Alex Teichman

A mathematical framework for online semi-supervised learning that enables systems to induce new object groups from a few user-provided examples during runtime.

Method to estimate the relative pose of two stationary depth sensors using only motion cues, requiring no calibration targets or specialized hardware.

CLAMS: a technique to calibrate depth sensor distortions by recording data from a handheld sensor and using SLAM, removing the need for calibration targets.

Extended journal paper describing algorithmic optimizations for real-time segmentation and tracking of arbitrary objects in RGB-D data, and applications for training object detectors with reduced annotation effort.

Invited workshop talk presenting early ideas on enabling long-term, online semi-supervised learning for object recognition systems, precursor to group induction.

A semi-supervised learning method that leverages tracking information to automatically find useful training examples, achieving similar accuracy with far less human labeling.

Conference version describing a model-free segmentation and tracking method for RGB-D data that supports subsequent semi-supervised learning approaches.

Overview of recent object recognition research in the Stanford autonomous driving lab, and discussion of challenges needed to make high-accuracy systems applicable beyond autonomous driving.

Original RSS conference paper introducing a method that uses tracking information to expand training datasets automatically, reducing labeling requirements by orders of magnitude.

A method that decomposes object recognition into segmentation, tracking, and track classification, enabling real-time classification of tracked objects (car, pedestrian, bicyclist, other).

A broad summary paper on systems and algorithms developed for Stanford's autonomous vehicle 'Junior', covering perception, sensor calibration, planning, and control.

Work on sparse coding in the exponential family with applications to self-taught learning, co-authored with Honglak Lee, Rajat Raina, and Andrew Ng.

Methods for recognizing configurations of modular robots, published in the International Journal of Robotics Research.