to be applied to “climate and space weather data collection”. By applying the Market-Based approach, we can optimize the mission by improving the communication; enhance data gathering, and collision avoidance with minimum cost.

Understanding Climate Change

A large volume of satellite images with high dimensionalities and complexities is being collected, processed and stored periodically by the National Oceanic and Atmospheric Administration (NOAA), National Climatic Data Center (NCDC) and other agencies. Due to the ‘curse of dimensionality’ of the images, large computer memory is required for storage and lot of computational resources is needed to archive the database. To overcome these limitations, our focus has been to develop a feature extraction technique that can capture only the dominant features of these satellite images and use the features to track the origin and estimate the intensity of a given tropical storm; and to speed up the search for similar images to a given satellite image of interest. We have developed a pyramidal wavelet decomposition to extract texture features of tropical storms (Debby 2006) and Defense Meteorological Satellite Program (DMSP) images. An automated algorithm capable of tracking the origin of a given storm is developed and tested on the Debby 2006. Finally, a modified Locality Sensitive Hashing is developed, implemented and tested on the DMSP images to find similar images to a query image. The results show a significant reduction in the volume of memory required for storage, a confirmation of the origin of Debby and the superiority of the similarity search algorithm over existing ones.

Expeditions in Computing NSF
Awards: 1029711, 1029166, 1029731, 1028746
This 5-year, $10 Million project is funded by an award from the National Science Foundation's Expeditions in computing program. The program, established in 2008 by NSF's Directorate for Computer and Information Science and Engineering (CISE), is aimed at pushing the boundaries of computer science research. The awards represent the single largest investments by the directorate in basic computer science research.

The Team
The project team, led by the University of Minnesota, includes faculty and researchers from Minnesota's College of Science and Engineering, College of Food, Agricultural and Natural Resource Sciences, College of Liberal Arts, and the Institute on the Environment, as well as researchers from North Carolina A & T State University, North Carolina State University, Northwestern University, and University of Tennessee/Oak Ridge National Laboratory