Motion Retargeting and Machine Learning for Humanoid Robotics

Accurate tracking and training of human movements can lead to efficient humanoid robot manipulation. Machine learning can pave the way for training these movements as it makes use of algorithms and statistical models to predict new movements relying on patterns and inference. We propose a model that explores the use of machine learning algorithms as a first step to improve humanoid robot manipulation. The implementation consists of two parts, firstly a motion tracking experimental setup that has been developed to accurately capture the movements of human body. Secondly, the development of a machine learning model to train the robot, especially wrist movements, for a given stimuli via music. We described the experimental challenges involved during the implementation such as setting up the experiment, extraction of relevant features for machine learning model and selection of fine-tuning parameters. Our initial results are encouraging thus supporting our hypothesis that the training of humanoid robot can be improved by using machine learning.