Tod Levitt, Senior Research Associate

Robotics Laboratory
Department of Computer Science
Stanford University
Stanford, CA 94305

Office: Cedar A7, (415) 723-4676

Research Interests

Computer Vision, Robotics, Uncertainty in AI (theories and practice of belief systems for machine intelligence). A key research area is the integration of machine perception with planning and action for robotics. Computer vision is still very limited; good biological and engineering models yield great uncertainty in scene interpretation. Bayesian inference is a fundamental approach to guiding robotic behavior from uncertain perceptions. Recent work involves extensions of Bayesian inference to hierarchical computer vision, connecting image processing and pattern recognition evidence with high level object understanding and manipulation. Binford's quasi-invariant image and geometric features are the basis for probabilistic estimates of evidence of scene objects. Hierarchical utility models and extended influence diagram techniques are being developed to form a physically deep, but combinatorially tractable, method of guiding robotics perception, inference and action.