The objective of this research is to investigate the fundamental issues of how to actually fuse multi-sensor data to represent the robot task environment. The principal investigator will first develop a hierarchical strategy for acquiring robot sensory information based on four distinct phases: "Far Away," "Close To," "Touching," and "Manipulation." Because each phase consists of common information and phase-specified information, the principal investigator will package the information for each phase into a phase-oriented distinct template. The next step will be the development of sensor models, sensor coordination, and fusion of multiple sensor data. A new "confidence distance measure" and "distance matrix" will be used as criteria for the detection of sensor errors before the beginning of the sensor fusion process. A mathematical model will be created to represent the level of confidence measures for determining the optimal fused sensor data. Following this, the principal investigator will group all information from the four phases to establish object information templates and complete the representation of the task environment. This approach permits data to be merged in both a low-level way (to minimize the influence of noisy data) and a high level way (constraints are put on the influence between sensors and the way the data is combined). This work can potentially clear the air on a number of issues: the usefulness of faulty sensor isolation, the accuracy of combined data sources, and the overall cohesiveness of the various templates for phased sensing.
