AI Lunch and Seminar

Juan Pablo Mendoza - Detection of Subtle and Context-Dependent Robot Model Inaccuracies

Autonomous robots frequently rely on models of their sensing and actions for intelligent decision-making. Unfortunately, in complex environments, robots are bound to encounter situations in which their models do not accurately represent the world. Furthermore, these context-dependent model inaccuracies may be subtle, such that multiple observations may be necessary to distinguish them from process noise. We explore the problem of detection and correction of such subtle contextual model inaccuracies in high-dimensional autonomous robot domains. Our solution relies on reasoning about these contextual inaccuracies as parametric Regions of Inaccurate Modeling (RIMs) in the robot’s context space, and developing optimization and search-based algorithms for finding these RIMs. We describe the solution in detail, and explore its application to model inaccuracy detection in the CoBot mobile service robots and the CMDragons autonomous soccer robot team.