Using Random Features for the pushing problem:
Probabilistic Data-Driven Models for the Pushing Problem by Maria Bauz à Villalonga
Pushing actions are common mechanisms present in most human and industry manipulations. Nevertheless, finding a precise description for the motion of pushed objects is still an open problem. In this work, we will develop the first data-driven models that can describe the pushing motion taking into account its uncertainty. We will also explain how we collected a high-quality data set for pushing using real experiments that will be available online to motivate research in the pushing domain. A key challenge to describe pushing is understanding friction properly. In most situations, friction makes systems stochastic and introduces uncertainty in our predictions. Moreover, in robot applications, sensors can also add noise into our observations making our state-estimations uncertain. In consequence, our work will consider probabilistic algorithms such as Gaussian Processes to introduce for the first time the uncertainty of our system into the modeling of pushing. In this thesis, we also investigate how these models behave for the particular case of a square object being pushed in a single contact point. This is a good starting point for future generalizations of our models and has already allowed us to simulate properly the motion of pushed objects and validate or refute most typical assumptions considered when trying to describe the pushing problem theoretically.
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.