Visual Data Mining of Robot Performance Data

We propose to design and develop VDM/RP, a visual data mining system that will enable analysts to acquire, store, query, analyze, and visualize recent and historical robot performance data. During mission operations, these capabilities will enable operators to more quickly and accurately detect and interpret data patterns that support or rebut candidate diagnoses or hypotheses about robot problems. During robot system development and experimentation, VDM/RP will enable analysts and robot designers to review robot test data to create and refine models that specify quantitative relationships among robot system health and status variables that hold for nominal and off-nominal modes. Key innovations include interactive arrays of timelines and graphs for visualizing multivariate, time-oriented data, temporal queries to search for significant data patterns, and intelligent assistance to simplify user selection of data, analyses, and visualizations. During Phase I, we prototyped visualizations to analyze K10 rover LIDAR scan failures. During Phase II, we will develop three successively more capable versions of VDM/RP for test usage and evaluation by NASA's Intelligent Robotics Group.