Machine Learning

Modeled picking and placing trash bin skill using manipulator arm of Locobot robotic platform as Gaussian Process (GP) to enable imitation based skill learning from single demonstration

Incorporated Dynamic Movement Primitives (DMP) approach to model stereo typical motion in data efficient manner and used that model to predict trajectory and goal location from a partially observed trajectory

Monocular Object-oriented Simultaneous Localization and Mapping (SLAM) using deep Convolutional Neural Network (CNN) and factor graph optimization

A multi-class linear SVM on extracted features

Implemented extension of Convolutional Neural Network to reduce number of paramteres for similar performance