Programming by Demonstration using Hidden Semi-Markov Models

Abstract

Despite the growth of robotics, robots are far from being used in everyday life. The major reasons for this slow integration include robots’ lack of sensitivity, safety and flexibility. The aim of this final project is to improve these three issues, by providing the user a tool for teaching tasks to robots. An advanced robotic arm with force and torque sensing capabilities is used. A Programming by Demonstration approach based on Hidden Semi-Markov Models is developed. The framework is adapted to cope with in-contact task, handling perturbations and user interaction with different control techniques. Experiments evaluate the performance of the approach, showing that it allows to successfully imitate not only trajectories, but also force profiles.