Trajectories in Hand Synergy Spaces for in-hand manipulation

In-hand regrasping is a challenging but important problem for dexterous robots equipped with humanoid hands. To facilitate control, synergy models have been proposed that use low-dimensional representations to deal with the many degrees of freedom (DoFs) present in robotic hands. In this work, we build upon a previously proposed synergy model that was used for in-hand regrasping directly in the synergy space of the model, and enhance it with additional metrics based on Riemannian geometry that allows us to impose additional constraints in the latent space. We present quantitative results that demonstrate that with the use of these metrics we are able to construct trajectories directly in the synergy space of the model with higher accumulated smoothness along the paths. Finally, we execute several in-hand regrasping trajectories, generated using our proposed models, in simulation. Implementation is available at https://github.com/JorgeTelo/Thesis