Hidden Markov Mixture Regression for Robot Manipulation

Building models that can generalize and reproduce desired tasks will  enable the robots to learn and do more, which will make our lives  easier. The aim of this project is to build such a model, able to  learn and reproduce different tasks. We chose Learning from  Demonstration as the way of teaching and we used Hidden Markov Models  (HMM) with a modified version of Gaussian Mixture Regression (GMR) in  order to teach a robot multiple types of trajectories together, with a  small number of demonstrations for each. The robot is then able to  decide on the type of trajectory it will reproduce from the state it  starts its execution and can reproduce the trajectory successfully.